fix(discovery): fall back to serving-endpoints for OSS models

OSS model discovery only queried UC model-services for `system.ai.*`
ids and, unlike the claude/gemini/codex families, had no AI-Gateway
fallback. On workspaces that expose OSS foundation models as regular
`databricks-*` serving endpoints (rather than `system.ai.*` UC
model-services), `oss_models` came back empty and pi/opencode never
offered GLM, Qwen, Gemma, inkling, etc.

Add `discover_oss_models`, which lists foundation-model endpoints
advertising `mlflow/v1/chat/completions` and keeps the OSS chat families
via the existing `_is_oss_chat_model` filter (the family filter is
required because Claude/Gemini endpoints can advertise the same dialect
on some workspaces). Wire it as the fallback in cli.py, mirroring the
other three families.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: 2 people

src/ucode/cli.py

@@ -37,6 +37,7 @@
     discover_codex_models,
     discover_gemini_models,
     discover_model_services,
+    discover_oss_models,
     ensure_ai_gateway_v2,
     ensure_databricks_auth,
     ensure_pat_bearer,
@@ -391,6 +392,8 @@ def configure_shared_state(
                     codex_models, codex_reason = discover_codex_models(workspace, token)
             if want_oss:
                 oss_models, oss_reason = ms_oss, ms_reason
+                if not oss_models:
+                    oss_models, oss_reason = discover_oss_models(workspace, token)
         if claude_models:
             opencode_models["anthropic"] = list(claude_models.values())
         if gemini_models:

src/ucode/databricks.py

@@ -2056,6 +2056,34 @@ def discover_codex_models(workspace: str, token: str) -> tuple[list[str], str |
     return discover_endpoints_with_api_type(workspace, token, "openai/v1/responses")
 
 
+def discover_oss_models(workspace: str, token: str) -> tuple[list[str], str | None]:
+    """Discover OSS chat models served as AI Gateway foundation-model endpoints.
+
+    Fallback for workspaces that don't register OSS foundation models as
+    `system.ai.*` UC model-services (see `discover_model_services`): those
+    workspaces expose the same models as regular `databricks-*` serving
+    endpoints instead. Lists every endpoint advertising the
+    `mlflow/v1/chat/completions` dialect, then keeps only the OSS chat families
+    (`_is_oss_chat_model`) — on some workspaces the Claude/Gemini endpoints also
+    advertise that dialect, so the family filter is what separates the OSS
+    cohort from them. Mirrors the AI-Gateway fallback the other families use
+    when the UC model-services listing is empty.
+    """
+    endpoints, reason = discover_endpoints_with_api_type(
+        workspace, token, "mlflow/v1/chat/completions"
+    )
+    if not endpoints:
+        return [], reason
+    oss = [e for e in endpoints if _is_oss_chat_model(e)]
+    if oss:
+        return oss, None
+    sample = ", ".join(endpoints[:5])
+    return [], (
+        "foundation-models exposing `mlflow/v1/chat/completions` matched no OSS "
+        f"chat family (got: {sample})"
+    )
+
+
 def fetch_gemini_models(workspace: str, token: str) -> list[str]:
     models, _ = discover_gemini_models(workspace, token)
     return models

tests/test_databricks.py

@@ -923,6 +923,82 @@ def test_codex_discovery_keeps_alphabetical_order(self, monkeypatch):
         assert models == ["databricks-gpt-4-1", "databricks-gpt-5-2-codex"]
 
 
+def _mlflow_chat_payload(names, *, api_type="mlflow/v1/chat/completions", v2=True):
+    return {
+        "endpoints": [
+            {
+                "name": name,
+                "config": {
+                    "served_entities": [
+                        {
+                            "foundation_model": {
+                                "ai_gateway_v2_supported": v2,
+                                "api_types": [api_type],
+                            }
+                        }
+                    ]
+                },
+            }
+            for name in names
+        ]
+    }
+
+
+class TestDiscoverOssModels:
+    def test_finds_oss_endpoints_via_foundation_models(self, monkeypatch):
+        # Mirrors a workspace with no system.ai UC model-services: OSS models are
+        # plain databricks-* serving endpoints under the mlflow chat dialect.
+        payload = _mlflow_chat_payload(
+            [
+                "databricks-glm-5-2",
+                "databricks-inkling",
+                "databricks-qwen35-122b-a10b",
+                "databricks-gemma-3-12b",
+            ]
+        )
+        monkeypatch.setattr(db_mod, "_http_get_json", lambda url, token: (payload, None))
+
+        models, reason = db_mod.discover_oss_models(WS, "token")
+
+        assert reason is None
+        assert "databricks-glm-5-2" in models
+        assert set(models) == {
+            "databricks-glm-5-2",
+            "databricks-inkling",
+            "databricks-qwen35-122b-a10b",
+            "databricks-gemma-3-12b",
+        }
+
+    def test_excludes_claude_and_gemini_sharing_the_mlflow_dialect(self, monkeypatch):
+        # On some workspaces every foundation model advertises the mlflow chat
+        # dialect, so the api_type filter alone is too broad — the OSS family
+        # filter must drop Claude/Gemini and keep only the OSS cohort.
+        payload = _mlflow_chat_payload(
+            [
+                "databricks-claude-opus-4-8",
+                "databricks-gemini-2-5-pro",
+                "databricks-glm-5-2",
+                "databricks-qwen3-embedding-0-6b",
+            ]
+        )
+        monkeypatch.setattr(db_mod, "_http_get_json", lambda url, token: (payload, None))
+
+        models, reason = db_mod.discover_oss_models(WS, "token")
+
+        assert reason is None
+        assert models == ["databricks-glm-5-2"]
+
+    def test_reports_reason_when_no_oss_family_matches(self, monkeypatch):
+        payload = _mlflow_chat_payload(["databricks-claude-opus-4-8"])
+        monkeypatch.setattr(db_mod, "_http_get_json", lambda url, token: (payload, None))
+
+        models, reason = db_mod.discover_oss_models(WS, "token")
+
+        assert models == []
+        assert reason is not None
+        assert "no OSS" in reason
+
+
 class TestResolvePatToken:
     def test_reads_pat_profile_token_from_cfg(self, monkeypatch, tmp_path):
         cfg = tmp_path / "databrickscfg"