Add local LLM providers (Ollama, vLLM) with zero-cost tracking

Author: johnteee

docs/acceptance.md

@@ -519,8 +519,4 @@ track p50/p95 latency across eval suites and detect regressions.*
 
 ## Next User Stories
 
-The following stories remain unimplemented.
-
-| ID | Story | Priority |
-|---|---|---|
-| AC-NEW-20 | Local model (Ollama/vLLM) with full governance | P2 |
+All currently tracked stories are implemented.

teaagent/llm/_config.py

@@ -43,6 +43,22 @@
         base_url='https://openrouter.ai/api/v1',
         base_url_env='OPENROUTER_BASE_URL',
     ),
+    'ollama': ProviderConfig(
+        name='ollama',
+        api_key_env='OLLAMA_API_KEY',
+        default_model='llama3.2',
+        base_url='http://localhost:11434/v1',
+        api_key='ollama',
+        base_url_env='OLLAMA_BASE_URL',
+    ),
+    'vllm': ProviderConfig(
+        name='vllm',
+        api_key_env='VLLM_API_KEY',
+        default_model='meta-llama/Llama-3.1-8B-Instruct',
+        base_url='http://localhost:8000/v1',
+        api_key='vllm',
+        base_url_env='VLLM_BASE_URL',
+    ),
     'opencodezen-go': ProviderConfig(
         name='opencodezen-go',
         api_key_env='OPENCODEZEN_API_KEY',
@@ -100,15 +116,19 @@ def check_llm_configuration(provider: str) -> tuple[bool, str]:
     'gpt': 0.00015,
     'gemini': 0.000075,
     'openrouter': 0.0005,
+    'ollama': 0.0,
     'opencodezen-go': 0.0005,
+    'vllm': 0.0,
 }
 
 PROVIDER_COST_PER_1K_OUTPUT: dict[str, float] = {
     'claude': 0.015,
     'gpt': 0.0006,
     'gemini': 0.0003,
     'openrouter': 0.002,
+    'ollama': 0.0,
     'opencodezen-go': 0.002,
+    'vllm': 0.0,
 }
 
 

teaagent/tui/__init__.py

@@ -32,7 +32,7 @@ def default_adapter_factory(provider: str, model: Optional[str]) -> LLMAdapter:
 HELP_TEXT = """Commands:
   help                      Show this help.
   doctor                    Check GraphQLite runtime.
-  provider <name>           Set model provider: claude, gpt, gemini, openrouter, opencodezen-go.
+  provider <name>           Set model provider: claude, gpt, gemini, openrouter, ollama, vllm, opencodezen-go.
   model <name|default>      Set or clear model override.
   route-model <on|off>      Enable or disable task-based model routing.
   route <task>              Preview model route for a task.

tests/test_llm.py

@@ -36,7 +36,37 @@ class LLMAdapterTests(unittest.TestCase):
     def test_available_providers_include_requested_adapters(self) -> None:
         self.assertEqual(
             available_providers(),
-            ['claude', 'gemini', 'gpt', 'opencodezen-go', 'openrouter'],
+            [
+                'claude',
+                'gemini',
+                'gpt',
+                'ollama',
+                'opencodezen-go',
+                'openrouter',
+                'vllm',
+            ],
+        )
+
+    def test_ollama_is_openai_compatible_without_api_key(self) -> None:
+        transport = FakeTransport({'choices': [{'message': {'content': 'ok'}}]})
+        with patch.dict(os.environ, {}, clear=True):
+            adapter = create_llm_adapter('ollama', transport=transport)
+            response = adapter.complete(LLMRequest(messages=[LLMMessage('user', 'hi')]))
+
+        self.assertEqual(response.content, 'ok')
+        self.assertEqual(
+            transport.calls[0]['url'], 'http://localhost:11434/v1/chat/completions'
+        )
+
+    def test_vllm_is_openai_compatible_without_api_key(self) -> None:
+        transport = FakeTransport({'choices': [{'message': {'content': 'ok'}}]})
+        with patch.dict(os.environ, {}, clear=True):
+            adapter = create_llm_adapter('vllm', transport=transport)
+            response = adapter.complete(LLMRequest(messages=[LLMMessage('user', 'hi')]))
+
+        self.assertEqual(response.content, 'ok')
+        self.assertEqual(
+            transport.calls[0]['url'], 'http://localhost:8000/v1/chat/completions'
         )
 
     def test_gpt_adapter_uses_openai_chat_completions_shape(self) -> None: