[loop-cycle-52] fix: remove all qwen3.5 references (#182) (#190)

.env.example

@@ -14,8 +14,8 @@
 # In production (docker-compose.prod.yml), this is set to http://ollama:11434 automatically.
 # OLLAMA_URL=http://localhost:11434
 
-# LLM model to use via Ollama (default: qwen3.5:latest)
+# LLM model to use via Ollama (default: qwen3:30b)
-# OLLAMA_MODEL=qwen3.5:latest
+# OLLAMA_MODEL=qwen3:30b
 
 # Ollama context window size (default: 4096 tokens)
 # Set higher for more context, lower to save RAM. 0 = model default.

README.md

@@ -18,15 +18,15 @@ make install              # create venv + install deps
 cp .env.example .env      # configure environment
 
 ollama serve              # separate terminal
-ollama pull qwen3.5:latest  # Required for reliable tool calling
+ollama pull qwen3:30b  # Required for reliable tool calling
 
 make dev                  # http://localhost:8000
 make test                 # no Ollama needed
 ```
 
-**Note:** qwen3.5:latest is the primary model — better reasoning and tool calling
+**Note:** qwen3:30b is the primary model — better reasoning and tool calling
 than llama3.1:8b-instruct while still running locally on modest hardware.
-Fallback: llama3.1:8b-instruct if qwen3.5:latest is not available.
+Fallback: llama3.1:8b-instruct if qwen3:30b is not available.
 llama3.2 (3B) was found to hallucinate tool output consistently in testing.
 
 ---
@@ -79,7 +79,7 @@ cp .env.example .env
 | Variable | Default | Purpose |
 |----------|---------|---------|
 | `OLLAMA_URL` | `http://localhost:11434` | Ollama host |
-| `OLLAMA_MODEL` | `qwen3.5:latest` | Primary model for reasoning and tool calling. Fallback: `llama3.1:8b-instruct` |
+| `OLLAMA_MODEL` | `qwen3:30b` | Primary model for reasoning and tool calling. Fallback: `llama3.1:8b-instruct` |
 | `DEBUG` | `false` | Enable `/docs` and `/redoc` |
 | `TIMMY_MODEL_BACKEND` | `ollama` | `ollama` \| `airllm` \| `auto` |
 | `AIRLLM_MODEL_SIZE` | `70b` | `8b` \| `70b` \| `405b` |

config/agents.yaml

@@ -20,7 +20,7 @@
 # ── Defaults ────────────────────────────────────────────────────────────────
 
 defaults:
-  model: qwen3.5:latest
+  model: qwen3:30b
   prompt_tier: lite
   max_history: 10
   tools: []

config/providers.yaml

@@ -25,7 +25,7 @@ providers:
     url: "http://localhost:11434"
     models:
       # Text + Tools models
-      - name: qwen3.5:latest
+      - name: qwen3:30b
         default: true
         context_window: 128000
         # Note: actual context is capped by OLLAMA_NUM_CTX (default 4096) to save RAM
@@ -114,13 +114,12 @@ fallback_chains:
   # Tool-calling models (for function calling)
   tools:
     - llama3.1:8b-instruct # Best tool use
-    - qwen3.5:latest       # Qwen 3.5 — strong tool use
     - qwen2.5:7b           # Reliable tools
     - llama3.2:3b          # Small but capable
   
   # General text generation (any model)
   text:
-    - qwen3.5:latest
+    - qwen3:30b
     - llama3.1:8b-instruct
     - qwen2.5:14b
     - deepseek-r1:1.5b

docs/SOVEREIGN_AGI_RESEARCH.md

@@ -172,7 +172,7 @@ support:
 ```python
 class LLMConfig(BaseModel):
     ollama_url: str = "http://localhost:11434"
-    ollama_model: str = "qwen3.5:latest"
+    ollama_model: str = "qwen3:30b"
     # ... all LLM settings
 
 class MemoryConfig(BaseModel):

src/config.py

@@ -16,11 +16,11 @@ class Settings(BaseSettings):
     ollama_url: str = "http://localhost:11434"
 
     # LLM model passed to Agno/Ollama — override with OLLAMA_MODEL
-    # qwen3.5:latest is the primary model — better reasoning and tool calling
+    # qwen3:30b is the primary model — better reasoning and tool calling
     # than llama3.1:8b-instruct while still running locally on modest hardware.
-    # Fallback: llama3.1:8b-instruct if qwen3.5:latest not available.
+    # Fallback: llama3.1:8b-instruct if qwen3:30b not available.
     # llama3.2 (3B) hallucinated tool output consistently in testing.
-    ollama_model: str = "qwen3.5:latest"
+    ollama_model: str = "qwen3:30b"
 
     # Context window size for Ollama inference — override with OLLAMA_NUM_CTX
     # qwen3:30b with default context eats 45GB on a 39GB Mac.
@@ -28,12 +28,11 @@ class Settings(BaseSettings):
     ollama_num_ctx: int = 4096
 
     # Fallback model chains — override with FALLBACK_MODELS / VISION_FALLBACK_MODELS
-    # as comma-separated strings, e.g. FALLBACK_MODELS="qwen3.5:latest,llama3.1"
+    # as comma-separated strings, e.g. FALLBACK_MODELS="qwen3:30b,llama3.1"
     # Or edit config/providers.yaml → fallback_chains for the canonical source.
     fallback_models: list[str] = [
         "llama3.1:8b-instruct",
         "llama3.1",
-        "qwen3.5:latest",
         "qwen2.5:14b",
         "qwen2.5:7b",
         "llama3.2:3b",

src/infrastructure/models/multimodal.py

@@ -93,18 +93,6 @@ KNOWN_MODEL_CAPABILITIES: dict[str, set[ModelCapability]] = {
         ModelCapability.VISION,
     },
     # Qwen series
-    "qwen3.5": {
-        ModelCapability.TEXT,
-        ModelCapability.TOOLS,
-        ModelCapability.JSON,
-        ModelCapability.STREAMING,
-    },
-    "qwen3.5:latest": {
-        ModelCapability.TEXT,
-        ModelCapability.TOOLS,
-        ModelCapability.JSON,
-        ModelCapability.STREAMING,
-    },
     "qwen2.5": {
         ModelCapability.TEXT,
         ModelCapability.TOOLS,
@@ -271,9 +259,8 @@ DEFAULT_FALLBACK_CHAINS: dict[ModelCapability, list[str]] = {
     ],
     ModelCapability.TOOLS: [
         "llama3.1:8b-instruct",  # Best tool use
-        "qwen3.5:latest",  # Qwen 3.5 — strong tool use
-        "llama3.2:3b",  # Smaller but capable
         "qwen2.5:7b",  # Reliable fallback
+        "llama3.2:3b",  # Smaller but capable
     ],
     ModelCapability.AUDIO: [
         # Audio models are less common in Ollama

src/timmy/tools.py

@@ -283,12 +283,12 @@ def create_aider_tool(base_path: Path):
         def __init__(self, base_dir: Path):
             self.base_dir = base_dir
 
-        def run_aider(self, prompt: str, model: str = "qwen3.5:latest") -> str:
+        def run_aider(self, prompt: str, model: str = "qwen3:30b") -> str:
             """Run Aider to generate code changes.
 
             Args:
                 prompt: What you want Aider to do (e.g., "add a fibonacci function")
-                model: Ollama model to use (default: qwen3.5:latest)
+                model: Ollama model to use (default: qwen3:30b)
 
             Returns:
                 Aider's response with the code changes made
@@ -788,7 +788,7 @@ def _ai_tool_catalog() -> dict:
         },
         "aider": {
             "name": "Aider AI Assistant",
-            "description": "Local AI coding assistant using Ollama (qwen3.5:latest or deepseek-coder)",
+            "description": "Local AI coding assistant using Ollama (qwen3:30b or deepseek-coder)",
             "available_in": ["forge", "orchestrator"],
         },
     }

src/timmy/tools_intro/__init__.py

@@ -63,7 +63,7 @@ def _get_ollama_model() -> str:
     1. /api/ps — models currently loaded in memory (most accurate)
     2. /api/tags — all installed models (fallback)
     Both use exact name match to avoid prefix collisions
-    (e.g. 'qwen3:30b' vs 'qwen3.5:latest').
+    (e.g. 'qwen3:8b' vs 'qwen3:30b').
     """
     from config import settings
 

tests/timmy/test_introspection.py

@@ -98,35 +98,35 @@ class TestGetOllamaModelExactMatch:
 
     @patch("timmy.tools_intro.httpx.get")
     def test_prefix_collision_returns_correct_model(self, mock_get):
-        """qwen3:30b configured — must NOT match qwen3.5:latest (prefix bug)."""
+        """qwen3:8b configured — must NOT match qwen3:30b (prefix bug)."""
         from timmy.tools_intro import _get_ollama_model
 
-        # /api/ps has both models loaded; configured is qwen3:30b
+        # /api/ps has both models loaded; configured is qwen3:8b
-        ps_resp = _mock_response({"models": [{"name": "qwen3.5:latest"}, {"name": "qwen3:30b"}]})
+        ps_resp = _mock_response({"models": [{"name": "qwen3:30b"}, {"name": "qwen3:8b"}]})
         mock_get.return_value = ps_resp
 
         with patch("config.settings") as mock_settings:
-            mock_settings.ollama_model = "qwen3:30b"
+            mock_settings.ollama_model = "qwen3:8b"
             mock_settings.ollama_url = "http://localhost:11434"
             result = _get_ollama_model()
 
-        assert result == "qwen3:30b", f"Got '{result}' — prefix collision bug!"
+        assert result == "qwen3:8b", f"Got '{result}' — prefix collision bug!"
 
     @patch("timmy.tools_intro.httpx.get")
     def test_configured_model_not_running_returns_actual(self, mock_get):
         """If configured model isn't loaded, report what IS running."""
         from timmy.tools_intro import _get_ollama_model
 
-        ps_resp = _mock_response({"models": [{"name": "qwen3.5:latest"}]})
+        ps_resp = _mock_response({"models": [{"name": "qwen3:30b"}]})
         mock_get.return_value = ps_resp
 
         with patch("config.settings") as mock_settings:
-            mock_settings.ollama_model = "qwen3:30b"
+            mock_settings.ollama_model = "qwen3:8b"
             mock_settings.ollama_url = "http://localhost:11434"
             result = _get_ollama_model()
 
         # Should report actual running model, not configured one
-        assert result == "qwen3.5:latest"
+        assert result == "qwen3:30b"
 
     @patch("timmy.tools_intro.httpx.get")
     def test_latest_suffix_match(self, mock_get):