Remove Claude Haiku 4.5 from the selectable model list

DeepSeek V4 Flash is now the only entry in AVAILABLE_MODELS; the picker
and /api/chat validation both derive from it, so Haiku is gone from the
UI and 422s ("Unknown model") for direct API callers. Anthropic support
stays in code (build_chat_model, tool catalog, pricing, preset rules),
so re-adding it later is a one-line revert. Tests that needed a
selectable model with web toggles re-admit Haiku via a patched
AVAILABLE_MODELS.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

AGENTS.md

@@ -6,7 +6,7 @@ This is the **canonical, tool-agnostic** instruction file for the repo. `CLAUDE.
 
 AI tutor for applied AI, LLMs, RAG, and Python. **Agentic RAG**: a LangChain/LangGraph agent grounds answers in a curated corpus of course + library docs, can browse a local file-based knowledge base, and (optionally) search the live web. One frontend: a **Next.js** UI (`frontend/`), served by a **FastAPI** backend (`app/api.py`) that streams in the Vercel AI SDK UI-message protocol. (A Gradio UI existed historically; it was removed to keep one rendering path.)
 
-ChromaDB for vectors; Cohere for embeddings/rerank; chat model is provider-configurable (DeepSeek V4 Flash through the first-party API by default, with a rescue-only in-app fallback to Gemini 2.5 Flash when a Gemini key is set; Claude Haiku 4.5 is also selectable; OpenAI and OpenRouter-compatible models are supported in code). **A conversation cannot change model mid-thread** — checkpoints store provider-native message blocks that no other provider can replay; see `build_chat_model` in `app/chat_service.py`. Python ≥3.13, managed with `uv`.
+ChromaDB for vectors; Cohere for embeddings/rerank; chat model is provider-configurable (DeepSeek V4 Flash through the first-party API by default, with a rescue-only in-app fallback to Gemini 2.5 Flash when a Gemini key is set; Anthropic, OpenAI, and OpenRouter-compatible models are supported in code but not user-selectable). **A conversation cannot change model mid-thread** — checkpoints store provider-native message blocks that no other provider can replay; see `build_chat_model` in `app/chat_service.py`. Python ≥3.13, managed with `uv`.
 
 ## Key URLs
 
@@ -62,7 +62,7 @@ Runtime guidance the agent follows is in `data/kb/AGENTS.md` (injected into the
 
 ## Sources & config
 
-`data/scraping_scripts/source_registry.py` is the **single source of truth** for sources (`SOURCE_CONFIGS`, key groupings, UI labels, defaults); `app/config.py` re-exports them and the frontend derives the picker from it (via `/api/tools`). Docs sources ingest via the GitHub API or `llms.txt`; course sources are Notion exports. To add a source: add it to the registry (+ the relevant grouping tuples), then run the matching workflow — no separate UI edit needed. Models live in `config.AVAILABLE_MODELS` (default `deepseek:deepseek-v4-flash`; Claude Haiku 4.5 is also selectable; OpenAI is supported in code). **`GEMINI_FALLBACK_MODEL_NAME` (`google-genai:gemini-2.5-flash`) is deliberately NOT in `AVAILABLE_MODELS`**: it serves real traffic as the DeepSeek rescue path but must never be user-selectable, because pre-Gemini-3 models cannot combine Gemini's built-in web tools with our two custom tools (that needs Gemini 3+ "tool context circulation"), so selecting it would be one web-search toggle away from a 400. The fallback is safe only because it never receives web tools. `build_chat_model` in `app/chat_service.py` accepts `deepseek:` for the first-party API, `openrouter:` for compatible experiment models, and `ollama:` for local SLM experiments, with pricing in `app/telemetry.MODEL_PRICING`; for the default DeepSeek model it also wires an in-app fallback to `google-genai:gemini-2.5-flash` whenever a Gemini key is configured (and substitutes it outright if `DEEPSEEK_API_KEY` is missing).
+`data/scraping_scripts/source_registry.py` is the **single source of truth** for sources (`SOURCE_CONFIGS`, key groupings, UI labels, defaults); `app/config.py` re-exports them and the frontend derives the picker from it (via `/api/tools`). Docs sources ingest via the GitHub API or `llms.txt`; course sources are Notion exports. To add a source: add it to the registry (+ the relevant grouping tuples), then run the matching workflow — no separate UI edit needed. Models live in `config.AVAILABLE_MODELS` (currently only the default `deepseek:deepseek-v4-flash`; Anthropic Claude Haiku 4.5 and OpenAI remain supported in code but are not selectable). **`GEMINI_FALLBACK_MODEL_NAME` (`google-genai:gemini-2.5-flash`) is deliberately NOT in `AVAILABLE_MODELS`**: it serves real traffic as the DeepSeek rescue path but must never be user-selectable, because pre-Gemini-3 models cannot combine Gemini's built-in web tools with our two custom tools (that needs Gemini 3+ "tool context circulation"), so selecting it would be one web-search toggle away from a 400. The fallback is safe only because it never receives web tools. `build_chat_model` in `app/chat_service.py` accepts `deepseek:` for the first-party API, `openrouter:` for compatible experiment models, and `ollama:` for local SLM experiments, with pricing in `app/telemetry.MODEL_PRICING`; for the default DeepSeek model it also wires an in-app fallback to `google-genai:gemini-2.5-flash` whenever a Gemini key is configured (and substitutes it outright if `DEEPSEEK_API_KEY` is missing).
 
 Memory policy is deliberately owned by `app/memory_presets.py`, not `app/config.py`. `PRODUCTION_MEMORY_PRESET` is the single primary-policy switch, `PRODUCTION_FALLBACK_MEMORY_PRESET` covers providers that cannot use it safely, and `PRODUCTION_LONG_CONTEXT_PROVIDERS` defines the model-aware allowlist. Keep historical presets immutable because saved eval runs refer to them by name.
 

app/config.py

@@ -99,7 +99,6 @@ AVAILABLE_MODELS: tuple[dict[str, str], ...] = (
         "id": DEEPSEEK_DIRECT_MODEL_NAME,
         "label": "DeepSeek V4 Flash",
     },
-    {"id": "anthropic:claude-haiku-4-5", "label": "Claude Haiku 4.5"},
 )
 
 

tests/test_api.py

@@ -208,16 +208,22 @@ class ApiTestCase(unittest.TestCase):
             ],
             "sourceKeys": ["langchain", "transformers"],
             "enabledTools": ["web_search", "not_a_real_tool"],
-            # A selectable model that offers web_search; gemini-2.5-flash is
-            # fallback-only and now 422s here (see test_config.py).
+            # The toggle passthrough below needs a model that offers web_search,
+            # and no selectable model does today (DeepSeek has no web toggles;
+            # gemini-2.5-flash is fallback-only). Re-admit Claude Haiku for this
+            # request only.
             "model": "anthropic:claude-haiku-4-5",
             "threadId": "thread_0",
         }
+        selectable_with_web_search = (
+            {"id": "anthropic:claude-haiku-4-5", "label": "Claude Haiku 4.5"},
+        )
 
-        with patch("app.api.stream_chat", fake_stream_chat):
-            with TestClient(app) as client:
-                with client.stream("POST", "/api/chat", json=payload) as response:
-                    body = "".join(response.iter_text())
+        with patch("app.api.AVAILABLE_MODELS", selectable_with_web_search):
+            with patch("app.api.stream_chat", fake_stream_chat):
+                with TestClient(app) as client:
+                    with client.stream("POST", "/api/chat", json=payload) as response:
+                        body = "".join(response.iter_text())
 
         self.assertEqual(response.status_code, 200)
         self.assertEqual(
@@ -1100,14 +1106,28 @@ class ApiTestCase(unittest.TestCase):
             )
         self.assertEqual(raised.exception.status_code, 422)
 
-        with self.assertRaises(HTTPException) as incompatible:
+        # Claude Haiku was removed from AVAILABLE_MODELS, so the model check
+        # rejects it before the preset check gets a say.
+        with self.assertRaises(HTTPException) as unselectable:
             build_chat_request(
-                ApiChatRequest(
-                    query="What is RAG?",
-                    model="anthropic:claude-haiku-4-5",
-                    memoryPreset="prod_v2",
-                )
+                ApiChatRequest(query="What is RAG?", model="anthropic:claude-haiku-4-5")
             )
+        self.assertEqual(unselectable.exception.status_code, 422)
+        self.assertEqual(unselectable.exception.detail, "Unknown model")
+
+        # The preset-provider incompatibility 422 needs a selectable model
+        # outside the long-context allowlist; none exists today, so re-admit
+        # Haiku for this request only.
+        haiku = ({"id": "anthropic:claude-haiku-4-5", "label": "Claude Haiku 4.5"},)
+        with patch("app.api.AVAILABLE_MODELS", haiku):
+            with self.assertRaises(HTTPException) as incompatible:
+                build_chat_request(
+                    ApiChatRequest(
+                        query="What is RAG?",
+                        model="anthropic:claude-haiku-4-5",
+                        memoryPreset="prod_v2",
+                    )
+                )
         self.assertEqual(incompatible.exception.status_code, 422)
         self.assertIn("does not support provider", incompatible.exception.detail)
 

tests/test_config.py

@@ -30,10 +30,7 @@ def test_gemini_fallback_model_is_deliberately_not_selectable() -> None:
     selectable = [model["id"] for model in config.AVAILABLE_MODELS]
 
     assert config.GEMINI_FALLBACK_MODEL_NAME not in selectable
-    assert selectable == [
-        "deepseek:deepseek-v4-flash",
-        "anthropic:claude-haiku-4-5",
-    ]
+    assert selectable == ["deepseek:deepseek-v4-flash"]
 
 
 def _patched_bundle(tmp_path: Path) -> ExitStack: