feat(llm): make tier-C/tier-B model + OpenRouter URL configurable from .env

LLM_MODEL_TIER_C, LLM_MODEL_TIER_B e OPENROUTER_BASE_URL ora override-abili
via env. Default invariati (back-compat). LLMClient accetta i tre valori
come kwargs opzionali; run_phase1 li propaga da Settings.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

.env.example

@@ -8,6 +8,11 @@ CERBERO_BOT_TAG=swarm-poc-phase1
 OPENROUTER_API_KEY=
 ANTHROPIC_API_KEY=
 
+# LLM models (override Phase 1 defaults if needed)
+LLM_MODEL_TIER_C=qwen/qwen-2.5-72b-instruct
+LLM_MODEL_TIER_B=claude-sonnet-4-6
+OPENROUTER_BASE_URL=https://openrouter.ai/api/v1
+
 # Run config
 RUN_NAME=phase1-spike-001
 DATA_DIR=./data

scripts/run_phase1.py

@@ -48,6 +48,9 @@ def main() -> None:
             settings.anthropic_api_key.get_secret_value()
             if settings.anthropic_api_key else None
         ),
+        model_tier_c=settings.llm_model_tier_c,
+        model_tier_b=settings.llm_model_tier_b,
+        openrouter_base_url=settings.openrouter_base_url,
     )
 
     cfg = RunConfig(

src/multi_swarm/config.py

@@ -26,6 +26,10 @@ class Settings(BaseSettings):
     openrouter_api_key: SecretStr
     anthropic_api_key: SecretStr | None = None
 
+    llm_model_tier_c: str = "qwen/qwen-2.5-72b-instruct"
+    llm_model_tier_b: str = "claude-sonnet-4-6"
+    openrouter_base_url: str = "https://openrouter.ai/api/v1"
+
     run_name: str = "phase1-spike-001"
     data_dir: Path = Field(default=Path("./data"))
     series_dir: Path = Field(default=Path("./series"))

src/multi_swarm/llm/client.py

@@ -45,8 +45,14 @@ class LLMClient:
         self,
         openrouter_api_key: str,
         anthropic_api_key: str | None = None,
+        model_tier_c: str = MODEL_TIER_C,
+        model_tier_b: str = MODEL_TIER_B,
+        openrouter_base_url: str = OPENROUTER_BASE_URL,
     ) -> None:
-        self._openrouter = OpenAI(api_key=openrouter_api_key, base_url=OPENROUTER_BASE_URL)
+        self.model_tier_c = model_tier_c
+        self.model_tier_b = model_tier_b
+        self.openrouter_base_url = openrouter_base_url
+        self._openrouter = OpenAI(api_key=openrouter_api_key, base_url=openrouter_base_url)
         self._anthropic = Anthropic(api_key=anthropic_api_key) if anthropic_api_key else None
 
     @retry(
@@ -64,7 +70,7 @@ class LLMClient:
     ) -> CompletionResult:
         if genome.model_tier == ModelTier.C:
             resp = self._openrouter.chat.completions.create(
-                model=MODEL_TIER_C,
+                model=self.model_tier_c,
                 messages=[
                     {"role": "system", "content": system},
                     {"role": "user", "content": user},
@@ -80,14 +86,14 @@ class LLMClient:
                 input_tokens=usage.prompt_tokens,
                 output_tokens=usage.completion_tokens,
                 tier=ModelTier.C,
-                model=MODEL_TIER_C,
+                model=self.model_tier_c,
             )
 
         if self._anthropic is None:
             raise RuntimeError("ANTHROPIC_API_KEY required for tier B genomes")
 
         msg = self._anthropic.messages.create(
-            model=MODEL_TIER_B,
+            model=self.model_tier_b,
             system=system,
             messages=[{"role": "user", "content": user}],
             temperature=genome.temperature,
@@ -100,5 +106,5 @@ class LLMClient:
             input_tokens=msg.usage.input_tokens,
             output_tokens=msg.usage.output_tokens,
             tier=ModelTier.B,
-            model=MODEL_TIER_B,
+            model=self.model_tier_b,
         )

tests/unit/test_config.py

@@ -40,3 +40,31 @@ def test_settings_requires_tokens(monkeypatch: pytest.MonkeyPatch) -> None:
         # Disable .env loading to keep the test deterministic regardless of
         # whether a developer's local .env exists and is populated.
         Settings(_env_file=None)  # type: ignore[call-arg]
+
+
+def test_settings_loads_llm_model_overrides(monkeypatch: pytest.MonkeyPatch) -> None:
+    monkeypatch.setenv("CERBERO_TESTNET_TOKEN", "tok-test")
+    monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
+    monkeypatch.setenv("LLM_MODEL_TIER_C", "deepseek/deepseek-chat")
+    monkeypatch.setenv("LLM_MODEL_TIER_B", "claude-opus-4-7")
+    monkeypatch.setenv("OPENROUTER_BASE_URL", "https://example.com/api/v1")
+
+    s = Settings(_env_file=None)  # type: ignore[call-arg]
+
+    assert s.llm_model_tier_c == "deepseek/deepseek-chat"
+    assert s.llm_model_tier_b == "claude-opus-4-7"
+    assert s.openrouter_base_url == "https://example.com/api/v1"
+
+
+def test_settings_llm_model_defaults(monkeypatch: pytest.MonkeyPatch) -> None:
+    monkeypatch.setenv("CERBERO_TESTNET_TOKEN", "tok-test")
+    monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
+    monkeypatch.delenv("LLM_MODEL_TIER_C", raising=False)
+    monkeypatch.delenv("LLM_MODEL_TIER_B", raising=False)
+    monkeypatch.delenv("OPENROUTER_BASE_URL", raising=False)
+
+    s = Settings(_env_file=None)  # type: ignore[call-arg]
+
+    assert s.llm_model_tier_c == "qwen/qwen-2.5-72b-instruct"
+    assert s.llm_model_tier_b == "claude-sonnet-4-6"
+    assert s.openrouter_base_url == "https://openrouter.ai/api/v1"

tests/unit/test_llm_client.py

@@ -75,6 +75,52 @@ def test_completion_retries_on_connection_error(mocker):
     assert fake_openai.chat.completions.create.call_count == 3
 
 
+def test_completion_uses_custom_model_tier_c(mocker):
+    fake_openai = mocker.MagicMock()
+    fake_response = mocker.MagicMock()
+    fake_response.choices = [
+        mocker.MagicMock(message=mocker.MagicMock(content="(strategy ...)"))
+    ]
+    fake_response.usage = mocker.MagicMock(prompt_tokens=10, completion_tokens=20)
+    fake_openai.chat.completions.create.return_value = fake_response
+    mocker.patch("multi_swarm.llm.client.OpenAI", return_value=fake_openai)
+
+    client = LLMClient(
+        openrouter_api_key="or-x",
+        anthropic_api_key=None,
+        model_tier_c="deepseek/deepseek-chat",
+    )
+    g = make_genome(ModelTier.C)
+    out = client.complete(g, system="sys", user="usr")
+
+    fake_openai.chat.completions.create.assert_called_once()
+    call_kwargs = fake_openai.chat.completions.create.call_args.kwargs
+    assert call_kwargs["model"] == "deepseek/deepseek-chat"
+    assert out.model == "deepseek/deepseek-chat"
+
+
+def test_completion_uses_custom_model_tier_b(mocker):
+    fake_anthropic = mocker.MagicMock()
+    fake_msg = mocker.MagicMock()
+    fake_msg.content = [mocker.MagicMock(text="(strategy ...)")]
+    fake_msg.usage = mocker.MagicMock(input_tokens=10, output_tokens=20)
+    fake_anthropic.messages.create.return_value = fake_msg
+    mocker.patch("multi_swarm.llm.client.Anthropic", return_value=fake_anthropic)
+
+    client = LLMClient(
+        openrouter_api_key="or-x",
+        anthropic_api_key="an-x",
+        model_tier_b="claude-opus-4-7",
+    )
+    g = make_genome(ModelTier.B)
+    out = client.complete(g, system="sys", user="usr")
+
+    fake_anthropic.messages.create.assert_called_once()
+    call_kwargs = fake_anthropic.messages.create.call_args.kwargs
+    assert call_kwargs["model"] == "claude-opus-4-7"
+    assert out.model == "claude-opus-4-7"
+
+
 @pytest.mark.slow
 def test_completion_succeeds_after_one_retry(mocker):
     """Dopo 1 fallimento transient, il retry riesce al 2 tentativo."""