chore(chat): tighten tool descriptions to reduce AI selection confusion

- plan_and_execute: restrict to 3+ step tasks; prevents over-triggering on simple requests
- memory_read: hint to call memory_lookup first to find the correct path

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

CLAUDE.md

@@ -8,7 +8,8 @@ a plugin/integration system (Stage 2+) and an encrypted vault (Stage 3+).
 ## Current Status
 
 **Stage 3 — Memory Database: complete** (2026-05-18)
-Next: Stage 4 — Vault Encryption
+**Stage 6 — Daemon infrastructure: in progress** (`feat/daemon` branch)
+Next: Stage 4 — Vault Encryption (skipped for now); messaging bots (Stage 6 remainder)
 
 ## Project Roadmap
 
@@ -19,11 +20,11 @@ memory in `~/.pyra/memory/`, and hard security boundaries around the vault.
 ### Stage 2 — Plugin Framework ✅ COMPLETE
 - `src/pyra/plugins/` package: `base.py`, `loader.py`, `registry.py`, `executor.py`, `install.py`
 - `src/pyra/bundled_plugins/` — ships bundled plugin scripts with pyra
-- `src/pyra/daemon/` stub (CLI surface only)
+- `src/pyra/daemon/` stub (CLI surface only; daemon itself is Stage 6)
 - Config: `PluginConfig` + `DaemonConfig` added to `PyraConfig`
 - Bootstrap: `~/.pyra/plugins/` and `~/.pyra/logs/` created on startup
 - Chat session: AI tool-use loop (up to 10 iterations), approval gate, plugin slash commands
-- CLI: `pyra plugin list/install/enable/disable/setup`, `pyra daemon *` stubs
+- CLI: `pyra plugin list/install/enable/disable/setup`, `pyra daemon *` (stubs at Stage 2; implemented in Stage 6)
 
 ### Stage 3 — Memory Database ✅ COMPLETE
 - `src/pyra/memory/database.py`: SQLite + FTS5 via `memory_meta` + `memory_fts` tables
@@ -99,6 +100,7 @@ the vault under namespaced keys (`plugin:{name}:{key}`).
 | `config/manager.py` | ruamel.yaml round-trip config read/write, chmod 600 enforced |
 | `config/dirs.py` | `bootstrap()` — creates `~/.pyra/` tree, checks vault sentinel every startup |
 | `chat/session.py` | prompt_toolkit REPL loop, AI tool-use loop, plugin slash commands |
+| `chat/planner.py` | `TaskPlanner` — multi-step plan approval loop, per-step AI execution and verification |
 | `chat/renderer.py` | Streaming + non-streaming markdown via rich, injection warning panel |
 | `chat/history.py` | Conversation list, token budget trimming, tool message support |
 | `memory/database.py` | SQLite+FTS5 — `init_db()`, `upsert()`, `remove()`, `search()`, `list_all()`, `migrate_from_files()` |
@@ -116,7 +118,11 @@ the vault under namespaced keys (`plugin:{name}:{key}`).
 | `plugins/executor.py` | Approval gate: scan args → prompt → execute → scan result → log |
 | `plugins/install.py` | Copies bundled plugins to `~/.pyra/plugins/` |
 | `bundled_plugins/` | Standalone plugin scripts shipped with pyra (installed on demand) |
-| `daemon/__init__.py` | Daemon package stub (implementation in Stage 2.4) |
+| `daemon/pid.py` | Atomic PID file — write, read, stale detection (POSIX + Windows), context manager |
+| `daemon/ipc.py` | IPC transport — Unix socket chmod 600 + UID-check (Linux/macOS) or TCP loopback + port file (Windows); newline-delimited JSON protocol |
+| `daemon/service.py` | OS service file generation + install/uninstall — launchd plist (macOS), systemd user unit (Linux), schtasks XML (Windows) |
+| `daemon/core.py` | asyncio event loop entry point, `PluginSupervisor` (per-task restart, max 10×, 5s back-off, reload), IPC command dispatch, signal handling |
+| `daemon/__init__.py` | Public daemon API exports |
 
 ### Runtime: `~/.pyra/`
 
@@ -243,7 +249,7 @@ uv pip install -e ".[all-plugins]" # Everything
 ## Running Tests
 
 ```bash
-pytest tests/ -v                           # all unit + security tests (161 tests)
+pytest tests/ -v                           # all unit + security tests
 pytest tests/integration/test_lmstudio.py  # requires LM Studio at localhost:1234
 ```
 
@@ -492,6 +498,40 @@ Dataclass: `MemoryFile(name, path, category, size_bytes, modified)`
 | `list_bundled_plugins` | `(bundled_dir: Path) -> list[str]` | Names of all bundled plugins that have a `manifest.json` |
 | `read_manifest` | `(plugin_dir: Path) -> dict` | Reads `manifest.json`; returns `{}` if missing |
 
+#### `daemon.core`
+
+| Function | Signature | Purpose |
+|----------|-----------|---------|
+| `run_foreground` | `() -> None` | Entry point for `pyra daemon run` — loads config + plugins, writes PID file, runs asyncio loop |
+| `start_background` | `() -> None` | Spawns `pyra daemon run` as a detached subprocess (`start_new_session` on POSIX, `DETACHED_PROCESS` on Windows) |
+
+#### `daemon.pid`
+
+| Function | Signature | Purpose |
+|----------|-----------|---------|
+| `resolve_pid_path` | `(cfg_path: str) -> Path` | Expand `~` and resolve to absolute Path |
+
+#### `daemon.ipc`
+
+| Function | Signature | Purpose |
+|----------|-----------|---------|
+| `send_command` | `(address, msg, timeout=5.0) -> IpcResponse` | Synchronous CLI helper — `asyncio.run(IpcClient.send(...))` |
+| `get_socket_path` | `(cfg: str) -> Path` | Expand `~` and return Unix socket path |
+| `is_unix_socket` | `() -> bool` | True on Linux/macOS (`sys.platform != 'nt'`) |
+| `get_port_file_path` | `() -> Path` | Path to `~/.pyra/daemon.port` (Windows TCP port file) |
+
+#### `daemon.service`
+
+| Function | Signature | Purpose |
+|----------|-----------|---------|
+| `detect_platform` | `() -> Literal["macos","linux","windows"]` | Detect current OS |
+| `find_pyra_executable` | `() -> str` | `shutil.which("pyra")` → sibling fallback → `sys.executable -m pyra` |
+| `install_service` | `() -> None` | Generate + register OS service (reads config for log/pid paths) |
+| `uninstall_service` | `() -> None` | Deregister OS service |
+| `render_launchd_plist` | `(exe, log_file, pid_file) -> str` | macOS plist template |
+| `render_systemd_unit` | `(exe, log_file) -> str` | Linux systemd unit template |
+| `render_schtasks_xml` | `(exe) -> str` | Windows Task Scheduler XML template (write as UTF-16) |
+
 #### `chat.renderer` — rendering functions and shared `console`
 
 Import `console` from here; do not create a second `rich.Console()` in new code.
@@ -514,7 +554,7 @@ Import `console` from here; do not create a second `rich.Console()` in new code.
 | `GeneralConfig` | `config.schema` | `general:` block — `user_name`, `assistant_name` |
 | `ProviderConfig` | `config.schema` | `ai:` block — `provider_id`, `model`, `base_url` |
 | `PluginConfig` | `config.schema` | `plugins:` block — `enabled`, `require_approval`, `log_executions` |
-| `DaemonConfig` | `config.schema` | `daemon:` block |
+| `DaemonConfig` | `config.schema` | `daemon:` block — `enabled`, `socket_path`, `log_file`, `pid_file`, `ipc_port` |
 | `MemoryConfig` | `config.schema` | `memory:` block — `max_tokens_in_context`, `auto_load` |
 | `SecurityConfig` | `config.schema` | `security:` block — `injection_detection`, `log_injections` |
 | `ConversationHistory` | `chat.history` | Holds message list; builds API payload via `build_for_api()`; trims to token budget |
@@ -524,3 +564,6 @@ Import `console` from here; do not create a second `rich.Console()` in new code.
 | `Tool` | `plugins.base` | Dataclass — `name`, `description`, `parameters` (JSON Schema), `handler`, `requires_approval` |
 | `PyraPlugin` | `plugins.base` | `@runtime_checkable` Protocol — the plugin interface |
 | `BasePlugin` | `plugins.base` | Concrete base with no-op defaults; plugins should inherit this |
+| `TaskPlanner` | `chat.planner` | Multi-step plan runner; `make_tool_handler()` returns the callable wired into the chat session; presents plan for user approval, executes each step via litellm with up to 5 tool-use iterations, verifies output before proceeding |
+| `PluginSupervisor` | `daemon.core` | asyncio supervisor — `add_task(name, factory)`, `start()`, `stop()`, `reload()`, `status()`; restarts crashed tasks up to 10× with 5s back-off |
+| `PidFile` | `daemon.pid` | `write()` (atomic), `read()`, `is_stale()`, `remove()`, context manager; `PidFileError(OSError)` raised when live PID already exists |

README.md

@@ -1,7 +1,7 @@
 # Pyra
 
-A personal AI assistant CLI with vault-first security. Combines multi-provider AI chat with
+A personal AI assistant CLI with vault-first security. Combines multi-provider AI chat,
-long-term memory and (coming) automation skills.
+long-term memory, and an extensible plugin system.
 
 ## Quick Start
 
@@ -31,6 +31,17 @@ pyra chat             # start talking
 | `pyra memory read <name>` | Read a memory file |
 | `pyra memory write <name> <content>` | Write a memory file |
 | `pyra memory append <name> <content>` | Append to a memory file |
+| `pyra plugin list` | List installed and available plugins |
+| `pyra plugin install <name>` | Install a bundled plugin |
+| `pyra plugin enable <name>` | Enable an installed plugin |
+| `pyra plugin disable <name>` | Disable a plugin (keeps it installed) |
+| `pyra plugin setup <name>` | Run a plugin's credential setup wizard |
+| `pyra daemon start` | Start the background daemon *(Stage 6, not yet implemented)* |
+| `pyra daemon stop` | Stop the running daemon *(Stage 6, not yet implemented)* |
+| `pyra daemon status` | Show daemon status *(Stage 6, not yet implemented)* |
+| `pyra daemon restart` | Restart the daemon *(Stage 6, not yet implemented)* |
+| `pyra daemon install` | Register Pyra as a system service *(Stage 6, not yet implemented)* |
+| `pyra daemon uninstall` | Remove the system service *(Stage 6, not yet implemented)* |
 
 ### In-chat slash commands
 
@@ -38,6 +49,7 @@ pyra chat             # start talking
 |---------|-------------|
 | `/help` | Show available commands |
 | `/memory list` | List memory files |
+| `/config` | Open the configuration TUI |
 | `/clear` | Clear conversation history |
 | `/quit` or `/exit` | Exit Pyra |
 
@@ -48,16 +60,41 @@ pyra chat             # start talking
 - **Prompt injection scanner** — warns on suspicious AI output, logs to `~/.pyra/security.log`
 - **Path sandboxing** — the AI can only reference memory files by name; traversal is blocked
 
+## Plugins
+
+Pyra has an extensible plugin system. Bundled plugins are shipped with Pyra and installed on
+demand; third-party plugins can be dropped into `~/.pyra/plugins/` directly.
+
+Each plugin is a directory containing a `manifest.json` and a `plugin.py`. Plugin credentials
+are stored in the vault under namespaced keys (`plugin:<name>:<key>`) — never in `config.yaml`.
+
+```bash
+pyra plugin list                  # see what's available
+pyra plugin install <name>        # copy a bundled plugin to ~/.pyra/plugins/
+pyra plugin setup <name>          # enter credentials (stored in vault)
+pyra plugin enable <name>         # activate for the next chat session
+```
+
+## Multi-step Planning
+
+When given a complex task the AI can propose a **multi-step plan** using the built-in
+`plan_and_execute` tool. Pyra prints the plan and asks for approval before executing
+anything. Each step runs as a separate AI call with access to enabled plugin tools; each
+result is verified before moving on to the next step. You can decline the plan or
+interrupt at any point.
+
 ## Memory
 
 Pyra reads your memory files at the start of each session and injects them as context.
-Files are plain Markdown stored in `~/.pyra/memory/`:
+Files are plain Markdown stored in `~/.pyra/memory/`, indexed by a SQLite full-text search
+database (`memory.db`) for fast in-chat lookup.
 
 ```
 ~/.pyra/memory/
 ├── user/profile.md     ← who you are
 ├── context/            ← ongoing projects
-└── knowledge/          ← general notes
+├── knowledge/          ← general notes
+└── memory.db           ← FTS5 search index (auto-managed)
 ```
 
 ## `~/.pyra/` Directory
@@ -67,15 +104,25 @@ Files are plain Markdown stored in `~/.pyra/memory/`:
 ├── config.yaml         ← provider + model (no secrets)
 ├── security.log        ← injection event log
 ├── memory/             ← AI-readable long-term memory
-├── skills/             ← automation scripts (Stage 2)
+│   └── memory.db       ← SQLite FTS5 search index
+├── plugins/            ← installed plugins
+│   └── <name>/
+│       ├── manifest.json
+│       └── plugin.py
+├── logs/               ← execution logs
+│   ├── tool_executions.log
+│   └── plugin_errors.log
 └── vault/              ← secure, AI-inaccessible storage
     └── secrets/api_keys.json
 ```
 
 ## Roadmap
 
-- **Stage 1** (now): Core CLI, multi-provider chat, memory, vault security
+- **Stage 1** ✅ Core CLI — multi-provider chat, memory, vault security
-- **Stage 2**: Skills — shell/PowerShell/Python automations with user approval gates
+- **Stage 2** ✅ Plugin Framework — extensible tools, slash commands, approval gates
-- **Stage 3**: Vault encryption with `age`
+- **Stage 3** ✅ Memory Database — SQLite + FTS5 full-text search index
-- **Stage 4**: Security audit sub-agent
+- **Stage 4** Vault Encryption — `age`-based encryption of `~/.pyra/vault/secrets/`
-- **Stage 5**: Web UI, embedding-based memory search
+- **Stage 5** Skills System — YAML-defined multi-plugin workflows with event triggers
+- **Stage 6** Daemon + Messaging Bots — always-on asyncio daemon, Matrix/Telegram/Signal bots
+- **Stage 7** Security Audit Sub-agent — automated scanning for injection, CVEs, permission drift
+- **Stage 8** Web UI — optional local interface, embedding-based memory search

pyproject.toml

@@ -35,6 +35,12 @@ gdrive    = ["google-api-python-client>=2.120.0", "google-auth-oauthlib>=1.2.0"]
 onedrive  = ["msal>=1.28.0"]
 dropbox   = ["dropbox>=12.0.0"]
 daemon    = ["aiofiles>=23.0.0"]
+email     = [
+    "imap-tools>=1.7.0",
+    "google-api-python-client>=2.120.0",
+    "google-auth-oauthlib>=1.2.0",
+    "O365>=2.0.36",
+]
 all-plugins = [
     "caldav>=1.3.0", "webdav4>=0.9.0", "vobject>=0.9.6",
     "matrix-nio>=0.24.0", "aiofiles>=23.0.0",
@@ -44,6 +50,8 @@ all-plugins = [
     "google-api-python-client>=2.120.0", "google-auth-oauthlib>=1.2.0",
     "msal>=1.28.0",
     "dropbox>=12.0.0",
+    "imap-tools>=1.7.0",
+    "O365>=2.0.36",
 ]
 
 [project.scripts]

src/pyra/chat/session.py

@@ -25,6 +25,7 @@ from pyra.plugins.executor import ToolExecutor
 from pyra.plugins.registry import PluginRegistry
 from pyra.security.injection import scan_response
 from pyra.setup.providers import get_provider
+from pyra.setup.wizard import fetch_loaded_models
 from pyra.utils.paths import pyra_home
 
 _HISTORY_FILE = pyra_home() / ".chat_history"
@@ -80,8 +81,9 @@ def start_chat() -> None:
         name="plan_and_execute",
         description=(
             "Decompose a multi-step task into sequential steps and execute each with "
-            "a focused sub-agent. Use when the request has multiple distinct phases. "
+            "a focused sub-agent. Use only for long tasks with 3 or more sequential "
-            "Specify 'agent' per step to route to a specialized agent."
+            "steps that need verification between them — for simple 2-step tasks, do "
+            "them directly. Specify 'agent' per step to route to a specialized agent."
         ),
         parameters={
             "type": "object",
@@ -124,7 +126,7 @@ def start_chat() -> None:
     ))
     registry.register_builtin(Tool(
         name="memory_read",
-        description="Read the full content of a memory file by its relative path (e.g. 'user/profile.md').",
+        description="Read the full content of a memory file by its relative path (e.g. 'user/profile.md'). Use memory_lookup first to find the correct path.",
         parameters={
             "type": "object",
             "properties": {
@@ -176,6 +178,16 @@ def start_chat() -> None:
         "[dim]Type /help for commands, /quit to exit.[/dim]"
     )
 
+    if provider.group == "Local":
+        loaded = fetch_loaded_models(provider)
+        if not loaded:
+            render_info(f"No model currently loaded in {provider.display_name}.")
+        elif cfg.ai.model not in loaded:
+            render_info(
+                f"Model '{cfg.ai.model}' not loaded in {provider.display_name}. "
+                f"Loaded: {', '.join(loaded)}"
+            )
+
     _flags: dict = {"use_tools": True}
 
     while True:

src/pyra/cli.py

@@ -174,7 +174,7 @@ def plugin_install(name: str) -> None:
         install_bundled_plugin(name, bundled_dir, plugins_dir)
         console.print(f"[green]Installed:[/green] {name}")
         console.print(f"  Enable:  [dim]pyra plugin enable {name}[/dim]")
-        console.print(f"  Confirm: [dim]pyra plugin setup {name}[/dim]")
+        console.print(f"  Configure: [dim]pyra plugin setup {name}[/dim]")
     except FileNotFoundError as exc:
         console.print(f"[red]Error:[/red] {exc}")
     except Exception as exc:
@@ -266,43 +266,144 @@ def daemon() -> None:
     _bootstrap_or_exit()
 
 
+@daemon.command("run", hidden=True)
+def daemon_run() -> None:
+    """Run daemon in foreground (used by service manager)."""
+    from pyra.daemon.core import run_foreground
+    run_foreground()
+
+
 @daemon.command("start")
 def daemon_start() -> None:
     """Start the Pyra daemon in the background."""
-    console.print("[yellow]Daemon (Stage 2.4) is not yet implemented.[/yellow]")
+    from pyra.daemon.core import start_background
+    try:
+        start_background()
+    except FileNotFoundError:
+        console.print("[red]Error:[/red] Run [dim]pyra setup[/dim] first.")
 
 
 @daemon.command("stop")
 def daemon_stop() -> None:
     """Stop the running Pyra daemon."""
-    console.print("[yellow]Daemon (Stage 2.4) is not yet implemented.[/yellow]")
+    _daemon_ipc("stop", success_msg="Daemon stopped.")
 
 
 @daemon.command("status")
 def daemon_status() -> None:
     """Show daemon status."""
-    console.print("[yellow]Daemon (Stage 2.4) is not yet implemented.[/yellow]")
+    _daemon_ipc("status")
 
 
 @daemon.command("restart")
 def daemon_restart() -> None:
     """Restart the Pyra daemon."""
-    console.print("[yellow]Daemon (Stage 2.4) is not yet implemented.[/yellow]")
+    import time
+    from pyra.daemon.core import start_background
+    _daemon_ipc("stop", success_msg=None)
+    time.sleep(1.5)
+    try:
+        start_background()
+    except FileNotFoundError:
+        console.print("[red]Error:[/red] Run [dim]pyra setup[/dim] first.")
 
 
 @daemon.command("install")
 def daemon_install() -> None:
-    """Install Pyra as a system service (launchd/systemd)."""
+    """Install Pyra as a system service (launchd/systemd/schtasks)."""
-    console.print("[yellow]Daemon service install (Stage 2.4) is not yet implemented.[/yellow]")
+    from pyra.daemon.service import detect_platform, install_service
+    try:
+        install_service()
+        console.print(f"[green]Service installed[/green] ({detect_platform()}).")
+    except Exception as exc:
+        console.print(f"[red]Install failed:[/red] {exc}")
 
 
 @daemon.command("uninstall")
 def daemon_uninstall() -> None:
     """Remove the Pyra system service."""
-    console.print("[yellow]Daemon service uninstall (Stage 2.4) is not yet implemented.[/yellow]")
+    from pyra.daemon.service import uninstall_service
+    try:
+        uninstall_service()
+        console.print("[green]Service removed.[/green]")
+    except Exception as exc:
+        console.print(f"[red]Uninstall failed:[/red] {exc}")
 
 
-@daemon.command("run", hidden=True)
+def _daemon_ipc(cmd: str, *, success_msg: str | None = None) -> None:
-def daemon_run() -> None:
+    """Send a command to the running daemon via IPC and render the response."""
-    """Run daemon in foreground (used by service manager)."""
+    from pyra.config.manager import load_config
-    console.print("[yellow]Daemon (Stage 2.4) is not yet implemented.[/yellow]")
+    from pyra.daemon.ipc import (
+        get_socket_path,
+        is_unix_socket,
+        get_port_file_path,
+        send_command,
+    )
+
+    try:
+        cfg = load_config()
+    except FileNotFoundError:
+        console.print("[red]Error:[/red] Run [dim]pyra setup[/dim] first.")
+        return
+
+    if is_unix_socket():
+        address = get_socket_path(cfg.daemon.socket_path)
+    else:
+        port = _read_windows_port()
+        if port is None:
+            console.print("[yellow]Daemon is not running.[/yellow]")
+            return
+        address = ("127.0.0.1", port)
+
+    try:
+        resp = send_command(address, {"cmd": cmd})
+    except (ConnectionRefusedError, FileNotFoundError, OSError):
+        console.print("[yellow]Daemon is not running.[/yellow]")
+        return
+    except ConnectionResetError:
+        console.print("[red]Permission denied:[/red] daemon rejected connection.")
+        return
+    except TimeoutError:
+        console.print("[red]Daemon did not respond in time.[/red]")
+        return
+
+    if not resp.get("ok"):
+        console.print(f"[red]Error:[/red] {resp.get('data', {}).get('error', 'unknown')}")
+        return
+
+    if cmd == "status":
+        _render_daemon_status(resp["data"])
+    elif success_msg:
+        console.print(f"[green]{success_msg}[/green]")
+
+
+def _read_windows_port() -> int | None:
+    from pyra.daemon.ipc import get_port_file_path
+    try:
+        return int(get_port_file_path().read_text().strip())
+    except (FileNotFoundError, ValueError):
+        return None
+
+
+def _render_daemon_status(data: dict) -> None:
+    from rich.table import Table
+
+    uptime = data.get("uptime", 0.0)
+    pid = data.get("pid", "?")
+    tasks = data.get("tasks", [])
+
+    hours, rem = divmod(int(uptime), 3600)
+    mins, secs = divmod(rem, 60)
+    uptime_str = f"{hours}h {mins}m {secs}s" if hours else f"{mins}m {secs}s"
+
+    console.print(f"[bold green]Daemon running[/bold green] — PID {pid}, uptime {uptime_str}")
+
+    if tasks:
+        table = Table("Task", "Alive", "Restarts", "Last error", show_header=True)
+        for t in tasks:
+            alive = "[green]yes[/green]" if t.get("alive") else "[red]no[/red]"
+            error = t.get("last_error") or "—"
+            table.add_row(t.get("name", "?"), alive, str(t.get("restart_count", 0)), error)
+        console.print(table)
+    else:
+        console.print("[dim]No plugin tasks registered.[/dim]")

src/pyra/config/schema.py

@@ -36,6 +36,7 @@ class DaemonConfig(BaseModel):
     socket_path: str = "~/.pyra/daemon.sock"
     log_file: str = "~/.pyra/daemon.log"
     pid_file: str = "~/.pyra/daemon.pid"
+    ipc_port: int = 0  # Windows TCP loopback: 0 = OS-assigned, written to ~/.pyra/daemon.port
 
 
 class PyraConfig(BaseModel):

src/pyra/daemon/__init__.py

@@ -0,0 +1,24 @@
+"""Pyra background daemon package."""
+
+from pyra.daemon.core import PluginSupervisor, run_foreground, start_background
+from pyra.daemon.events import publish, subscribe_forever
+from pyra.daemon.ipc import IpcClient, IpcServer, send_command
+from pyra.daemon.pid import PidFile, PidFileError, resolve_pid_path
+from pyra.daemon.service import detect_platform, install_service, uninstall_service
+
+__all__ = [
+    "run_foreground",
+    "start_background",
+    "PluginSupervisor",
+    "publish",
+    "subscribe_forever",
+    "IpcClient",
+    "IpcServer",
+    "send_command",
+    "PidFile",
+    "PidFileError",
+    "resolve_pid_path",
+    "detect_platform",
+    "install_service",
+    "uninstall_service",
+]

src/pyra/daemon/core.py

@@ -0,0 +1,313 @@
+"""Pyra daemon core — asyncio event loop, plugin task supervisor, signal handling."""
+
+from __future__ import annotations
+
+import asyncio
+import logging
+import logging.handlers
+import os
+import signal
+import subprocess
+import sys
+import time
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Callable, Coroutine
+
+from pyra.utils.paths import pyra_home, safe_chmod
+
+
+_log = logging.getLogger("pyra.daemon")
+_start_time: float = 0.0
+
+
+# ── Plugin task supervisor ────────────────────────────────────────────────────
+
+@dataclass
+class TaskRecord:
+    name: str
+    coro_factory: Callable[[], Coroutine]  # type: ignore[type-arg]
+    task: asyncio.Task | None = field(default=None, repr=False)
+    restart_count: int = 0
+    last_error: str | None = None
+
+    def is_alive(self) -> bool:
+        return self.task is not None and not self.task.done()
+
+
+class PluginSupervisor:
+    _RESTART_DELAY: float = 5.0
+    _MAX_RESTARTS: int = 10
+
+    def __init__(self) -> None:
+        self._records: list[TaskRecord] = []
+        self._shutdown = asyncio.Event()
+
+    def add_task(self, name: str, factory: Callable[[], Coroutine]) -> None:  # type: ignore[type-arg]
+        self._records.append(TaskRecord(name=name, coro_factory=factory))
+
+    async def start(self) -> None:
+        for record in self._records:
+            record.task = asyncio.create_task(
+                self._supervise(record), name=record.name
+            )
+        _log.info("Supervisor started with %d plugin task(s).", len(self._records))
+
+    async def run_until_shutdown(self) -> None:
+        await self._shutdown.wait()
+        _log.info("Shutdown requested — stopping supervisor.")
+
+    def request_shutdown(self) -> None:
+        self._shutdown.set()
+
+    async def stop(self) -> None:
+        for record in self._records:
+            if record.task and not record.task.done():
+                record.task.cancel()
+        tasks = [r.task for r in self._records if r.task]
+        if tasks:
+            await asyncio.gather(*tasks, return_exceptions=True)
+
+    async def reload(self) -> None:
+        """Cancel all running tasks and restart them with fresh coroutines."""
+        for record in self._records:
+            if record.task and not record.task.done():
+                record.task.cancel()
+                try:
+                    await record.task
+                except (asyncio.CancelledError, Exception):
+                    pass
+            record.restart_count = 0
+            record.last_error = None
+            record.task = asyncio.create_task(
+                self._supervise(record), name=record.name
+            )
+        _log.info("Reloaded %d plugin task(s).", len(self._records))
+
+    def status(self) -> list[dict]:
+        return [
+            {
+                "name": r.name,
+                "alive": r.is_alive(),
+                "restart_count": r.restart_count,
+                "last_error": r.last_error,
+            }
+            for r in self._records
+        ]
+
+    async def _supervise(self, record: TaskRecord) -> None:
+        while not self._shutdown.is_set():
+            try:
+                await record.coro_factory()
+                _log.info("Plugin task %s completed normally.", record.name)
+                return
+            except asyncio.CancelledError:
+                return
+            except Exception as exc:
+                record.restart_count += 1
+                record.last_error = f"{type(exc).__name__}: {exc}"
+                _log.error(
+                    "Plugin task %s crashed (restart #%d): %s",
+                    record.name, record.restart_count, exc,
+                    exc_info=True,
+                )
+                if record.restart_count >= self._MAX_RESTARTS:
+                    _log.critical(
+                        "Plugin task %s exceeded max restarts (%d). Giving up.",
+                        record.name, self._MAX_RESTARTS,
+                    )
+                    return
+                try:
+                    await asyncio.wait_for(
+                        asyncio.sleep(self._RESTART_DELAY),
+                        timeout=self._RESTART_DELAY + 1,
+                    )
+                except asyncio.CancelledError:
+                    return
+
+
+# ── IPC command dispatch ──────────────────────────────────────────────────────
+
+def _make_ipc_handler(supervisor: PluginSupervisor):
+    async def handler(msg: dict) -> dict:
+        cmd = msg.get("cmd", "")
+        match cmd:
+            case "ping":
+                return {"ok": True, "data": {"pong": True}}
+            case "status":
+                return {
+                    "ok": True,
+                    "data": {
+                        "uptime": time.monotonic() - _start_time,
+                        "pid": os.getpid(),
+                        "tasks": supervisor.status(),
+                    },
+                }
+            case "stop":
+                supervisor.request_shutdown()
+                return {"ok": True, "data": {}}
+            case "reload":
+                await supervisor.reload()
+                return {"ok": True, "data": {"tasks_reloaded": len(supervisor._records)}}
+            case _:
+                return {"ok": False, "data": {"error": f"unknown command: {cmd}"}}
+
+    return handler
+
+
+# ── Main async entrypoint ─────────────────────────────────────────────────────
+
+async def _run_daemon(cfg, supervisor: PluginSupervisor) -> None:
+    from pyra.daemon.ipc import IpcServer, get_socket_path, is_unix_socket
+
+    # Install signal handlers now that the event loop is running.
+    _install_signal_handlers(supervisor)
+
+    if is_unix_socket():
+        address = get_socket_path(cfg.daemon.socket_path)
+    else:
+        address = ("127.0.0.1", cfg.daemon.ipc_port)
+
+    server = IpcServer(address, _make_ipc_handler(supervisor))
+
+    await supervisor.start()
+
+    async with asyncio.TaskGroup() as tg:
+        tg.create_task(server.start(), name="ipc_server")
+        tg.create_task(supervisor.run_until_shutdown(), name="shutdown_waiter")
+
+    await server.stop()
+    await supervisor.stop()
+
+
+# ── Foreground entry point (pyra daemon run) ──────────────────────────────────
+
+def run_foreground() -> None:
+    """Run the daemon in the foreground. Called by `pyra daemon run`."""
+    from pyra.config.manager import load_config
+    from pyra.daemon.pid import PidFile, PidFileError, resolve_pid_path
+    from pyra.plugins.registry import PluginRegistry
+
+    global _start_time
+
+    cfg = load_config()
+    _setup_logging(cfg.daemon.log_file)
+    pid_path = resolve_pid_path(cfg.daemon.pid_file)
+    pid_file = PidFile(pid_path)
+
+    existing = pid_file.read()
+    if existing is not None and not pid_file.is_stale():
+        _log.error("Daemon already running (PID %d). Exiting.", existing)
+        sys.exit(1)
+
+    registry = PluginRegistry()
+    from pyra.utils.paths import pyra_home as _pyra_home
+    plugins_dir = _pyra_home() / "plugins"
+    if plugins_dir.exists():
+        registry.load_all(plugins_dir, cfg.plugins.enabled)
+
+    supervisor = PluginSupervisor()
+    for name, factory in registry.get_daemon_task_factories():
+        supervisor.add_task(name, factory)
+
+    _start_time = time.monotonic()
+
+    try:
+        with pid_file:
+            _log.info("Pyra daemon starting (PID %d).", os.getpid())
+            try:
+                asyncio.run(_run_daemon(cfg, supervisor))
+            except KeyboardInterrupt:
+                pass
+            _log.info("Pyra daemon stopped.")
+    except PidFileError as exc:
+        _log.error("Could not acquire PID file: %s", exc)
+        sys.exit(1)
+
+
+# ── Background spawn (pyra daemon start) ─────────────────────────────────────
+
+def start_background() -> None:
+    """Spawn `pyra daemon run` as a detached background process."""
+    from pyra.config.manager import load_config
+    from pyra.daemon.pid import PidFile, resolve_pid_path
+    from pyra.daemon.service import find_pyra_executable
+
+    cfg = load_config()
+    pid_path = resolve_pid_path(cfg.daemon.pid_file)
+    pid_file = PidFile(pid_path)
+
+    existing = pid_file.read()
+    if existing is not None and not pid_file.is_stale():
+        from pyra.chat.renderer import console
+        console.print(f"[yellow]Daemon already running (PID {existing}).[/yellow]")
+        return
+
+    exe = find_pyra_executable()
+    log_path = Path(cfg.daemon.log_file).expanduser()
+    log_path.parent.mkdir(parents=True, exist_ok=True)
+
+    with open(log_path, "a") as log_fh:
+        if sys.platform == "win32":
+            DETACHED_PROCESS = 0x00000008
+            CREATE_NEW_PROCESS_GROUP = 0x00000200
+            subprocess.Popen(
+                [exe, "daemon", "run"],
+                creationflags=DETACHED_PROCESS | CREATE_NEW_PROCESS_GROUP,
+                stdout=log_fh,
+                stderr=log_fh,
+                close_fds=True,
+            )
+        else:
+            subprocess.Popen(
+                [exe, "daemon", "run"],
+                start_new_session=True,
+                stdout=log_fh,
+                stderr=log_fh,
+                stdin=subprocess.DEVNULL,
+                close_fds=True,
+            )
+
+    from pyra.chat.renderer import console
+
+    # Poll the PID file for up to 3 seconds to confirm startup.
+    for _ in range(30):
+        time.sleep(0.1)
+        pid = pid_file.read()
+        if pid is not None:
+            console.print(f"[green]Daemon started (PID {pid}).[/green]")
+            return
+
+    console.print("[yellow]Daemon process spawned but PID file not yet written.[/yellow]")
+
+
+# ── Signal handling ───────────────────────────────────────────────────────────
+
+def _install_signal_handlers(supervisor: PluginSupervisor) -> None:
+    if sys.platform == "win32":
+        signal.signal(signal.SIGTERM, lambda *_: supervisor.request_shutdown())
+        return
+
+    loop = asyncio.get_running_loop()
+    loop.add_signal_handler(signal.SIGTERM, supervisor.request_shutdown)
+    loop.add_signal_handler(signal.SIGHUP, supervisor.request_shutdown)
+
+
+# ── Logging setup ─────────────────────────────────────────────────────────────
+
+def _setup_logging(log_file_str: str) -> None:
+    log_path = Path(log_file_str).expanduser()
+    log_path.parent.mkdir(parents=True, exist_ok=True)
+
+    handler = logging.handlers.RotatingFileHandler(
+        log_path, maxBytes=5 * 1024 * 1024, backupCount=3
+    )
+    handler.setFormatter(
+        logging.Formatter("%(asctime)s %(levelname)s %(name)s %(message)s")
+    )
+
+    root = logging.getLogger("pyra")
+    root.addHandler(handler)
+    root.setLevel(logging.INFO)
+
+    safe_chmod(log_path, 0o600)

src/pyra/daemon/events.py

@@ -0,0 +1,46 @@
+"""Async notification bus for inter-plugin communication in the daemon.
+
+Plugins publish events to a shared asyncio.Queue; other plugins (e.g. messaging
+bots) consume them via subscribe_forever().  No direct plugin-to-plugin imports
+are needed — both sides just use this module.
+
+Event shape (by convention):
+    {"type": "new_email", "priority": int, "from": str, "subject": str,
+     "summary": str, "uid": str, "folder": str}
+    {"type": "new_message", "bot": str, "user_id": str, "text": str}
+"""
+from __future__ import annotations
+
+import asyncio
+from typing import Any, AsyncGenerator
+
+_queue: asyncio.Queue[dict[str, Any]] | None = None
+
+
+def get_queue() -> asyncio.Queue[dict[str, Any]]:
+    global _queue
+    if _queue is None:
+        _queue = asyncio.Queue(maxsize=200)
+    return _queue
+
+
+async def publish(event: dict[str, Any]) -> None:
+    """Emit an event. Drops silently if the queue is full (daemon is overloaded)."""
+    q = get_queue()
+    try:
+        q.put_nowait(event)
+    except asyncio.QueueFull:
+        pass
+
+
+async def subscribe_forever() -> AsyncGenerator[dict[str, Any], None]:
+    """Async generator — yields events as they arrive. Intended for daemon tasks."""
+    q = get_queue()
+    while True:
+        yield await q.get()
+
+
+def reset() -> None:
+    """Discard the current queue and create a fresh one. FOR TESTS ONLY."""
+    global _queue
+    _queue = None

src/pyra/daemon/ipc.py

@@ -0,0 +1,241 @@
+"""IPC transport for the Pyra daemon.
+
+Linux/macOS: Unix domain socket at ~/.pyra/daemon.sock (chmod 600, UID-checked).
+Windows:     TCP loopback on an OS-assigned port; actual port written to
+             ~/.pyra/daemon.port so clients can connect without knowing it ahead
+             of time.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+import os
+import struct
+import sys
+from pathlib import Path
+from typing import Any, Awaitable, Callable
+
+
+# ── Protocol types ────────────────────────────────────────────────────────────
+
+IpcMessage = dict[str, Any]   # must have "cmd" key
+IpcResponse = dict[str, Any]  # must have "ok" and "data" keys
+
+
+# ── Encode / decode ───────────────────────────────────────────────────────────
+
+def encode_message(msg: IpcMessage) -> bytes:
+    return (json.dumps(msg) + "\n").encode()
+
+
+def decode_message(line: bytes) -> IpcMessage:
+    try:
+        return json.loads(line.rstrip(b"\n"))
+    except json.JSONDecodeError as exc:
+        raise ValueError(f"Invalid IPC message: {exc}") from exc
+
+
+# ── Address helpers ───────────────────────────────────────────────────────────
+
+def is_unix_socket() -> bool:
+    return sys.platform != "win32"
+
+
+def get_socket_path(cfg_socket_path: str) -> Path:
+    """Expand ~ and return the Unix socket path."""
+    return Path(cfg_socket_path).expanduser()
+
+
+def get_port_file_path() -> Path:
+    from pyra.utils.paths import pyra_home
+    return pyra_home() / "daemon.port"
+
+
+def _read_windows_port() -> int | None:
+    port_file = get_port_file_path()
+    try:
+        return int(port_file.read_text().strip())
+    except (FileNotFoundError, ValueError):
+        return None
+
+
+# ── Server ────────────────────────────────────────────────────────────────────
+
+class IpcServer:
+    def __init__(
+        self,
+        address: Path | tuple[str, int],
+        handler: Callable[[IpcMessage], Awaitable[IpcResponse]],
+    ) -> None:
+        self._address = address
+        self._handler = handler
+        self._server: asyncio.Server | None = None
+
+    async def start(self) -> None:
+        if is_unix_socket():
+            assert isinstance(self._address, Path)
+            sock_path = self._address
+            if sock_path.exists():
+                sock_path.unlink()
+            self._server = await asyncio.start_unix_server(
+                self._handle_client, path=str(sock_path)
+            )
+            os.chmod(sock_path, 0o600)
+        else:
+            host, port = self._address if isinstance(self._address, tuple) else ("127.0.0.1", 0)
+            self._server = await asyncio.start_server(
+                self._handle_client, host=host, port=port
+            )
+            actual_port = self._server.sockets[0].getsockname()[1]
+            port_file = get_port_file_path()
+            port_file.write_text(str(actual_port))
+
+        await self._server.start_serving()
+
+    async def stop(self) -> None:
+        if self._server is not None:
+            self._server.close()
+            try:
+                await asyncio.wait_for(self._server.wait_closed(), timeout=5.0)
+            except asyncio.TimeoutError:
+                pass
+        if is_unix_socket() and isinstance(self._address, Path):
+            try:
+                self._address.unlink()
+            except FileNotFoundError:
+                pass
+        else:
+            port_file = get_port_file_path()
+            try:
+                port_file.unlink()
+            except FileNotFoundError:
+                pass
+
+    async def _handle_client(
+        self, reader: asyncio.StreamReader, writer: asyncio.StreamWriter
+    ) -> None:
+        try:
+            if is_unix_socket() and not self._check_peer_uid(writer):
+                writer.close()
+                return
+
+            line = await asyncio.wait_for(reader.readline(), timeout=5.0)
+            if not line:
+                return
+
+            try:
+                msg = decode_message(line)
+            except ValueError:
+                resp: IpcResponse = {"ok": False, "data": {"error": "invalid JSON"}}
+            else:
+                resp = await self._handler(msg)
+
+            writer.write(encode_message(resp))
+            await writer.drain()
+        except (asyncio.TimeoutError, ConnectionResetError, BrokenPipeError):
+            pass
+        finally:
+            try:
+                writer.close()
+                await writer.wait_closed()
+            except Exception:
+                pass
+
+    def _check_peer_uid(self, writer: asyncio.StreamWriter) -> bool:
+        """Return True if the peer's UID matches ours. Falls back to True on error."""
+        try:
+            peer_uid = _get_peer_uid(writer)
+            if peer_uid is None:
+                return True  # can't determine — allow (socket perms are the guard)
+            return peer_uid == os.getuid()
+        except Exception:
+            return True  # don't crash the server on unexpected errors
+
+
+# ── Client ────────────────────────────────────────────────────────────────────
+
+class IpcClient:
+    def __init__(self, address: Path | tuple[str, int]) -> None:
+        self._address = address
+
+    async def send(self, msg: IpcMessage, timeout: float = 5.0) -> IpcResponse:
+        if is_unix_socket():
+            assert isinstance(self._address, Path)
+            reader, writer = await asyncio.wait_for(
+                asyncio.open_unix_connection(str(self._address)), timeout=timeout
+            )
+        else:
+            host, port = self._address if isinstance(self._address, tuple) else ("127.0.0.1", 0)
+            reader, writer = await asyncio.wait_for(
+                asyncio.open_connection(host, port), timeout=timeout
+            )
+
+        try:
+            writer.write(encode_message(msg))
+            await writer.drain()
+            line = await asyncio.wait_for(reader.readline(), timeout=timeout)
+            return decode_message(line)
+        finally:
+            writer.close()
+            try:
+                await writer.wait_closed()
+            except Exception:
+                pass
+
+
+def send_command(
+    address: Path | tuple[str, int],
+    msg: IpcMessage,
+    timeout: float = 5.0,
+) -> IpcResponse:
+    """Synchronous wrapper around IpcClient.send() for CLI callers."""
+    return asyncio.run(IpcClient(address).send(msg, timeout=timeout))
+
+
+# ── Peer UID detection ────────────────────────────────────────────────────────
+
+def _get_peer_uid(writer: asyncio.StreamWriter) -> int | None:
+    """Return the connecting peer's UID, or None if unavailable."""
+    try:
+        sock = writer.get_extra_info("socket")
+        if sock is None:
+            return None
+
+        if sys.platform == "linux":
+            # SO_PEERCRED: struct { pid_t pid; uid_t uid; gid_t gid; }
+            SO_PEERCRED = 17
+            cred = sock.getsockopt(
+                socket_module().SOL_SOCKET, SO_PEERCRED, struct.calcsize("3i")
+            )
+            _pid, uid, _gid = struct.unpack("3i", cred)
+            return uid
+
+        if sys.platform == "darwin":
+            return _macos_peer_uid(sock.fileno())
+
+    except Exception:
+        pass
+    return None
+
+
+def socket_module():  # lazy import to avoid top-level import on non-Unix
+    import socket
+    return socket
+
+
+def _macos_peer_uid(fd: int) -> int | None:
+    """Use getpeereid(2) via ctypes to retrieve the peer UID on macOS."""
+    import ctypes
+    import ctypes.util
+
+    libc_name = ctypes.util.find_library("c")
+    if not libc_name:
+        return None
+    libc = ctypes.CDLL(libc_name)
+
+    euid = ctypes.c_uint32(0)
+    egid = ctypes.c_uint32(0)
+    if libc.getpeereid(fd, ctypes.byref(euid), ctypes.byref(egid)) != 0:
+        return None
+    return euid.value

src/pyra/daemon/pid.py

@@ -0,0 +1,94 @@
+"""PID file management for the Pyra daemon."""
+
+from __future__ import annotations
+
+import os
+import sys
+from pathlib import Path
+
+
+class PidFileError(OSError):
+    """Raised when a PID file operation fails due to a live conflicting process."""
+
+
+class PidFile:
+    def __init__(self, path: Path) -> None:
+        self._path = path
+
+    def write(self) -> None:
+        """Write the current PID atomically.
+
+        Raises PidFileError if a non-stale PID file already exists.
+        """
+        existing = self.read()
+        if existing is not None and not self.is_stale():
+            raise PidFileError(
+                f"Daemon already running with PID {existing} "
+                f"(PID file: {self._path})"
+            )
+        tmp = self._path.with_suffix(".pid.tmp")
+        tmp.write_text(str(os.getpid()))
+        tmp.replace(self._path)
+
+    def read(self) -> int | None:
+        """Return the PID from the file, or None if the file is absent or unreadable."""
+        try:
+            return int(self._path.read_text().strip())
+        except (FileNotFoundError, ValueError):
+            return None
+
+    def is_stale(self) -> bool:
+        """True when the PID file exists but the process no longer runs."""
+        pid = self.read()
+        if pid is None:
+            return False
+        return not _process_is_alive(pid)
+
+    def remove(self) -> None:
+        """Delete the PID file, ignoring FileNotFoundError."""
+        try:
+            self._path.unlink()
+        except FileNotFoundError:
+            pass
+
+    def __enter__(self) -> "PidFile":
+        self.write()
+        return self
+
+    def __exit__(self, *_: object) -> None:
+        self.remove()
+
+
+def resolve_pid_path(cfg_path: str) -> Path:
+    """Expand ~ and return an absolute Path."""
+    return Path(cfg_path).expanduser().resolve()
+
+
+# ── Platform-specific process liveness check ─────────────────────────────────
+
+def _process_is_alive(pid: int) -> bool:
+    if sys.platform == "win32":
+        return _win_process_is_alive(pid)
+    return _posix_process_is_alive(pid)
+
+
+def _posix_process_is_alive(pid: int) -> bool:
+    try:
+        os.kill(pid, 0)
+        return True
+    except ProcessLookupError:
+        return False
+    except PermissionError:
+        # Process exists but is owned by another user — still alive.
+        return True
+
+
+def _win_process_is_alive(pid: int) -> bool:
+    import ctypes
+
+    SYNCHRONIZE = 0x00100000
+    handle = ctypes.windll.kernel32.OpenProcess(SYNCHRONIZE, False, pid)  # type: ignore[attr-defined]
+    if handle == 0:
+        return False
+    ctypes.windll.kernel32.CloseHandle(handle)  # type: ignore[attr-defined]
+    return True

src/pyra/daemon/service.py

@@ -0,0 +1,212 @@
+"""OS-specific service file generation and install/uninstall for the Pyra daemon."""
+
+from __future__ import annotations
+
+import platform
+import shutil
+import subprocess
+import sys
+from pathlib import Path
+from typing import Literal
+
+from pyra.utils.paths import safe_chmod
+
+
+def detect_platform() -> Literal["macos", "linux", "windows"]:
+    s = platform.system()
+    if s == "Darwin":
+        return "macos"
+    if s == "Linux":
+        return "linux"
+    if s == "Windows":
+        return "windows"
+    raise RuntimeError(f"Unsupported platform: {s}")
+
+
+def find_pyra_executable() -> str:
+    """Return the full path to the active pyra executable.
+
+    Tries, in order:
+    1. shutil.which("pyra") — works when pyra is on PATH (activated venv)
+    2. sys.executable's sibling "pyra" script — covers editable installs
+    3. Fallback: sys.executable -m pyra
+    """
+    found = shutil.which("pyra")
+    if found:
+        return found
+
+    sibling = Path(sys.executable).parent / "pyra"
+    if sibling.exists():
+        return str(sibling)
+
+    return f"{sys.executable} -m pyra"
+
+
+# ── Template generators ───────────────────────────────────────────────────────
+
+def render_launchd_plist(exe: str, log_file: str, pid_file: str) -> str:
+    log = str(Path(log_file).expanduser())
+    return f"""<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN"
+    "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+    <key>Label</key>
+    <string>com.pyra.daemon</string>
+    <key>ProgramArguments</key>
+    <array>
+        <string>{exe}</string>
+        <string>daemon</string>
+        <string>run</string>
+    </array>
+    <key>RunAtLoad</key>
+    <true/>
+    <key>KeepAlive</key>
+    <true/>
+    <key>StandardOutPath</key>
+    <string>{log}</string>
+    <key>StandardErrorPath</key>
+    <string>{log}</string>
+    <key>ProcessType</key>
+    <string>Background</string>
+</dict>
+</plist>
+"""
+
+
+def render_systemd_unit(exe: str, log_file: str) -> str:
+    log = str(Path(log_file).expanduser())
+    return f"""[Unit]
+Description=Pyra Personal AI Assistant Daemon
+After=default.target
+
+[Service]
+Type=simple
+ExecStart={exe} daemon run
+Restart=on-failure
+RestartSec=5s
+StandardOutput=append:{log}
+StandardError=append:{log}
+
+[Install]
+WantedBy=default.target
+"""
+
+
+def render_schtasks_xml(exe: str) -> str:
+    return f"""<?xml version="1.0" encoding="UTF-16"?>
+<Task version="1.2" xmlns="http://schemas.microsoft.com/windows/2004/02/mit/task">
+  <RegistrationInfo>
+    <Description>Pyra Personal AI Assistant Daemon</Description>
+  </RegistrationInfo>
+  <Triggers>
+    <LogonTrigger>
+      <Enabled>true</Enabled>
+    </LogonTrigger>
+  </Triggers>
+  <Settings>
+    <MultipleInstancesPolicy>IgnoreNew</MultipleInstancesPolicy>
+    <DisallowStartIfOnBatteries>false</DisallowStartIfOnBatteries>
+    <StopIfGoingOnBatteries>false</StopIfGoingOnBatteries>
+    <ExecutionTimeLimit>PT0S</ExecutionTimeLimit>
+    <RestartOnFailure>
+      <Interval>PT1M</Interval>
+      <Count>999</Count>
+    </RestartOnFailure>
+  </Settings>
+  <Actions Context="Author">
+    <Exec>
+      <Command>{exe}</Command>
+      <Arguments>daemon run</Arguments>
+    </Exec>
+  </Actions>
+</Task>
+"""
+
+
+# ── Install / uninstall ───────────────────────────────────────────────────────
+
+def install_service() -> None:
+    """Generate and register the OS service for the current platform."""
+    from pyra.config.manager import load_config
+
+    cfg = load_config()
+    exe = find_pyra_executable()
+    plat = detect_platform()
+
+    if plat == "macos":
+        _install_launchd(exe, cfg.daemon.log_file, cfg.daemon.pid_file)
+    elif plat == "linux":
+        _install_systemd(exe, cfg.daemon.log_file)
+    else:
+        _install_windows(exe)
+
+
+def uninstall_service() -> None:
+    """Deregister the OS service for the current platform."""
+    plat = detect_platform()
+    if plat == "macos":
+        _uninstall_launchd()
+    elif plat == "linux":
+        _uninstall_systemd()
+    else:
+        _uninstall_windows()
+
+
+# ── macOS launchd ─────────────────────────────────────────────────────────────
+
+_PLIST_PATH = Path.home() / "Library" / "LaunchAgents" / "com.pyra.daemon.plist"
+
+
+def _install_launchd(exe: str, log_file: str, pid_file: str) -> None:
+    _PLIST_PATH.parent.mkdir(parents=True, exist_ok=True)
+    _PLIST_PATH.write_text(render_launchd_plist(exe, log_file, pid_file))
+    safe_chmod(_PLIST_PATH, 0o644)  # launchd requires 644, not 600
+    subprocess.run(["launchctl", "load", str(_PLIST_PATH)], check=True)
+
+
+def _uninstall_launchd() -> None:
+    if _PLIST_PATH.exists():
+        subprocess.run(["launchctl", "unload", str(_PLIST_PATH)], check=False)
+        _PLIST_PATH.unlink()
+
+
+# ── Linux systemd ─────────────────────────────────────────────────────────────
+
+_SYSTEMD_UNIT = Path.home() / ".config" / "systemd" / "user" / "pyra.service"
+
+
+def _install_systemd(exe: str, log_file: str) -> None:
+    _SYSTEMD_UNIT.parent.mkdir(parents=True, exist_ok=True)
+    _SYSTEMD_UNIT.write_text(render_systemd_unit(exe, log_file))
+    subprocess.run(["systemctl", "--user", "daemon-reload"], check=True)
+    subprocess.run(["systemctl", "--user", "enable", "pyra"], check=True)
+
+
+def _uninstall_systemd() -> None:
+    subprocess.run(
+        ["systemctl", "--user", "disable", "--now", "pyra"], check=False
+    )
+    if _SYSTEMD_UNIT.exists():
+        _SYSTEMD_UNIT.unlink()
+    subprocess.run(["systemctl", "--user", "daemon-reload"], check=False)
+
+
+# ── Windows Task Scheduler ────────────────────────────────────────────────────
+
+def _install_windows(exe: str) -> None:
+    from pyra.utils.paths import pyra_home
+
+    xml_path = pyra_home() / "daemon_task.xml"
+    # schtasks /Create /XML requires UTF-16 encoding
+    xml_path.write_text(render_schtasks_xml(exe), encoding="utf-16")
+    subprocess.run(
+        ["schtasks", "/Create", "/TN", "PyraAssistant", "/XML", str(xml_path), "/F"],
+        check=True,
+    )
+
+
+def _uninstall_windows() -> None:
+    subprocess.run(
+        ["schtasks", "/Delete", "/TN", "PyraAssistant", "/F"], check=False
+    )

src/pyra/plugins/registry.py

@@ -75,6 +75,32 @@ class PluginRegistry:
                 pass
         return tasks
 
+    def get_daemon_task_factories(
+        self,
+    ) -> list[tuple[str, Callable[[], Coroutine]]]:  # type: ignore[type-arg]
+        """Return (name, factory) pairs for all plugin daemon tasks.
+
+        Each factory re-calls plugin.daemon_tasks() to produce a fresh coroutine,
+        enabling the supervisor to restart crashed tasks without changing the plugin
+        protocol.
+        """
+        factories: list[tuple[str, Callable[[], Coroutine]]] = []  # type: ignore[type-arg]
+        for plugin in self._plugins.values():
+            try:
+                initial = plugin.daemon_tasks()
+                n_tasks = len(initial)
+                for c in initial:
+                    c.close()  # prevent "coroutine never awaited" RuntimeWarning
+            except Exception:
+                continue
+            for i in range(n_tasks):
+                name = f"{plugin.name}.task_{i}"
+                # Capture plugin and index by value so each closure is independent.
+                def _factory(p=plugin, idx=i) -> Coroutine:  # type: ignore[type-arg]
+                    return p.daemon_tasks()[idx]
+                factories.append((name, _factory))
+        return factories
+
     def find_tool(self, name: str) -> Tool | None:
         return self._tools.get(name)
 

src/pyra/setup/wizard.py

@@ -1,3 +1,6 @@
+import contextlib
+import json
+
 import httpx
 import questionary
 from rich.console import Console
@@ -7,6 +10,7 @@ from rich.text import Text
 from pyra.config.manager import save_config
 from pyra.config.schema import GeneralConfig, ProviderConfig, PyraConfig
 from pyra.setup.providers import PROVIDERS, Provider, get_provider
+from pyra.utils.paths import pyra_home, safe_chmod
 
 console = Console()
 
@@ -20,6 +24,72 @@ _USE_CASE_PLUGINS: dict[str, list[str]] = {
 }
 
 
+_DRAFT_FILE = "setup.draft.json"
+
+
+def _draft_path():
+    return pyra_home() / _DRAFT_FILE
+
+
+def _save_draft(state: dict) -> None:
+    path = _draft_path()
+    path.write_text(json.dumps(state, indent=2))
+    safe_chmod(path, 0o600)
+
+
+def _load_draft() -> dict | None:
+    path = _draft_path()
+    if not path.exists():
+        return None
+    try:
+        return json.loads(path.read_text())
+    except Exception:
+        return None
+
+
+def _delete_draft() -> None:
+    with contextlib.suppress(FileNotFoundError):
+        _draft_path().unlink()
+
+
+def _mark_step_done(state: dict, step: str) -> None:
+    state.setdefault("completed_steps", [])
+    if step not in state["completed_steps"]:
+        state["completed_steps"].append(step)
+
+
+def _offer_resume(draft: dict) -> bool:
+    """Show a summary of the incomplete setup and ask Resume / Start fresh."""
+    completed = draft.get("completed_steps", [])
+    step_display = {
+        "profile":    f"Profile:    {draft.get('user_name', '?')}",
+        "provider":   f"Provider:   {draft.get('provider_id', '?')}",
+        "model":      f"Model:      {draft.get('model', '?')}",
+        "api_key":    "API key:    stored in vault",
+        "connection": "Connection: test passed",
+    }
+    lines = ["[bold]An incomplete setup was found.[/bold]\n"]
+    for step, label in step_display.items():
+        if step in completed:
+            lines.append(f"  [green]✓[/green] {label}")
+        else:
+            lines.append(f"  [dim]○ {label.split(':')[0].strip()}: pending[/dim]")
+
+    console.print(Panel("\n".join(lines), title="Incomplete setup", border_style="yellow"))
+    console.print()
+
+    action = questionary.select(
+        "What would you like to do?",
+        choices=[
+            questionary.Choice("Resume from where you left off", value="resume"),
+            questionary.Choice("Start fresh", value="fresh"),
+        ],
+    ).ask()
+    if action is None:
+        raise SystemExit(0)
+    return action == "resume"
+
+
 def run_setup() -> None:
     console.print(Panel(
         Text("Welcome to Pyra Setup", justify="center", style="bold cyan"),
@@ -28,36 +98,92 @@ def run_setup() -> None:
     ))
     console.print()
 
-    user_name, purpose, use_cases = _collect_user_profile()
+    state: dict = {}
+    draft = _load_draft()
+    if draft:
+        if _offer_resume(draft):
+            state = draft
+        else:
+            _delete_draft()
 
-    provider = _choose_provider()
+    try:
-    model = _choose_model(provider)
+        # ── Step 1: profile ────────────────────────────────────────────────
+        if "profile" in state.get("completed_steps", []):
+            user_name = state["user_name"]
+            purpose = state["purpose"]
+            use_cases = state["use_cases"]
+            console.print(f"  [dim]✓ Profile: {user_name}[/dim]")
+        else:
+            user_name, purpose, use_cases = _collect_user_profile()
+            state.update(user_name=user_name, purpose=purpose, use_cases=use_cases)
+            _mark_step_done(state, "profile")
+            _save_draft(state)
 
-    if provider.requires_key:
+        # ── Step 2: provider ───────────────────────────────────────────────
-        _collect_api_key(provider)
+        if "provider" in state.get("completed_steps", []):
+            provider = get_provider(state["provider_id"])
+            console.print(f"  [dim]✓ Provider: {provider.display_name}[/dim]")
+        else:
+            provider = _choose_provider()
+            state.update(provider_id=provider.id)
+            _mark_step_done(state, "provider")
+            _save_draft(state)
 
-    _test_connection(provider, model)
+        # ── Step 3: model ──────────────────────────────────────────────────
+        if "model" in state.get("completed_steps", []):
+            model = state["model"]
+            console.print(f"  [dim]✓ Model: {model}[/dim]")
+        else:
+            model = _choose_model(provider)
+            state.update(model=model)
+            _mark_step_done(state, "model")
+            _save_draft(state)
 
-    cfg = PyraConfig(
+        # ── Step 4: API key ────────────────────────────────────────────────
-        ai=ProviderConfig(
+        if "api_key" not in state.get("completed_steps", []) and provider.requires_key:
-            provider_id=provider.id,
+            from pyra.vault.reader import get_key as _get_key
-            model=model,
+            if not _get_key(provider.id):
-            base_url=provider.base_url,
+                _collect_api_key(provider)
-        ),
+            _mark_step_done(state, "api_key")
-        general=GeneralConfig(user_name=user_name, purpose=purpose),
+            _save_draft(state)
-    )
-    save_config(cfg)
 
-    _suggest_plugins(use_cases)
+        # ── Step 5: connection test ────────────────────────────────────────
+        if "connection" not in state.get("completed_steps", []):
+            model = _test_connection(provider, model)
+            state["model"] = model
+            _mark_step_done(state, "connection")
+            _save_draft(state)
 
-    console.print()
+        # ── Finalise ───────────────────────────────────────────────────────
-    console.print(Panel(
+        cfg = PyraConfig(
-        f"[green]Setup complete![/green]\n\n"
+            ai=ProviderConfig(
-        f"Provider: [bold]{provider.display_name}[/bold]\n"
+                provider_id=provider.id,
-        f"Model:    [bold]{model}[/bold]\n\n"
+                model=model,
-        "Run [bold cyan]pyra chat[/bold cyan] to start talking.",
+                base_url=provider.base_url,
-        border_style="green",
+            ),
-    ))
+            general=GeneralConfig(user_name=user_name, purpose=purpose),
+        )
+        save_config(cfg)
+        _delete_draft()
+
+        _suggest_plugins(use_cases)
+
+        console.print()
+        console.print(Panel(
+            f"[green]Setup complete![/green]\n\n"
+            f"Provider: [bold]{provider.display_name}[/bold]\n"
+            f"Model:    [bold]{model}[/bold]\n\n"
+            "Run [bold cyan]pyra chat[/bold cyan] to start talking.",
+            border_style="green",
+        ))
+
+    except SystemExit:
+        if state.get("completed_steps"):
+            console.print()
+            console.print(
+                "  [dim]Setup paused — run [bold]pyra setup[/bold] to resume.[/dim]"
+            )
+        raise
 
 
 def _collect_user_profile() -> tuple[str, str, list[str]]:
@@ -135,29 +261,254 @@ def _choose_provider() -> Provider:
 
     if provider.connectivity_check:
         _check_local_server(provider)
+        _show_local_model_status(provider)
 
     return provider
 
 
+def _classify_error(exc: Exception) -> tuple[str, str]:
+    """Return (short_label, resolution_hint) for a provider or network error."""
+    name = type(exc).__name__
+    module = type(exc).__module__ or ""
+    is_llm = "litellm" in module or "openai" in module
+
+    if is_llm:
+        if "AuthenticationError" in name:
+            return (
+                "Invalid API key",
+                "The provider rejected your API key.\n"
+                "Double-check it on the provider dashboard and re-enter it.",
+            )
+        if "NotFoundError" in name:
+            return (
+                "Model not found",
+                "The model name doesn't exist for this provider.\n"
+                "Check the exact model identifier on the provider's model list.",
+            )
+        if "RateLimitError" in name:
+            return (
+                "Rate limit reached",
+                "You've exceeded the provider's request rate.\n"
+                "Wait a few seconds and retry.",
+            )
+        if "ServiceUnavailable" in name:
+            return (
+                "Service temporarily unavailable",
+                "The provider's servers returned a 5xx error.\n"
+                "This is usually transient — wait a minute and retry.",
+            )
+        if "APIConnectionError" in name or "ConnectError" in name:
+            return (
+                "Cannot reach provider",
+                "A network error prevented the connection.\n"
+                "Check your internet connection and firewall settings.",
+            )
+        if "Timeout" in name:
+            return (
+                "Request timed out",
+                "The provider did not respond in time.\n"
+                "The service may be overloaded — retry in a moment.",
+            )
+        if "BadRequestError" in name or "InvalidRequest" in name:
+            return (
+                "Bad request",
+                "The request was rejected — the model name or parameters may be wrong.\n"
+                "Verify the exact model identifier.",
+            )
+        return ("Provider error", str(exc)[:300])
+
+    if "httpx" in module:
+        if "ConnectError" in name or "ConnectTimeout" in name:
+            return (
+                "Server not reachable",
+                "Could not connect to the local server.\n"
+                "Make sure it is running and listening on the expected address.",
+            )
+        if "Timeout" in name:
+            return (
+                "Connection timed out",
+                "The local server did not respond in time.\n"
+                "It may still be starting up — wait a moment and retry.",
+            )
+        if "HTTPStatusError" in name:
+            code = getattr(getattr(exc, "response", None), "status_code", 0)
+            if code == 401:
+                return ("Unauthorized (401)", "The server requires credentials that were not provided.")
+            if code == 404:
+                return ("Endpoint not found (404)", "The API endpoint was not found — check the server version.")
+            return (f"HTTP {code}", f"The server returned an unexpected error (HTTP {code}).")
+
+    return ("Unexpected error", str(exc)[:300])
+
+
 def _check_local_server(provider: Provider) -> None:
-    console.print(f"  Checking connection to [bold]{provider.display_name}[/bold]...", end=" ")
+    while True:
-    try:
-        resp = httpx.get(provider.connectivity_check, timeout=3.0)
-        resp.raise_for_status()
-        console.print("[green]✓[/green]")
-    except Exception:
-        console.print("[yellow]✗ (server not reachable)[/yellow]")
         console.print(
-            f"  [yellow]Warning:[/yellow] Could not reach {provider.base_url}.\n"
+            f"  Checking connection to [bold]{provider.display_name}[/bold]...", end=" "
-            f"  Make sure {provider.display_name} is running before using Pyra."
         )
+        try:
+            resp = httpx.get(provider.connectivity_check, timeout=3.0)
+            resp.raise_for_status()
+            console.print("[green]✓[/green]")
+            return
+        except Exception as exc:
+            label, hint = _classify_error(exc)
+            console.print("[yellow]✗[/yellow]")
+            console.print()
+            console.print(Panel(
+                f"[bold yellow]{label}[/bold yellow]\n\n{hint}",
+                title="Connection problem",
+                border_style="yellow",
+            ))
+            action = questionary.select(
+                "How would you like to proceed?",
+                choices=[
+                    questionary.Choice("Retry", value="retry"),
+                    questionary.Choice(
+                        "Continue anyway (model list may be unavailable)", value="continue"
+                    ),
+                    questionary.Choice("Abort setup", value="abort"),
+                ],
+            ).ask()
+            if action is None or action == "abort":
+                raise SystemExit(0)
+            if action == "continue":
+                return
+            # "retry" → loop
+
+
+def fetch_loaded_models(provider: Provider) -> list[str]:
+    """Return models currently loaded in RAM from a local provider's API."""
+    if not provider.base_url:
+        return []
+    try:
+        if provider.id == "ollama":
+            resp = httpx.get(f"{provider.base_url}/api/ps", timeout=3.0)
+            resp.raise_for_status()
+            return [m["name"] for m in resp.json().get("models", [])]
+        elif provider.id == "lmstudio":
+            resp = httpx.get("http://localhost:1234/api/v0/models", timeout=3.0)
+            resp.raise_for_status()
+            return [
+                m["id"] for m in resp.json().get("data", [])
+                if m.get("state") == "loaded"
+            ]
+        else:  # llamacpp — /models returns only the active loaded model
+            resp = httpx.get(f"{provider.base_url}/models", timeout=3.0)
+            resp.raise_for_status()
+            return [m["id"] for m in resp.json().get("data", [])]
+    except Exception:
+        return []
+
+
+def _show_local_model_status(provider: Provider) -> None:
+    """Print a one-line status showing which models are currently loaded."""
+    models = fetch_loaded_models(provider)
+    if not models:
+        console.print("  [yellow]No model currently loaded[/yellow]")
+    elif len(models) == 1:
+        console.print(f"  [green]Loaded model:[/green] {models[0]}")
+    else:
+        names = ", ".join(models)
+        console.print(f"  [green]{len(models)} models loaded:[/green] {names}")
+
+
+def _fetch_local_models(provider: Provider) -> list[str]:
+    """Return currently loaded models from a local provider's API."""
+    if not provider.base_url:
+        return []
+    try:
+        if provider.id == "ollama":
+            resp = httpx.get(f"{provider.base_url}/api/tags", timeout=3.0)
+            resp.raise_for_status()
+            return [m["name"] for m in resp.json().get("models", [])]
+        elif provider.id == "lmstudio":
+            resp = httpx.get("http://localhost:1234/api/v0/models", timeout=3.0)
+            resp.raise_for_status()
+            return [
+                m["id"] for m in resp.json().get("data", [])
+                if m.get("state") == "loaded"
+            ]
+        else:
+            resp = httpx.get(f"{provider.base_url}/models", timeout=3.0)
+            resp.raise_for_status()
+            return [m["id"] for m in resp.json().get("data", [])]
+    except Exception:
+        return []
+
+
+def _fetch_lmstudio_available_models() -> list[str]:
+    """Return all downloaded (not necessarily loaded) models from LM Studio's beta API."""
+    try:
+        resp = httpx.get("http://localhost:1234/api/v0/models", timeout=3.0)
+        resp.raise_for_status()
+        return [m["id"] for m in resp.json().get("data", [])]
+    except Exception:
+        return []
+
+
+def _load_lmstudio_model(model_id: str) -> bool:
+    """Attempt to load a model via LM Studio's beta API. Returns True on success."""
+    try:
+        resp = httpx.post(
+            "http://localhost:1234/api/v0/models/load",
+            json={"identifier": model_id},
+            timeout=60.0,
+        )
+        return resp.is_success
+    except Exception:
+        return False
 
 
 def _choose_model(provider: Provider) -> str:
-    model = questionary.text(
+    if provider.group != "Local":
-        "Model name:",
+        model = questionary.text("Model name:", default=provider.default_model).ask()
-        default=provider.default_model,
+        if model is None:
-    ).ask()
+            raise SystemExit(0)
+        return model.strip()
+
+    _MANUAL = "__manual__"
+    loaded = _fetch_local_models(provider)
+
+    if loaded:
+        choices = loaded + [questionary.Choice("── Enter manually ──", value=_MANUAL)]
+        selected = questionary.select("Select model:", choices=choices).ask()
+        if selected is None:
+            raise SystemExit(0)
+        if selected != _MANUAL:
+            return selected
+
+    elif provider.id == "lmstudio":
+        console.print("  [yellow]No model currently loaded in LM Studio.[/yellow]")
+        available = _fetch_lmstudio_available_models()
+        if available:
+            choices = available + [questionary.Choice("── Enter manually ──", value=_MANUAL)]
+            selected = questionary.select(
+                "Select a downloaded model to load:", choices=choices
+            ).ask()
+            if selected is None:
+                raise SystemExit(0)
+            if selected != _MANUAL:
+                console.print(f"  Loading [bold]{selected}[/bold]...", end=" ")
+                if _load_lmstudio_model(selected):
+                    console.print("[green]✓ Loaded[/green]")
+                else:
+                    console.print(
+                        "[yellow]Could not load via API — "
+                        "please load the model manually in LM Studio.[/yellow]"
+                    )
+                return selected
+        else:
+            console.print(Panel(
+                "No models are loaded or downloaded in LM Studio.\n"
+                "Open LM Studio → Local Server tab → load a model, then re-run setup.",
+                border_style="yellow",
+            ))
+
+    else:
+        console.print(f"  [yellow]No models found at {provider.base_url}.[/yellow]")
+
+    model = questionary.text("Model name:", default=provider.default_model).ask()
     if model is None:
         raise SystemExit(0)
     return model.strip()
@@ -180,26 +531,68 @@ def _collect_api_key(provider: Provider) -> None:
     console.print("  [green]✓ Key stored in vault[/green]")
 
 
-def _test_connection(provider: Provider, model: str) -> None:
+def _test_connection(provider: Provider, model: str) -> str:
     from pyra.vault.reader import get_key
 
-    console.print("\n  Running test call...", end=" ")
+    while True:
-    try:
+        console.print("\n  Running connection test...", end=" ")
-        import litellm
+        try:
+            import litellm
 
-        # Local providers don't need a real key but litellm still requires the field
+            api_key = get_key(provider.id) if provider.requires_key else "local"
-        api_key = get_key(provider.id) if provider.requires_key else "local"
+            kwargs: dict = {
-        kwargs: dict = {
+                "model": f"{provider.litellm_prefix}{model}",
-            "model": f"{provider.litellm_prefix}{model}",
+                "messages": [{"role": "user", "content": "Reply with exactly: OK"}],
-            "messages": [{"role": "user", "content": "Reply with exactly: OK"}],
+                "max_tokens": 10,
-            "max_tokens": 10,
+                "api_key": api_key,
-            "api_key": api_key,
+            }
-        }
+            if provider.base_url:
-        if provider.base_url:
+                kwargs["api_base"] = provider.base_url
-            kwargs["api_base"] = provider.base_url
 
-        litellm.completion(**kwargs)
+            litellm.completion(**kwargs)
-        console.print("[green]✓ Connection OK[/green]")
+            console.print("[green]✓ Connection OK[/green]")
-    except Exception as exc:
+            return model
-        console.print(f"[yellow]✗ Test call failed: {exc}[/yellow]")
+
-        console.print("  [dim]You can still proceed — check your config with 'pyra setup' again.[/dim]")
+        except Exception as exc:
+            label, hint = _classify_error(exc)
+            console.print("[red]✗[/red]")
+            console.print()
+            console.print(Panel(
+                f"[bold red]{label}[/bold red]\n\n{hint}",
+                title="Test call failed",
+                border_style="red",
+            ))
+
+            exc_name = type(exc).__name__
+            is_auth_error = "AuthenticationError" in exc_name
+            is_model_error = any(
+                kw in exc_name for kw in ("NotFoundError", "BadRequestError", "InvalidRequest")
+            )
+
+            choices = [questionary.Choice("Retry", value="retry")]
+            if is_model_error:
+                choices.append(questionary.Choice("Change model", value="change_model"))
+            if provider.requires_key and is_auth_error:
+                choices.append(questionary.Choice("Re-enter API key", value="rekey"))
+            choices += [
+                questionary.Choice("Skip test and continue setup", value="skip"),
+                questionary.Choice("Abort setup", value="abort"),
+            ]
+
+            action = questionary.select(
+                "How would you like to proceed?",
+                choices=choices,
+            ).ask()
+
+            if action is None or action == "abort":
+                raise SystemExit(0)
+            if action == "skip":
+                console.print(
+                    "  [dim]Test skipped — run [bold]pyra setup[/bold] again if chat doesn't work.[/dim]"
+                )
+                return model
+            if action == "change_model":
+                model = _choose_model(provider)
+            elif action == "rekey":
+                _collect_api_key(provider)
+            # loop → retry (with possibly new model or key)

tests/integration/test_lmstudio.py

@@ -1,31 +1,41 @@
 """
 Live integration test against LM Studio at localhost:1234.
-Skipped automatically if LM Studio is not running.
+Skipped automatically if LM Studio is not running or no model is loaded.
 """
+import httpx
 import pytest
 
-LMSTUDIO_MODEL = "gemma-4-e4b-uncensored-hauhaucs-aggressive"
+_LMSTUDIO_BASE_URL = "http://localhost:1234/v1"
-LMSTUDIO_BASE_URL = "http://localhost:1234/v1"
 
 
-@pytest.fixture(autouse=True)
+def _get_loaded_model() -> str | None:
-def require_lmstudio():
+    """Return the first currently loaded model ID from LM Studio, or None."""
-    import httpx
     try:
-        r = httpx.get(f"{LMSTUDIO_BASE_URL}/models", timeout=2.0)
+        resp = httpx.get(f"{_LMSTUDIO_BASE_URL}/models", timeout=2.0)
-        r.raise_for_status()
+        resp.raise_for_status()
+        models = resp.json().get("data", [])
+        return models[0]["id"] if models else None
     except Exception:
-        pytest.skip("LM Studio not reachable at localhost:1234")
+        return None
 
 
-def test_basic_completion():
+@pytest.fixture()
+def lmstudio_model() -> str:
+    """Resolve the first loaded model in LM Studio; skip if none available."""
+    model = _get_loaded_model()
+    if model is None:
+        pytest.skip("LM Studio not reachable or no model currently loaded")
+    return model
+
+
+def test_basic_completion(lmstudio_model):
     import litellm
     litellm.suppress_debug_info = True
 
     response = litellm.completion(
-        model=f"openai/{LMSTUDIO_MODEL}",
+        model=f"openai/{lmstudio_model}",
         messages=[{"role": "user", "content": "Reply with exactly the word: PONG"}],
-        api_base=LMSTUDIO_BASE_URL,
+        api_base=_LMSTUDIO_BASE_URL,
         api_key="lm-studio",
         max_tokens=20,
         stream=False,
@@ -34,14 +44,14 @@ def test_basic_completion():
     assert text and len(text) > 0
 
 
-def test_streaming_completion():
+def test_streaming_completion(lmstudio_model):
     import litellm
     litellm.suppress_debug_info = True
 
     stream = litellm.completion(
-        model=f"openai/{LMSTUDIO_MODEL}",
+        model=f"openai/{lmstudio_model}",
         messages=[{"role": "user", "content": "Count from 1 to 3."}],
-        api_base=LMSTUDIO_BASE_URL,
+        api_base=_LMSTUDIO_BASE_URL,
         api_key="lm-studio",
         max_tokens=50,
         stream=True,
@@ -52,30 +62,29 @@ def test_streaming_completion():
     assert len(full_text) > 0
 
 
-def test_injection_scan_on_live_response(tmp_pyra_home):
+def test_injection_scan_on_live_response(tmp_pyra_home, lmstudio_model):
     """Verify injection scanner runs on real model output without false positives."""
     import litellm
     from pyra.security.injection import scan_response
     litellm.suppress_debug_info = True
 
     response = litellm.completion(
-        model=f"openai/{LMSTUDIO_MODEL}",
+        model=f"openai/{lmstudio_model}",
         messages=[{"role": "user", "content": "Explain what a list is in Python."}],
-        api_base=LMSTUDIO_BASE_URL,
+        api_base=_LMSTUDIO_BASE_URL,
         api_key="lm-studio",
         max_tokens=200,
         stream=False,
     )
     text = response.choices[0].message.content
     warnings = scan_response(text)
-    # Normal responses about Python lists should not trigger injection warnings
-    for w in warnings:
-        print(f"[warning] {w.pattern_label}: {w.matched_text!r}")
     # Not asserting zero warnings — some models may have quirky phrasing —
     # but at least the scanner must not crash on real output
+    for w in warnings:
+        print(f"[warning] {w.pattern_label}: {w.matched_text!r}")
 
 
-def test_pyra_chat_session_with_lmstudio(tmp_pyra_home):
+def test_pyra_chat_session_with_lmstudio(tmp_pyra_home, lmstudio_model):
     """Full stack: config → vault → history → litellm → injection scan."""
     from pyra.config.schema import PyraConfig, ProviderConfig
     from pyra.config.manager import save_config
@@ -87,8 +96,8 @@ def test_pyra_chat_session_with_lmstudio(tmp_pyra_home):
     cfg = PyraConfig(
         ai=ProviderConfig(
             provider_id="lmstudio",
-            model=LMSTUDIO_MODEL,
+            model=lmstudio_model,
-            base_url=LMSTUDIO_BASE_URL,
+            base_url=_LMSTUDIO_BASE_URL,
         )
     )
     save_config(cfg)
@@ -98,9 +107,9 @@ def test_pyra_chat_session_with_lmstudio(tmp_pyra_home):
     messages = history.build_for_api()
 
     response = litellm.completion(
-        model=f"openai/{LMSTUDIO_MODEL}",
+        model=f"openai/{lmstudio_model}",
         messages=messages,
-        api_base=LMSTUDIO_BASE_URL,
+        api_base=_LMSTUDIO_BASE_URL,
         api_key="lm-studio",
         max_tokens=30,
         stream=False,

tests/unit/test_daemon_core.py

@@ -0,0 +1,226 @@
+"""Unit tests for the daemon core — PluginSupervisor and IPC handler dispatch."""
+
+from __future__ import annotations
+
+import asyncio
+
+import pytest
+
+from pyra.daemon.core import PluginSupervisor, _make_ipc_handler
+
+
+# ── Helpers ───────────────────────────────────────────────────────────────────
+
+async def _drain(n: int = 20) -> None:
+    """Yield to the event loop n times to let scheduled tasks run."""
+    for _ in range(n):
+        await asyncio.sleep(0)
+
+
+# ── PluginSupervisor — lifecycle ──────────────────────────────────────────────
+
+async def test_supervisor_empty_starts_and_stops_cleanly() -> None:
+    sup = PluginSupervisor()
+    await sup.start()
+    await sup.stop()
+    assert sup.status() == []
+
+
+async def test_supervisor_runs_task_to_completion() -> None:
+    done = asyncio.Event()
+
+    async def task():
+        done.set()
+
+    sup = PluginSupervisor()
+    sup._RESTART_DELAY = 0.0
+    sup.add_task("t", task)
+    await sup.start()
+
+    await asyncio.wait_for(done.wait(), timeout=1.0)
+    await sup.stop()
+
+    assert sup._records[0].restart_count == 0
+    assert sup._records[0].last_error is None
+
+
+async def test_supervisor_restarts_crashed_task() -> None:
+    call_count = 0
+    completed = asyncio.Event()
+
+    async def flaky():
+        nonlocal call_count
+        call_count += 1
+        if call_count == 1:
+            raise RuntimeError("first call fails")
+        completed.set()
+
+    sup = PluginSupervisor()
+    sup._RESTART_DELAY = 0.0
+    sup.add_task("flaky", flaky)
+    await sup.start()
+
+    await asyncio.wait_for(completed.wait(), timeout=1.0)
+    await sup.stop()
+
+    assert sup._records[0].restart_count == 1
+    assert "RuntimeError" in (sup._records[0].last_error or "")
+
+
+async def test_supervisor_gives_up_after_max_restarts() -> None:
+    async def always_fails():
+        raise ValueError("always")
+
+    sup = PluginSupervisor()
+    sup._RESTART_DELAY = 0.0
+    sup._MAX_RESTARTS = 3
+    sup.add_task("failing", always_fails)
+    await sup.start()
+
+    # Allow enough iterations for 3 restarts + give-up.
+    for _ in range(200):
+        await asyncio.sleep(0)
+        if sup._records[0].task and sup._records[0].task.done():
+            break
+
+    await sup.stop()
+
+    assert sup._records[0].restart_count == 3
+    assert sup._records[0].last_error is not None
+
+
+# ── PluginSupervisor — status ─────────────────────────────────────────────────
+
+async def test_supervisor_status_returns_correct_shape() -> None:
+    sup = PluginSupervisor()
+    sup._RESTART_DELAY = 0.0
+
+    async def noop():
+        pass
+
+    sup.add_task("noop", noop)
+    await sup.start()
+    await _drain()
+
+    statuses = sup.status()
+    assert len(statuses) == 1
+    s = statuses[0]
+    assert set(s.keys()) == {"name", "alive", "restart_count", "last_error"}
+    assert s["name"] == "noop"
+    assert isinstance(s["alive"], bool)
+    assert isinstance(s["restart_count"], int)
+
+    await sup.stop()
+
+
+async def test_supervisor_status_empty_when_no_tasks() -> None:
+    sup = PluginSupervisor()
+    await sup.start()
+    assert sup.status() == []
+    await sup.stop()
+
+
+# ── PluginSupervisor — reload ─────────────────────────────────────────────────
+
+async def test_supervisor_reload_restarts_tasks() -> None:
+    call_count = 0
+
+    async def counting():
+        nonlocal call_count
+        call_count += 1
+        # Hang until cancelled so reload can cancel it.
+        await asyncio.sleep(10)
+
+    sup = PluginSupervisor()
+    sup._RESTART_DELAY = 0.0
+    sup.add_task("c", counting)
+    await sup.start()
+
+    await _drain()
+    assert call_count == 1
+
+    await sup.reload()
+    await _drain()
+
+    # After reload, the task should have been restarted (called a second time).
+    assert call_count == 2
+    assert sup._records[0].restart_count == 0  # reset by reload
+
+    await sup.stop()
+
+
+async def test_supervisor_reload_resets_restart_count() -> None:
+    call_count = 0
+
+    async def flaky():
+        nonlocal call_count
+        call_count += 1
+        if call_count <= 2:
+            raise RuntimeError("crash")
+        await asyncio.sleep(10)
+
+    sup = PluginSupervisor()
+    sup._RESTART_DELAY = 0.0
+    sup.add_task("f", flaky)
+    await sup.start()
+
+    # Wait for 2 crashes to accumulate.
+    for _ in range(200):
+        await asyncio.sleep(0)
+        if sup._records[0].restart_count >= 2:
+            break
+
+    assert sup._records[0].restart_count == 2
+
+    await sup.reload()
+    # Reload must reset the counter.
+    assert sup._records[0].restart_count == 0
+
+    await sup.stop()
+
+
+# ── IPC command handler ───────────────────────────────────────────────────────
+
+async def test_ipc_handler_ping() -> None:
+    sup = PluginSupervisor()
+    handler = _make_ipc_handler(sup)
+    resp = await handler({"cmd": "ping"})
+    assert resp["ok"] is True
+    assert resp["data"]["pong"] is True
+
+
+async def test_ipc_handler_status_shape() -> None:
+    sup = PluginSupervisor()
+    handler = _make_ipc_handler(sup)
+    resp = await handler({"cmd": "status"})
+    assert resp["ok"] is True
+    assert "uptime" in resp["data"]
+    assert "pid" in resp["data"]
+    assert "tasks" in resp["data"]
+    assert isinstance(resp["data"]["tasks"], list)
+
+
+async def test_ipc_handler_stop_signals_shutdown() -> None:
+    sup = PluginSupervisor()
+    handler = _make_ipc_handler(sup)
+    assert not sup._shutdown.is_set()
+    resp = await handler({"cmd": "stop"})
+    assert resp["ok"] is True
+    assert sup._shutdown.is_set()
+
+
+async def test_ipc_handler_reload_returns_task_count() -> None:
+    sup = PluginSupervisor()
+    handler = _make_ipc_handler(sup)
+    resp = await handler({"cmd": "reload"})
+    assert resp["ok"] is True
+    assert resp["data"]["tasks_reloaded"] == 0
+
+
+async def test_ipc_handler_unknown_command() -> None:
+    sup = PluginSupervisor()
+    handler = _make_ipc_handler(sup)
+    resp = await handler({"cmd": "bogus"})
+    assert resp["ok"] is False
+    assert "error" in resp["data"]
+    assert "bogus" in resp["data"]["error"]

tests/unit/test_daemon_ipc.py

@@ -0,0 +1,162 @@
+"""Unit tests for the IPC layer."""
+
+from __future__ import annotations
+
+import asyncio
+import os
+import sys
+import tempfile
+from pathlib import Path
+
+import pytest
+
+from pyra.daemon.ipc import (
+    IpcClient,
+    IpcMessage,
+    IpcResponse,
+    IpcServer,
+    decode_message,
+    encode_message,
+    is_unix_socket,
+)
+
+
+@pytest.fixture
+def sock_path():
+    """Short socket path that fits within macOS's 104-char AF_UNIX limit."""
+    with tempfile.TemporaryDirectory(dir="/tmp") as d:
+        yield Path(d) / "t.sock"
+
+
+# ── Protocol encode / decode ──────────────────────────────────────────────────
+
+def test_encode_appends_newline() -> None:
+    data = encode_message({"cmd": "ping"})
+    assert data.endswith(b"\n")
+
+
+def test_encode_is_valid_json() -> None:
+    import json
+    data = encode_message({"cmd": "status", "extra": 42})
+    assert json.loads(data) == {"cmd": "status", "extra": 42}
+
+
+def test_decode_roundtrip() -> None:
+    msg: IpcMessage = {"cmd": "stop"}
+    assert decode_message(encode_message(msg)) == msg
+
+
+def test_decode_strips_newline() -> None:
+    assert decode_message(b'{"cmd": "stop"}\n')["cmd"] == "stop"
+
+
+def test_decode_raises_on_bad_json() -> None:
+    with pytest.raises(ValueError, match="Invalid IPC message"):
+        decode_message(b"not json\n")
+
+
+def test_decode_raises_on_empty_line() -> None:
+    with pytest.raises(ValueError):
+        decode_message(b"\n")
+
+
+# ── is_unix_socket ────────────────────────────────────────────────────────────
+
+def test_is_unix_socket_matches_platform() -> None:
+    if sys.platform == "win32":
+        assert not is_unix_socket()
+    else:
+        assert is_unix_socket()
+
+
+# ── Server + client roundtrip (Unix only) ─────────────────────────────────────
+
+@pytest.mark.skipif(sys.platform == "win32", reason="Unix socket test")
+async def test_server_client_ping(sock_path: Path) -> None:
+    async def handler(msg: IpcMessage) -> IpcResponse:
+        return {"ok": True, "data": {"pong": True}}
+
+    server = IpcServer(sock_path, handler)
+    await server.start()
+    try:
+        resp = await IpcClient(sock_path).send({"cmd": "ping"})
+        assert resp["ok"] is True
+        assert resp["data"]["pong"] is True
+    finally:
+        await server.stop()
+
+
+@pytest.mark.skipif(sys.platform == "win32", reason="Unix socket test")
+async def test_server_echoes_error_for_bad_json(sock_path: Path) -> None:
+    async def handler(msg: IpcMessage) -> IpcResponse:
+        return {"ok": True, "data": {}}
+
+    server = IpcServer(sock_path, handler)
+    await server.start()
+    try:
+        reader, writer = await asyncio.open_unix_connection(str(sock_path))
+        writer.write(b"not valid json\n")
+        await writer.drain()
+        line = await asyncio.wait_for(reader.readline(), timeout=3.0)
+        resp = decode_message(line)
+        assert resp["ok"] is False
+        assert "error" in resp["data"]
+    finally:
+        try:
+            writer.close()
+        except Exception:
+            pass
+        await server.stop()
+
+
+@pytest.mark.skipif(sys.platform == "win32", reason="Unix socket test")
+async def test_handler_response_returned_to_client(sock_path: Path) -> None:
+    async def handler(msg: IpcMessage) -> IpcResponse:
+        if msg.get("cmd") == "status":
+            return {"ok": True, "data": {"uptime": 99.0}}
+        return {"ok": False, "data": {"error": "unknown"}}
+
+    server = IpcServer(sock_path, handler)
+    await server.start()
+    try:
+        resp = await IpcClient(sock_path).send({"cmd": "status"})
+        assert resp["ok"] is True
+        assert resp["data"]["uptime"] == 99.0
+
+        resp2 = await IpcClient(sock_path).send({"cmd": "bogus"})
+        assert resp2["ok"] is False
+    finally:
+        await server.stop()
+
+
+@pytest.mark.skipif(sys.platform == "win32", reason="Unix socket test")
+async def test_client_raises_when_no_server(sock_path: Path) -> None:
+    client = IpcClient(sock_path)
+    with pytest.raises((ConnectionRefusedError, FileNotFoundError, OSError)):
+        await client.send({"cmd": "ping"})
+
+
+@pytest.mark.skipif(sys.platform == "win32", reason="Unix socket test")
+async def test_socket_file_chmod_600(sock_path: Path) -> None:
+    async def handler(msg: IpcMessage) -> IpcResponse:
+        return {"ok": True, "data": {}}
+
+    server = IpcServer(sock_path, handler)
+    await server.start()
+    try:
+        mode = oct(sock_path.stat().st_mode & 0o777)
+        assert mode == oct(0o600), f"Expected 0o600, got {mode}"
+    finally:
+        await server.stop()
+
+
+@pytest.mark.skipif(sys.platform == "win32", reason="Unix socket test")
+async def test_stop_removes_socket_file(sock_path: Path) -> None:
+    async def handler(msg: IpcMessage) -> IpcResponse:
+        return {"ok": True, "data": {}}
+
+    server = IpcServer(sock_path, handler)
+    await server.start()
+    assert sock_path.exists()
+    await server.stop()
+    assert not sock_path.exists()

tests/unit/test_daemon_pid.py

@@ -0,0 +1,103 @@
+"""Unit tests for daemon PID file management."""
+
+from __future__ import annotations
+
+import os
+from pathlib import Path
+
+import pytest
+
+from pyra.daemon.pid import PidFile, PidFileError, resolve_pid_path
+
+
+def test_write_creates_file(tmp_path: Path) -> None:
+    p = PidFile(tmp_path / "daemon.pid")
+    p.write()
+    assert (tmp_path / "daemon.pid").exists()
+    assert int((tmp_path / "daemon.pid").read_text().strip()) == os.getpid()
+
+
+def test_read_returns_none_when_absent(tmp_path: Path) -> None:
+    p = PidFile(tmp_path / "daemon.pid")
+    assert p.read() is None
+
+
+def test_read_returns_pid_when_present(tmp_path: Path) -> None:
+    pid_file = tmp_path / "daemon.pid"
+    pid_file.write_text("12345")
+    p = PidFile(pid_file)
+    assert p.read() == 12345
+
+
+def test_read_returns_none_on_bad_content(tmp_path: Path) -> None:
+    pid_file = tmp_path / "daemon.pid"
+    pid_file.write_text("not-a-number")
+    p = PidFile(pid_file)
+    assert p.read() is None
+
+
+def test_is_stale_false_for_self(tmp_path: Path) -> None:
+    p = PidFile(tmp_path / "daemon.pid")
+    p.write()
+    assert not p.is_stale()
+
+
+def test_is_stale_true_for_dead_pid(tmp_path: Path) -> None:
+    pid_file = tmp_path / "daemon.pid"
+    pid_file.write_text("999999999")  # unrealistically large PID
+    p = PidFile(pid_file)
+    assert p.is_stale()
+
+
+def test_is_stale_false_when_file_absent(tmp_path: Path) -> None:
+    p = PidFile(tmp_path / "daemon.pid")
+    assert not p.is_stale()
+
+
+def test_remove_deletes_file(tmp_path: Path) -> None:
+    p = PidFile(tmp_path / "daemon.pid")
+    p.write()
+    p.remove()
+    assert not (tmp_path / "daemon.pid").exists()
+
+
+def test_remove_is_idempotent(tmp_path: Path) -> None:
+    p = PidFile(tmp_path / "daemon.pid")
+    p.remove()  # must not raise
+
+
+def test_context_manager_writes_and_removes(tmp_path: Path) -> None:
+    pid_file = tmp_path / "daemon.pid"
+    p = PidFile(pid_file)
+    with p:
+        assert pid_file.exists()
+        assert int(pid_file.read_text().strip()) == os.getpid()
+    assert not pid_file.exists()
+
+
+def test_write_raises_when_live_pid_exists(tmp_path: Path) -> None:
+    p = PidFile(tmp_path / "daemon.pid")
+    p.write()  # writes self PID (which is alive)
+    p2 = PidFile(tmp_path / "daemon.pid")
+    with pytest.raises(PidFileError):
+        p2.write()
+
+
+def test_write_succeeds_over_stale_pid(tmp_path: Path) -> None:
+    pid_file = tmp_path / "daemon.pid"
+    pid_file.write_text("999999999")  # stale
+    p = PidFile(pid_file)
+    p.write()  # should not raise
+    assert int(pid_file.read_text().strip()) == os.getpid()
+
+
+def test_resolve_pid_path_expands_tilde() -> None:
+    result = resolve_pid_path("~/.pyra/daemon.pid")
+    assert not str(result).startswith("~")
+    assert result.is_absolute()
+
+
+def test_resolve_pid_path_absolute_unchanged(tmp_path: Path) -> None:
+    path = tmp_path / "daemon.pid"
+    result = resolve_pid_path(str(path))
+    assert result == path

tests/unit/test_daemon_service.py

@@ -0,0 +1,189 @@
+"""Unit tests for daemon service file generation and platform detection."""
+
+from __future__ import annotations
+
+import subprocess
+import sys
+from pathlib import Path
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from pyra.daemon.service import (
+    detect_platform,
+    find_pyra_executable,
+    render_launchd_plist,
+    render_systemd_unit,
+    render_schtasks_xml,
+)
+
+
+# ── Template rendering ────────────────────────────────────────────────────────
+
+def test_render_launchd_plist_contains_exe() -> None:
+    xml = render_launchd_plist("/usr/local/bin/pyra", "~/.pyra/daemon.log", "~/.pyra/daemon.pid")
+    assert "/usr/local/bin/pyra" in xml
+    assert "<string>daemon</string>" in xml
+    assert "<string>run</string>" in xml
+    assert "com.pyra.daemon" in xml
+    assert "<true/>" in xml  # KeepAlive and RunAtLoad
+
+
+def test_render_launchd_plist_expands_log_tilde() -> None:
+    xml = render_launchd_plist("/bin/pyra", "~/.pyra/daemon.log", "~/.pyra/daemon.pid")
+    assert "~" not in xml
+
+
+def test_render_systemd_unit_contains_exe() -> None:
+    unit = render_systemd_unit("/usr/local/bin/pyra", "~/.pyra/daemon.log")
+    assert "ExecStart=/usr/local/bin/pyra daemon run" in unit
+    assert "Restart=on-failure" in unit
+    assert "Type=simple" in unit
+    assert "WantedBy=default.target" in unit
+
+
+def test_render_systemd_unit_expands_log_tilde() -> None:
+    unit = render_systemd_unit("/bin/pyra", "~/.pyra/daemon.log")
+    assert "~" not in unit
+
+
+def test_render_schtasks_xml_contains_exe() -> None:
+    xml = render_schtasks_xml("C:\\Users\\test\\pyra.exe")
+    assert "C:\\Users\\test\\pyra.exe" in xml
+    assert "LogonTrigger" in xml
+    assert "daemon run" in xml
+    assert "IgnoreNew" in xml
+
+
+def test_render_schtasks_xml_no_time_limit() -> None:
+    xml = render_schtasks_xml("pyra.exe")
+    assert "PT0S" in xml  # ExecutionTimeLimit=PT0S means unlimited
+
+
+# ── Platform detection ────────────────────────────────────────────────────────
+
+def test_detect_platform_returns_known_value() -> None:
+    result = detect_platform()
+    assert result in ("macos", "linux", "windows")
+
+
+@pytest.mark.parametrize("system,expected", [
+    ("Darwin", "macos"),
+    ("Linux", "linux"),
+    ("Windows", "windows"),
+])
+def test_detect_platform_maps_correctly(system: str, expected: str) -> None:
+    with patch("platform.system", return_value=system):
+        assert detect_platform() == expected
+
+
+def test_detect_platform_raises_on_unknown() -> None:
+    with patch("platform.system", return_value="FreeBSD"):
+        with pytest.raises(RuntimeError, match="Unsupported platform"):
+            detect_platform()
+
+
+# ── Executable detection ──────────────────────────────────────────────────────
+
+def test_find_pyra_executable_returns_string() -> None:
+    result = find_pyra_executable()
+    assert isinstance(result, str)
+    assert len(result) > 0
+
+
+def test_find_pyra_executable_uses_which_when_available(tmp_path: Path) -> None:
+    fake_pyra = tmp_path / "pyra"
+    fake_pyra.touch()
+    with patch("shutil.which", return_value=str(fake_pyra)):
+        assert find_pyra_executable() == str(fake_pyra)
+
+
+def test_find_pyra_executable_falls_back_to_sibling(tmp_path: Path) -> None:
+    fake_python = tmp_path / "python3"
+    fake_pyra = tmp_path / "pyra"
+    fake_pyra.touch()
+    with patch("shutil.which", return_value=None):
+        with patch("sys.executable", str(fake_python)):
+            assert find_pyra_executable() == str(fake_pyra)
+
+
+def test_find_pyra_executable_falls_back_to_module(tmp_path: Path) -> None:
+    fake_python = tmp_path / "python3"
+    with patch("shutil.which", return_value=None):
+        with patch("sys.executable", str(fake_python)):
+            result = find_pyra_executable()
+            assert result == f"{fake_python} -m pyra"
+
+
+# ── Install / uninstall (subprocess mocked) ───────────────────────────────────
+
+@pytest.mark.skipif(sys.platform == "win32", reason="launchd install is macOS-only")
+def test_install_launchd_writes_plist_and_calls_launchctl(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    import pyra.daemon.service as svc
+
+    plist_path = tmp_path / "Library" / "LaunchAgents" / "com.pyra.daemon.plist"
+    monkeypatch.setattr(svc, "_PLIST_PATH", plist_path)
+
+    calls: list[list[str]] = []
+    monkeypatch.setattr(subprocess, "run", lambda cmd, **kw: calls.append(cmd))
+
+    svc._install_launchd("/usr/local/bin/pyra", "~/.pyra/daemon.log", "~/.pyra/daemon.pid")
+
+    assert plist_path.exists()
+    assert "com.pyra.daemon" in plist_path.read_text()
+    assert any("launchctl" in c[0] for c in calls)
+
+
+@pytest.mark.skipif(sys.platform == "win32", reason="systemd install is Linux-only")
+def test_install_systemd_writes_unit_and_calls_systemctl(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    import pyra.daemon.service as svc
+
+    unit_path = tmp_path / ".config" / "systemd" / "user" / "pyra.service"
+    monkeypatch.setattr(svc, "_SYSTEMD_UNIT", unit_path)
+
+    calls: list[list[str]] = []
+    monkeypatch.setattr(subprocess, "run", lambda cmd, **kw: calls.append(cmd))
+
+    svc._install_systemd("/usr/local/bin/pyra", "~/.pyra/daemon.log")
+
+    assert unit_path.exists()
+    assert "ExecStart" in unit_path.read_text()
+    assert any("systemctl" in c[0] for c in calls)
+
+
+@pytest.mark.skipif(sys.platform == "win32", reason="launchd uninstall is macOS-only")
+def test_uninstall_launchd_removes_plist(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    import pyra.daemon.service as svc
+
+    plist_path = tmp_path / "Library" / "LaunchAgents" / "com.pyra.daemon.plist"
+    plist_path.parent.mkdir(parents=True)
+    plist_path.write_text("<plist/>")
+    monkeypatch.setattr(svc, "_PLIST_PATH", plist_path)
+    monkeypatch.setattr(subprocess, "run", lambda cmd, **kw: None)
+
+    svc._uninstall_launchd()
+
+    assert not plist_path.exists()
+
+
+@pytest.mark.skipif(sys.platform == "win32", reason="systemd uninstall is Linux-only")
+def test_uninstall_systemd_removes_unit(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    import pyra.daemon.service as svc
+
+    unit_path = tmp_path / ".config" / "systemd" / "user" / "pyra.service"
+    unit_path.parent.mkdir(parents=True)
+    unit_path.write_text("[Service]")
+    monkeypatch.setattr(svc, "_SYSTEMD_UNIT", unit_path)
+    monkeypatch.setattr(subprocess, "run", lambda cmd, **kw: None)
+
+    svc._uninstall_systemd()
+
+    assert not unit_path.exists()

tests/unit/test_setup_wizard.py

@@ -1,4 +1,6 @@
-"""Tests for setup wizard personalization helpers."""
+"""Tests for setup wizard personalization and model-discovery helpers."""
+
+from unittest.mock import MagicMock
 
 import pytest
 
@@ -90,3 +92,438 @@ def test_suggest_plugins_multiple_categories(monkeypatch):
     combined = " ".join(str(p) for p in panels)
     assert "email" in combined
     assert "ssh_tool" in combined
+
+
+# ── _fetch_local_models ────────────────────────────────────────────────────────
+
+def test_fetch_local_models_lmstudio_returns_loaded_model_ids(monkeypatch):
+    import pyra.setup.wizard as wiz
+    mock_resp = MagicMock()
+    mock_resp.json.return_value = {
+        "data": [
+            {"id": "gemma-4b", "state": "loaded"},
+            {"id": "llama3", "state": "not_loaded"},
+        ]
+    }
+    mock_resp.raise_for_status = lambda: None
+    monkeypatch.setattr(wiz.httpx, "get", lambda *a, **kw: mock_resp)
+    from pyra.setup.providers import get_provider
+    assert wiz._fetch_local_models(get_provider("lmstudio")) == ["gemma-4b"]
+
+
+def test_fetch_local_models_lmstudio_filters_unloaded(monkeypatch):
+    import pyra.setup.wizard as wiz
+    mock_resp = MagicMock()
+    mock_resp.json.return_value = {
+        "data": [
+            {"id": "model-a", "state": "not_loaded"},
+            {"id": "model-b", "state": "not_loaded"},
+        ]
+    }
+    mock_resp.raise_for_status = lambda: None
+    monkeypatch.setattr(wiz.httpx, "get", lambda *a, **kw: mock_resp)
+    from pyra.setup.providers import get_provider
+    assert wiz._fetch_local_models(get_provider("lmstudio")) == []
+
+
+def test_fetch_local_models_ollama_returns_model_names(monkeypatch):
+    import pyra.setup.wizard as wiz
+    mock_resp = MagicMock()
+    mock_resp.json.return_value = {"models": [{"name": "llama3:latest"}, {"name": "mistral"}]}
+    mock_resp.raise_for_status = lambda: None
+    monkeypatch.setattr(wiz.httpx, "get", lambda *a, **kw: mock_resp)
+    from pyra.setup.providers import get_provider
+    assert wiz._fetch_local_models(get_provider("ollama")) == ["llama3:latest", "mistral"]
+
+
+def test_fetch_local_models_returns_empty_on_connection_error(monkeypatch):
+    import pyra.setup.wizard as wiz
+    monkeypatch.setattr(wiz.httpx, "get", MagicMock(side_effect=Exception("conn refused")))
+    from pyra.setup.providers import get_provider
+    assert wiz._fetch_local_models(get_provider("lmstudio")) == []
+
+
+def test_fetch_local_models_returns_empty_when_no_base_url():
+    import pyra.setup.wizard as wiz
+    from pyra.setup.providers import Provider
+    provider = Provider(
+        id="test", display_name="Test", requires_key=False,
+        default_model="x", litellm_prefix="openai/", group="Local",
+    )
+    assert wiz._fetch_local_models(provider) == []
+
+
+# ── _fetch_lmstudio_available_models ──────────────────────────────────────────
+
+def test_fetch_lmstudio_available_models_returns_ids(monkeypatch):
+    import pyra.setup.wizard as wiz
+    mock_resp = MagicMock()
+    mock_resp.json.return_value = {"data": [{"id": "model-a"}, {"id": "model-b"}]}
+    mock_resp.raise_for_status = lambda: None
+    monkeypatch.setattr(wiz.httpx, "get", lambda *a, **kw: mock_resp)
+    assert wiz._fetch_lmstudio_available_models() == ["model-a", "model-b"]
+
+
+def test_fetch_lmstudio_available_models_returns_empty_on_error(monkeypatch):
+    import pyra.setup.wizard as wiz
+    monkeypatch.setattr(wiz.httpx, "get", MagicMock(side_effect=Exception("not found")))
+    assert wiz._fetch_lmstudio_available_models() == []
+
+
+def test_fetch_lmstudio_available_models_empty_data(monkeypatch):
+    import pyra.setup.wizard as wiz
+    mock_resp = MagicMock()
+    mock_resp.json.return_value = {"data": []}
+    mock_resp.raise_for_status = lambda: None
+    monkeypatch.setattr(wiz.httpx, "get", lambda *a, **kw: mock_resp)
+    assert wiz._fetch_lmstudio_available_models() == []
+
+
+# ── _load_lmstudio_model ──────────────────────────────────────────────────────
+
+def test_load_lmstudio_model_returns_true_on_success(monkeypatch):
+    import pyra.setup.wizard as wiz
+    mock_resp = MagicMock()
+    mock_resp.is_success = True
+    monkeypatch.setattr(wiz.httpx, "post", lambda *a, **kw: mock_resp)
+    assert wiz._load_lmstudio_model("gemma-4b") is True
+
+
+def test_load_lmstudio_model_returns_false_on_api_failure(monkeypatch):
+    import pyra.setup.wizard as wiz
+    mock_resp = MagicMock()
+    mock_resp.is_success = False
+    monkeypatch.setattr(wiz.httpx, "post", lambda *a, **kw: mock_resp)
+    assert wiz._load_lmstudio_model("gemma-4b") is False
+
+
+def test_load_lmstudio_model_returns_false_on_exception(monkeypatch):
+    import pyra.setup.wizard as wiz
+    monkeypatch.setattr(wiz.httpx, "post", MagicMock(side_effect=Exception("timeout")))
+    assert wiz._load_lmstudio_model("gemma-4b") is False
+
+
+# ── _classify_error ───────────────────────────────────────────────────────────
+
+def _fake_llm_exc(name: str) -> Exception:
+    """Create a fake litellm exception with the given class name."""
+    cls = type(name, (Exception,), {"__module__": "litellm.exceptions"})
+    return cls("test error")
+
+
+def test_classify_auth_error():
+    from pyra.setup.wizard import _classify_error
+    label, hint = _classify_error(_fake_llm_exc("AuthenticationError"))
+    assert "key" in label.lower()
+    assert "provider" in hint.lower() or "dashboard" in hint.lower()
+
+
+def test_classify_not_found_error():
+    from pyra.setup.wizard import _classify_error
+    label, hint = _classify_error(_fake_llm_exc("NotFoundError"))
+    assert "model" in label.lower()
+    assert "model" in hint.lower()
+
+
+def test_classify_rate_limit_error():
+    from pyra.setup.wizard import _classify_error
+    label, _ = _classify_error(_fake_llm_exc("RateLimitError"))
+    assert "rate" in label.lower()
+
+
+def test_classify_service_unavailable():
+    from pyra.setup.wizard import _classify_error
+    label, _ = _classify_error(_fake_llm_exc("ServiceUnavailableError"))
+    assert "unavailable" in label.lower()
+
+
+def test_classify_api_connection_error():
+    from pyra.setup.wizard import _classify_error
+    label, hint = _classify_error(_fake_llm_exc("APIConnectionError"))
+    assert "reach" in label.lower() or "network" in label.lower() or "connect" in label.lower()
+    assert "internet" in hint.lower() or "network" in hint.lower()
+
+
+def test_classify_timeout_error():
+    from pyra.setup.wizard import _classify_error
+    label, _ = _classify_error(_fake_llm_exc("TimeoutError"))
+    assert "time" in label.lower()
+
+
+def test_classify_bad_request_error():
+    from pyra.setup.wizard import _classify_error
+    label, hint = _classify_error(_fake_llm_exc("BadRequestError"))
+    assert "request" in label.lower() or "bad" in label.lower()
+    assert "model" in hint.lower()
+
+
+def test_classify_httpx_connect_error():
+    import httpx
+    from pyra.setup.wizard import _classify_error
+    label, hint = _classify_error(httpx.ConnectError("refused"))
+    assert "reach" in label.lower() or "reachable" in label.lower()
+    assert "running" in hint.lower() or "listening" in hint.lower()
+
+
+def test_classify_httpx_timeout():
+    import httpx
+    from pyra.setup.wizard import _classify_error
+    label, _ = _classify_error(httpx.TimeoutException("timeout"))
+    assert "time" in label.lower()
+
+
+def test_classify_generic_error():
+    from pyra.setup.wizard import _classify_error
+    label, hint = _classify_error(ValueError("something went wrong"))
+    assert len(label) > 0
+    assert "something went wrong" in hint
+
+
+def test_classify_error_always_returns_two_strings():
+    from pyra.setup.wizard import _classify_error
+    for exc in (RuntimeError("boom"), OSError("disk"), KeyError("k")):
+        label, hint = _classify_error(exc)
+        assert isinstance(label, str) and len(label) > 0
+        assert isinstance(hint, str) and len(hint) > 0
+
+
+# ── _check_local_server retry behaviour ──────────────────────────────────────
+
+def _silent_print(*a, **kw):
+    pass
+
+
+def test_check_local_server_success(monkeypatch):
+    import pyra.setup.wizard as wiz
+    mock_resp = MagicMock()
+    mock_resp.raise_for_status = lambda: None
+    monkeypatch.setattr(wiz.httpx, "get", lambda *a, **kw: mock_resp)
+    monkeypatch.setattr(wiz.console, "print", _silent_print)
+    from pyra.setup.providers import get_provider
+    wiz._check_local_server(get_provider("lmstudio"))  # must not raise
+
+
+def test_check_local_server_retry_then_success(monkeypatch):
+    import pyra.setup.wizard as wiz
+    from pyra.setup.providers import get_provider
+
+    call_count = {"n": 0}
+
+    def flaky_get(*a, **kw):
+        call_count["n"] += 1
+        if call_count["n"] == 1:
+            raise ConnectionError("refused")
+        m = MagicMock()
+        m.raise_for_status = lambda: None
+        return m
+
+    monkeypatch.setattr(wiz.httpx, "get", flaky_get)
+    monkeypatch.setattr(wiz.console, "print", _silent_print)
+    monkeypatch.setattr(wiz.questionary, "select",
+                        lambda *a, **kw: MagicMock(ask=lambda: "retry"))
+
+    wiz._check_local_server(get_provider("lmstudio"))
+    assert call_count["n"] == 2
+
+
+def test_check_local_server_abort_raises_system_exit(monkeypatch):
+    import pyra.setup.wizard as wiz
+    from pyra.setup.providers import get_provider
+
+    monkeypatch.setattr(wiz.httpx, "get", MagicMock(side_effect=ConnectionError("refused")))
+    monkeypatch.setattr(wiz.console, "print", _silent_print)
+    monkeypatch.setattr(wiz.questionary, "select",
+                        lambda *a, **kw: MagicMock(ask=lambda: "abort"))
+
+    with pytest.raises(SystemExit):
+        wiz._check_local_server(get_provider("lmstudio"))
+
+
+def test_check_local_server_continue_returns(monkeypatch):
+    import pyra.setup.wizard as wiz
+    from pyra.setup.providers import get_provider
+
+    monkeypatch.setattr(wiz.httpx, "get", MagicMock(side_effect=ConnectionError("refused")))
+    monkeypatch.setattr(wiz.console, "print", _silent_print)
+    monkeypatch.setattr(wiz.questionary, "select",
+                        lambda *a, **kw: MagicMock(ask=lambda: "continue"))
+
+    wiz._check_local_server(get_provider("lmstudio"))  # must return without raising
+
+
+# ── draft persistence ─────────────────────────────────────────────────────────
+
+def test_save_and_load_draft(tmp_pyra_home):
+    import pyra.setup.wizard as wiz
+    state = {"completed_steps": ["profile"], "user_name": "Alice"}
+    wiz._save_draft(state)
+    loaded = wiz._load_draft()
+    assert loaded == state
+
+
+def test_load_draft_returns_none_when_no_file(tmp_pyra_home):
+    import pyra.setup.wizard as wiz
+    assert wiz._load_draft() is None
+
+
+def test_delete_draft_removes_file(tmp_pyra_home):
+    import pyra.setup.wizard as wiz
+    wiz._save_draft({"completed_steps": []})
+    wiz._delete_draft()
+    assert wiz._load_draft() is None
+
+
+def test_delete_draft_is_idempotent(tmp_pyra_home):
+    import pyra.setup.wizard as wiz
+    wiz._delete_draft()
+    wiz._delete_draft()
+
+
+def test_mark_step_done_appends_once(tmp_pyra_home):
+    import pyra.setup.wizard as wiz
+    state = {}
+    wiz._mark_step_done(state, "profile")
+    wiz._mark_step_done(state, "profile")
+    assert state["completed_steps"].count("profile") == 1
+
+
+def test_draft_file_has_correct_permissions(tmp_pyra_home):
+    import os
+    import pyra.setup.wizard as wiz
+    wiz._save_draft({"completed_steps": []})
+    path = wiz._draft_path()
+    if os.name != "nt":
+        mode = oct(path.stat().st_mode)[-3:]
+        assert mode == "600"
+
+
+# ── fetch_loaded_models ───────────────────────────────────────────────────────
+
+def test_fetch_loaded_models_ollama_uses_api_ps(monkeypatch):
+    import pyra.setup.wizard as wiz
+    from pyra.setup.providers import get_provider
+    mock_resp = MagicMock()
+    mock_resp.json.return_value = {"models": [{"name": "llama3:latest"}, {"name": "mistral"}]}
+    mock_resp.raise_for_status = lambda: None
+    calls = []
+    monkeypatch.setattr(wiz.httpx, "get", lambda url, **kw: (calls.append(url), mock_resp)[1])
+    result = wiz.fetch_loaded_models(get_provider("ollama"))
+    assert result == ["llama3:latest", "mistral"]
+    assert any("/api/ps" in u for u in calls)
+
+
+def test_fetch_loaded_models_lmstudio_uses_beta_api_and_filters(monkeypatch):
+    import pyra.setup.wizard as wiz
+    from pyra.setup.providers import get_provider
+    mock_resp = MagicMock()
+    mock_resp.json.return_value = {
+        "data": [
+            {"id": "gemma-4b", "state": "loaded"},
+            {"id": "llama3", "state": "not_loaded"},
+        ]
+    }
+    mock_resp.raise_for_status = lambda: None
+    calls = []
+    monkeypatch.setattr(wiz.httpx, "get", lambda url, **kw: (calls.append(url), mock_resp)[1])
+    result = wiz.fetch_loaded_models(get_provider("lmstudio"))
+    assert result == ["gemma-4b"]
+    assert any("/api/v0/models" in u for u in calls)
+
+
+def test_fetch_loaded_models_lmstudio_filters_unloaded(monkeypatch):
+    import pyra.setup.wizard as wiz
+    from pyra.setup.providers import get_provider
+    mock_resp = MagicMock()
+    mock_resp.json.return_value = {
+        "data": [
+            {"id": "model-a", "state": "not_loaded"},
+            {"id": "model-b", "state": "not_loaded"},
+        ]
+    }
+    mock_resp.raise_for_status = lambda: None
+    monkeypatch.setattr(wiz.httpx, "get", lambda url, **kw: mock_resp)
+    assert wiz.fetch_loaded_models(get_provider("lmstudio")) == []
+
+
+def test_fetch_loaded_models_returns_empty_on_error(monkeypatch):
+    import pyra.setup.wizard as wiz
+    from pyra.setup.providers import get_provider
+    monkeypatch.setattr(wiz.httpx, "get", MagicMock(side_effect=Exception("conn refused")))
+    assert wiz.fetch_loaded_models(get_provider("ollama")) == []
+
+
+def test_fetch_loaded_models_returns_empty_when_no_base_url():
+    import pyra.setup.wizard as wiz
+    from pyra.setup.providers import Provider
+    provider = Provider(
+        id="test", display_name="Test", requires_key=False,
+        default_model="x", litellm_prefix="openai/", group="Local",
+    )
+    assert wiz.fetch_loaded_models(provider) == []
+
+
+# ── _show_local_model_status ──────────────────────────────────────────────────
+
+def test_show_local_model_status_one_model(monkeypatch):
+    import pyra.setup.wizard as wiz
+    from pyra.setup.providers import get_provider
+    monkeypatch.setattr(wiz, "fetch_loaded_models", lambda p: ["gemma-4b"])
+    printed = []
+    monkeypatch.setattr(wiz.console, "print", lambda *a, **kw: printed.append(str(a)))
+    wiz._show_local_model_status(get_provider("lmstudio"))
+    assert any("gemma-4b" in s for s in printed)
+
+
+def test_show_local_model_status_none(monkeypatch):
+    import pyra.setup.wizard as wiz
+    from pyra.setup.providers import get_provider
+    monkeypatch.setattr(wiz, "fetch_loaded_models", lambda p: [])
+    printed = []
+    monkeypatch.setattr(wiz.console, "print", lambda *a, **kw: printed.append(str(a)))
+    wiz._show_local_model_status(get_provider("lmstudio"))
+    assert any("No model" in s for s in printed)
+
+
+def test_show_local_model_status_multiple(monkeypatch):
+    import pyra.setup.wizard as wiz
+    from pyra.setup.providers import get_provider
+    monkeypatch.setattr(wiz, "fetch_loaded_models", lambda p: ["a", "b", "c"])
+    printed = []
+    monkeypatch.setattr(wiz.console, "print", lambda *a, **kw: printed.append(str(a)))
+    wiz._show_local_model_status(get_provider("lmstudio"))
+    combined = " ".join(printed)
+    assert "3" in combined or ("a" in combined and "b" in combined)
+
+
+# ── _test_connection model re-entry ───────────────────────────────────────────
+
+def test_test_connection_change_model(tmp_pyra_home, monkeypatch):
+    import pyra.setup.wizard as wiz
+    from pyra.setup.providers import get_provider
+
+    call_count = {"n": 0}
+
+    class FakeNotFound(Exception):
+        pass
+    FakeNotFound.__module__ = "litellm.exceptions"
+    FakeNotFound.__name__ = "NotFoundError"
+
+    def fake_completion(**kwargs):
+        call_count["n"] += 1
+        if call_count["n"] == 1:
+            raise FakeNotFound("model not found")
+
+    import litellm
+    monkeypatch.setattr(litellm, "completion", fake_completion)
+    monkeypatch.setattr(wiz.console, "print", _silent_print)
+    monkeypatch.setattr(wiz.questionary, "select",
+                        lambda *a, **kw: MagicMock(ask=lambda: "change_model"))
+    monkeypatch.setattr(wiz.questionary, "text",
+                        lambda *a, **kw: MagicMock(ask=lambda: "new-model"))
+    monkeypatch.setattr(wiz, "_fetch_local_models", lambda p: [])
+    monkeypatch.setattr("pyra.vault.reader.get_key", lambda k: "sk-test")
+
+    provider = get_provider("anthropic")
+    result = wiz._test_connection(provider, "old-model")
+    assert result == "new-model"
+    assert call_count["n"] == 2