homeassistant-ai
diff --git a/‎homeassistant-addon-dev/config.yaml‎
Lines changed: 0 additions & 1 deletion b/‎homeassistant-addon-dev/config.yaml‎
Lines changed: 0 additions & 1 deletion
diff --git a/‎homeassistant-addon-dev/translations/en.yaml‎
Lines changed: 0 additions & 7 deletions b/‎homeassistant-addon-dev/translations/en.yaml‎
Lines changed: 0 additions & 7 deletions
diff --git a/‎homeassistant-addon/config.yaml‎
Lines changed: 0 additions & 1 deletion b/‎homeassistant-addon/config.yaml‎
Lines changed: 0 additions & 1 deletion
diff --git a/‎homeassistant-addon/start.py‎
Lines changed: 51 additions & 7 deletions b/‎homeassistant-addon/start.py‎
Lines changed: 51 additions & 7 deletions
diff --git a/‎homeassistant-addon/translations/en.yaml‎
Lines changed: 0 additions & 7 deletions b/‎homeassistant-addon/translations/en.yaml‎
Lines changed: 0 additions & 7 deletions
diff --git a/‎src/ha_mcp/client/rest_client.py‎
Lines changed: 28 additions & 8 deletions b/‎src/ha_mcp/client/rest_client.py‎
Lines changed: 28 additions & 8 deletions
diff --git a/‎src/ha_mcp/tools/backup.py‎
Lines changed: 14 additions & 2 deletions b/‎src/ha_mcp/tools/backup.py‎
Lines changed: 14 additions & 2 deletions
diff --git a/‎src/ha_mcp/tools/tools_utility.py‎
Lines changed: 25 additions & 2 deletions b/‎src/ha_mcp/tools/tools_utility.py‎
Lines changed: 25 additions & 2 deletions
@@ -86,6 +86,5 @@ schema:
   disabled_tools: str?
   pinned_tools: str?
   verify_ssl: bool?
-  advanced_debug_logging: bool?
 ports:
   9583/tcp: 9583
@@ -261,10 +261,3 @@ configuration:
       certificate or hostname mismatch. Disabling weakens transport
       security — leave on unless you know you need it. Requires restart
       to take effect.
-  advanced_debug_logging:
-    name: Advanced debug logging
-    description: >-
-      Captures extra diagnostic info on kill/shutdown — sender PID
-      and memory state. Useful when reporting unexplained add-on
-      stops or crash loops. Off by default to keep logs concise.
-      Requires restart to take effect.
@@ -70,7 +70,6 @@ schema:
   auto_backup_throttle_minutes: int(0,1440)?
   auto_backup_retain_per_entity: int(1,10000)?
   verify_ssl: bool?
-  advanced_debug_logging: bool?
 # Add-on exposes HTTP port for MCP communication (fixed internal port)
 ports:
   9583/tcp: 9583
@@ -202,6 +202,40 @@ def resolve_bool_option(config: dict[str, Any], key: str, default: bool) -> bool
     return default
 
 
+def resolve_effective_log_level() -> int:
+    """Return the root log level from ha-mcp's effective settings.
+
+    Reads the web Settings UI "Log level" advanced setting (persisted
+    under ``/data`` and applied by ``get_global_settings()``) so the
+    addon log actually honors it — ``main()`` configures logging before
+    ha-mcp is imported, so it can only hardcode INFO at that point and
+    must re-apply the real level after the import (#1721).
+
+    Call only after ``main()`` has exported ``HOMEASSISTANT_URL`` /
+    ``HOMEASSISTANT_TOKEN``: ``get_global_settings()`` caches a
+    singleton, so an earlier call would pin placeholder connection
+    settings for the process. Any failure falls back to INFO — logging
+    config must never block addon startup.
+    """
+    import logging
+
+    try:
+        from ha_mcp.config import get_global_settings
+
+        return getattr(logging, get_global_settings().log_level, logging.INFO)
+    except Exception as e:
+        # Loud fallback: without this line, a user who set DEBUG in the
+        # web UI can't tell "I'm on INFO" from "my DEBUG request crashed
+        # on load" — the same silent-no-op class this fix exists to kill.
+        # print-based so it reaches the addon log regardless of logging
+        # state.
+        log_warning(
+            f"Could not resolve effective log level from settings; "
+            f"defaulting to INFO (web-UI Log level not applied): {e!r}"
+        )
+        return logging.INFO
+
+
 _DEV_ADDON_BETA_KEYS = (
     "enable_yaml_config_editing",
     # Per-key sub-gates of enable_yaml_config_editing. Kept in lockstep
@@ -353,7 +387,6 @@ def main() -> int:
     disabled_tools_raw = ""  # default
     pinned_tools_raw = ""  # default
     verify_ssl = True  # default
-    advanced_debug_logging = False  # default
 
     if config_file.exists():
         try:
@@ -493,9 +526,6 @@ def main() -> int:
             raw_pinned = config.get("pinned_tools", "")
             pinned_tools_raw = raw_pinned if isinstance(raw_pinned, str) else ""
             verify_ssl = resolve_bool_option(config, "verify_ssl", True)
-            advanced_debug_logging = resolve_bool_option(
-                config, "advanced_debug_logging", False
-            )
         except Exception as e:
             log_error(f"Failed to read config: {e}, using defaults")
             # Persistent "you lost your features" line so an operator
@@ -528,7 +558,6 @@ def main() -> int:
 
     log_info(f"Backup hint mode: {backup_hint}")
     log_info(f"Verify SSL: {verify_ssl}")
-    log_info(f"Advanced debug logging: {advanced_debug_logging}")
 
     # Set up environment for ha-mcp
     os.environ["HOMEASSISTANT_URL"] = "http://supervisor/core"
@@ -711,7 +740,22 @@ def main() -> int:
     # FastMCP surfaces its update notice in these same startup logs.
     _log_startup_version()
 
-    if advanced_debug_logging:
+    # Re-apply the effective log level now that ha_mcp is imported —
+    # the basicConfig above could only hardcode INFO. Without this, the
+    # web Settings UI "Log level" setting never reaches the addon log
+    # (#1721).
+    effective_log_level = resolve_effective_log_level()
+    logging.getLogger().setLevel(effective_log_level)
+    # Proof-of-application canary: emitted through the logging system
+    # (not print), so it reaches the addon log ONLY when the DEBUG level
+    # actually took effect. Support threads can key on this line to tell
+    # "user set DEBUG" apart from "DEBUG is really active".
+    logging.getLogger("addon_start").debug(
+        "Debug logging active (log_level applied from settings)"
+    )
+
+    if effective_log_level == logging.DEBUG:
+        log_info("Debug log level active — arming kill-signal diagnostics")
         # Defers SA_SIGINFO install until uvicorn's capture_signals has
         # run. Otherwise uvicorn's signal.signal() call would overwrite
         # our handler before any signal arrived.
@@ -723,7 +767,7 @@ def main() -> int:
 
             schedule_install_after_uvicorn()
         except Exception as e:
-            log_error(f"advanced_debug_logging install failed: {e!r}; continuing")
+            log_error(f"kill-signal diagnostics install failed: {e!r}; continuing")
 
     register_browser_landing(mcp, secret_path)
     # Mount settings UI routes both at root (for HA ingress proxy) and
 
@@ -99,13 +99,6 @@ configuration:
       certificate or hostname mismatch. Disabling weakens transport
       security — leave on unless you know you need it. Requires restart
       to take effect.
-  advanced_debug_logging:
-    name: Advanced debug logging
-    description: >-
-      Captures extra diagnostic info on kill/shutdown — sender PID
-      and memory state. Useful when reporting unexplained add-on
-      stops or crash loops. Off by default to keep logs concise.
-      Requires restart to take effect.
   tool_search_max_results:
     name: Tool search max results
     description: >-
 
@@ -606,7 +606,7 @@ async def _is_supervised_install(self) -> bool:
         self._supervised_detected = is_supervised
         return is_supervised
 
-    async def get_addon_logs(self, slug: str) -> str:
+    async def get_addon_logs(self, slug: str, lines: int | None = None) -> str:
         """Fetch an add-on's container logs.
 
         Branch on ``is_running_in_addon()`` (which keys off ``SUPERVISOR_TOKEN``
@@ -621,6 +621,15 @@ async def get_addon_logs(self, slug: str) -> str:
         HA URL), falls back to the HA Core proxy path. That path requires an
         admin LLA but works fine when not invoked from the add-on container.
 
+        ``lines`` sets the journald window Supervisor serves via its
+        ``?lines=`` query param (supervisor/api/host.py). Without it,
+        Supervisor returns only its default window (``DEFAULT_LINES = 100``
+        in supervisor/api/const.py) — which silently capped every larger
+        caller-side limit before this param existed. Both branches carry
+        it: the direct endpoint parses the query natively, and HA Core's
+        hassio proxy forwards ``request.query`` upstream
+        (homeassistant/components/hassio/http.py).
+
         Both branches return ``text/plain`` log content.
 
         Raises:
@@ -633,17 +642,18 @@ async def get_addon_logs(self, slug: str) -> str:
             HomeAssistantConnectionError: Network, timeout, or transport error.
         """
         if is_running_in_addon():
-            return await self._get_addon_logs_via_supervisor(slug)
+            return await self._get_addon_logs_via_supervisor(slug, lines=lines)
 
         logger.debug(f"Fetching addon logs for slug={slug} via HA Core proxy")
         response = await self._raw_request(
             "GET",
             f"/hassio/addons/{slug}/logs",
             headers={"Accept": "text/plain"},
+            params={"lines": lines} if lines is not None else None,
         )
         return response.text
 
-    async def _supervisor_logs_get(self, path: str) -> str:
+    async def _supervisor_logs_get(self, path: str, lines: int | None = None) -> str:
         """Fetch ``text/plain`` logs from a Supervisor REST endpoint.
 
         ``path`` is everything between ``http://supervisor/`` and ``/logs``:
@@ -652,6 +662,9 @@ async def _supervisor_logs_get(self, path: str) -> str:
         - ``"<service>"`` (where service ∈ {supervisor, host, core, dns, audio,
           cli, multicast, observer}) for system-service logs
 
+        ``lines`` maps to the endpoint's ``?lines=`` journald-window query
+        param; omitted → Supervisor's 100-line default window.
+
         Bypasses ``HomeAssistantClient.httpx_client`` because the Supervisor
         endpoint uses a different base URL (``http://supervisor``) and a
         different token (``SUPERVISOR_TOKEN``) than HA Core REST. Both
@@ -681,9 +694,10 @@ async def _supervisor_logs_get(self, path: str) -> str:
 
         relative_path = f"/{path}/logs"
         logger.debug(
-            "Fetching %s%s via Supervisor direct",
+            "Fetching %s%s via Supervisor direct (lines=%s)",
             get_supervisor_base_url(),
             relative_path,
+            lines,
         )
 
         try:
@@ -694,6 +708,7 @@ async def _supervisor_logs_get(self, path: str) -> str:
                 response = await client.get(
                     relative_path,
                     headers={"Accept": "text/plain"},
+                    params={"lines": lines} if lines is not None else None,
                 )
         except httpx.TimeoutException as e:
             raise HomeAssistantConnectionError(
@@ -753,7 +768,9 @@ async def _supervisor_logs_get(self, path: str) -> str:
             )
         return response.text
 
-    async def _get_addon_logs_via_supervisor(self, slug: str) -> str:
+    async def _get_addon_logs_via_supervisor(
+        self, slug: str, lines: int | None = None
+    ) -> str:
         """Fetch add-on container logs directly from Supervisor's REST API.
 
         Distinct from ``tools_bug_report._fetch_addon_logs``: that helper is
@@ -766,9 +783,11 @@ async def _get_addon_logs_via_supervisor(self, slug: str) -> str:
         Delegates to ``_supervisor_logs_get`` so error handling stays in
         lockstep with ``_get_system_service_logs``.
         """
-        return await self._supervisor_logs_get(f"addons/{slug}")
+        return await self._supervisor_logs_get(f"addons/{slug}", lines=lines)
 
-    async def _get_system_service_logs(self, service: str) -> str:
+    async def _get_system_service_logs(
+        self, service: str, lines: int | None = None
+    ) -> str:
         """Fetch HA system-service logs.
 
         ``service`` must be one of the eight Supervisor-managed services:
@@ -796,13 +815,14 @@ async def _get_system_service_logs(self, service: str) -> str:
         logs) call kept working through its own Core-proxy fallback.
         """
         if is_running_in_addon():
-            return await self._supervisor_logs_get(service)
+            return await self._supervisor_logs_get(service, lines=lines)
 
         logger.debug(f"Fetching {service} logs via HA Core proxy")
         response = await self._raw_request(
             "GET",
             f"/hassio/{service}/logs",
             headers={"Accept": "text/plain"},
+            params={"lines": lines} if lines is not None else None,
         )
         return response.text
 
 
@@ -329,11 +329,19 @@ async def _poll_backup_completion(
         )
 
         if event_state == "failed":
+            # Surface the failure event verbatim — HA's failed event
+            # carries the cause (e.g. another backup already in
+            # progress), and dropping it left CI failures reading
+            # "Backup creation failed: Backup creation failed" with
+            # nothing to diagnose.
             raise_tool_error(
                 create_error_response(
                     ErrorCode.SERVICE_CALL_FAILED,
                     "Backup creation failed",
-                    context={"backup_job_id": backup_job_id},
+                    context={
+                        "backup_job_id": backup_job_id,
+                        "last_action_event": last_event,
+                    },
                 )
             )
 
@@ -410,7 +418,11 @@ async def _poll_backup_completion(
                 create_error_response(
                     ErrorCode.SERVICE_CALL_FAILED,
                     "Backup creation failed (observed at post-timeout lookup)",
-                    context={"backup_job_id": backup_job_id, "name": name},
+                    context={
+                        "backup_job_id": backup_job_id,
+                        "name": name,
+                        "last_action_event": last_event,
+                    },
                 )
             )
         if state != "idle":
 
@@ -53,6 +53,11 @@
 
 DEFAULT_LIMIT = 50
 DEFAULT_LOG_LIMIT = 100
+# Journald window to request from Supervisor when a search filter is
+# active: matches are found within the fetched window only, so a
+# search over the caller's (often small) limit needs more history
+# behind it than the limit itself.
+SUPERVISOR_SEARCH_WINDOW_LINES = 2000
 MAX_LIMIT = 500
 
 # Regex to match log level at the start of a log line
@@ -752,8 +757,18 @@ async def _get_supervisor_log(
             limit, default=DEFAULT_LOG_LIMIT, suggestion_example="100"
         )
 
+        # Request a journald window sized to the caller's limit —
+        # Supervisor's /logs endpoints default to their last-100-lines
+        # window, which silently capped any larger limit before ?lines=
+        # was plumbed through (found via #1721's e2e).
+        fetch_lines = (
+            max(effective_limit, SUPERVISOR_SEARCH_WINDOW_LINES)
+            if search
+            else effective_limit
+        )
+
         try:
-            log_text = await self._client.get_addon_logs(slug)
+            log_text = await self._client.get_addon_logs(slug, lines=fetch_lines)
 
             lines = log_text.splitlines() if log_text else []
 
@@ -892,8 +907,16 @@ async def _get_system_service_log(
             limit, default=DEFAULT_LOG_LIMIT, suggestion_example="100"
         )
 
+        fetch_lines = (
+            max(effective_limit, SUPERVISOR_SEARCH_WINDOW_LINES)
+            if search
+            else effective_limit
+        )
+
         try:
-            log_text = await self._client._get_system_service_logs(service)
+            log_text = await self._client._get_system_service_logs(
+                service, lines=fetch_lines
+            )
 
             lines = log_text.splitlines() if log_text else []