feat(catalog): overlay local route visibility on mirrored sources

Author: André Lange

CHANGELOG.md

@@ -1,5 +1,17 @@
 # fusionAIze Gate Changelog
 
+## v1.13.0 - Unreleased
+
+### Added
+
+- Expanded the provider source catalog scope beyond `blackbox`, `kilo`, and `openai` so Gate can also track mirrored official source data for `anthropic`, `deepseek`, and `google`
+- Added local models-endpoint overlays per configured route, which lets Gate compare what a specific key can really see against the mirrored global provider catalog
+
+### Changed
+
+- Provider source alerts now distinguish more clearly between global catalog drift and key-specific route/model visibility drift
+- Catalog summaries now include local route counts, local visible model counts, and route-vs-catalog mismatch hints instead of only source freshness and change counts
+
 ## v1.12.0 - 2026-03-29
 
 ### Added

docs/CONFIGURATION.md

@@ -384,7 +384,10 @@ provider_source_refresh:
   timeout_seconds: 10.0
   interval_seconds: 21600
   providers:
+    - anthropic
     - blackbox
+    - deepseek
+    - google
     - kilo
     - openai
 ```
@@ -393,4 +396,5 @@ Notes:
 - startup refresh is best-effort and should not block the service if docs are unavailable
 - `interval_seconds` controls the conservative background refresh loop after startup
 - source snapshots live in the same local DB as metrics
+- for providers with a usable local `models` endpoint, Gate also mirrors key-specific model visibility per configured route and compares that against the global source snapshot
 - local billing overlays such as subscription or quota windows belong in the local account profile layer, not in the global provider snapshot

docs/FAIGATE-ROADMAP.md

@@ -37,29 +37,31 @@ The detailed design lives in [Adaptive model orchestration](./ADAPTIVE-ORCHESTRA
 
 The next block should stay disciplined: build on the workstation baseline, keep packaging practical, and avoid turning fusionAIze Gate into a sprawling platform.
 
-## Current release target: `v1.12.0`
+## Current release target: `v1.13.0`
 
-The next release should land as a clean operational release, not as another loose pile of runtime slices.
+`v1.12.0` closed the first operator-facing catalog and release-hardening loop. The
+next release should make that catalog meaningfully more alive instead of merely
+more visible.
 
-`v1.12.0` should close around three themes that now fit together:
+`v1.13.0` should close around three tightly related themes:
 
-- provider source cataloging and alerting as a first-class operator surface
-- clearer aggregator behavior for Kilo and BLACKBOX, especially where "free", "budget", "wallet", and explicit paid lanes are easy to conflate
-- hardened release automation after the `v1.11.x` release failures
+- provider source catalog moves from mirrored docs pages to a more living operator dataset
+- local key and route visibility are overlaid against global provider source snapshots
+- provider drift gets classified more clearly as global docs drift, key-specific access drift, or route-level mismatch
 
 The release should feel coherent from an operator point of view:
 
-- Quick Setup, Doctor, Provider Probe, Dashboard, and route preview all explain drift or route pressure using the same language
-- Kilo explicit Sonnet/Opus lanes are visible as deliberate routing choices instead of hidden aggregator magic
-- release prep, tag validation, and publish dry-runs are boring and repeatable again
+- Doctor, Provider Probe, Dashboard, Quick Setup, and `/api/provider-catalog` tell the same story about what changed globally and what is only true for this key or route
+- free-tier, paid-tier, wallet, and BYOK assumptions are treated as per-key operational facts instead of being inferred blindly from public pricing tables
+- the provider source catalog becomes a trustworthy early-warning surface before route selection starts leaning on outdated assumptions
 
-What is intentionally not in scope for `v1.12.0`:
+What is intentionally not in scope for `v1.13.0`:
 
 - the virtual key layer
 - gateway-level response caching
-- fully automated external provider-source crawling on a long-running schedule
+- a large new bridge or client-surface expansion
 
-Those stay as follow-on tracks once the operator surfaces, release path, and aggregator semantics are stable enough to trust.
+Those stay as follow-on tracks once the provider catalog and route-availability overlay are stable enough to trust under real operator workflows.
 
 ## Shipped: `v1.8.0` – `v1.9.1`
 

faigate/config.py

@@ -1704,9 +1704,10 @@ def _normalize_provider_source_refresh(data: dict[str, Any]) -> dict[str, Any]:
     if interval_seconds <= 0:
         raise ConfigError("'provider_source_refresh.interval_seconds' must be positive")
 
-    providers = raw.get("providers", ["blackbox", "kilo", "openai"])
+    default_providers = ["anthropic", "blackbox", "deepseek", "google", "kilo", "openai"]
+    providers = raw.get("providers", default_providers)
     if providers in (None, ""):
-        providers = ["blackbox", "kilo", "openai"]
+        providers = default_providers
     if not isinstance(providers, list) or any(
         not isinstance(item, str) or not item.strip() for item in providers
     ):
@@ -1881,7 +1882,14 @@ def provider_source_refresh(self) -> dict:
                 "on_startup": True,
                 "timeout_seconds": 10.0,
                 "interval_seconds": 21600,
-                "providers": ["blackbox", "kilo", "openai"],
+                "providers": [
+                    "anthropic",
+                    "blackbox",
+                    "deepseek",
+                    "google",
+                    "kilo",
+                    "openai",
+                ],
             },
         )
 

faigate/main.py

@@ -38,6 +38,10 @@
 )
 from .lane_registry import get_provider_lane_binding, get_route_add_recommendations
 from .metrics import MetricsStore, calc_cost
+from .provider_availability import (
+    record_availability_from_config,
+    refresh_local_model_availability,
+)
 from .provider_catalog import (
     build_provider_catalog_report,
     build_provider_discovery_view,
@@ -74,6 +78,10 @@
 _provider_catalog_refresh_task: asyncio.Task[None] | None = None
 
 
+def _provider_catalog_config_path() -> str:
+    return str(os.environ.get("FAIGATE_CONFIG_FILE") or "config.yaml")
+
+
 class PayloadTooLargeError(ValueError):
     """Raised when one request or upload exceeds configured size limits."""
 
@@ -204,6 +212,21 @@ async def _refresh_provider_source_catalog(*, force: bool = False) -> list[dict[
         provider_ids=target_ids,
         timeout_seconds=float(source_refresh_cfg.get("timeout_seconds") or 10.0),
     )
+    await asyncio.to_thread(
+        record_availability_from_config,
+        _provider_catalog_store,
+        config_path=_provider_catalog_config_path(),
+        health_payload={
+            "providers": {item["name"]: item for item in _build_provider_inventory()}
+        },
+    )
+    await asyncio.to_thread(
+        refresh_local_model_availability,
+        _provider_catalog_store,
+        config_path=_provider_catalog_config_path(),
+        provider_ids=target_ids,
+        timeout_seconds=float(source_refresh_cfg.get("timeout_seconds") or 10.0),
+    )
     ok_count = sum(1 for item in refresh_results if item.ok)
     logger.info(
         "Provider source refresh completed: %s/%s source endpoints succeeded (%s)",
@@ -1908,6 +1931,14 @@ async def provider_catalog():
         "priority_next": {},
     }
     if _provider_catalog_store is not None:
+        await asyncio.to_thread(
+            record_availability_from_config,
+            _provider_catalog_store,
+            config_path=_provider_catalog_config_path(),
+            health_payload={
+                "providers": {item["name"]: item for item in _build_provider_inventory()}
+            },
+        )
         source_catalog = build_catalog_summary(
             _provider_catalog_store,
             provider_ids=list(_config.provider_source_refresh.get("providers") or []),