Model name normalization and provider prefix handling

Fixes on record

Merged to main. A stamp counts the issues a fix is linked to.

• #7187 fix(model): salvage model/provider resolution PRs #6860 #6844 #6863 #6943

  merged to main

  files: agent/auxiliary_client.py, agent/error_classifier.py, cli.py, hermes_cli/main.py, hermes_cli/model_normalize.py, hermes_cli/model_switch.py, run_agent.py, tests/agent/test_auxiliary_named_custom_providers.py, tests/agent/test_error_classifier.py, tests/hermes_cli/test_codex_models.py, tests/hermes_cli/test_custom_provider_model_switch.py, tests/hermes_cli/test_model_normalize.py, tests/hermes_cli/test_opencode_go_in_model_list.py, tests/run_agent/test_fallback_model.py, tests/run_agent/test_run_agent.py, tests/run_agent/test_switch_model_context.py

• #7935 fix(custom-providers): propagate model field from config so API receives the correct model name

  merged to main

  files: cli.py, hermes_cli/config.py, hermes_cli/runtime_provider.py, tests/hermes_cli/test_runtime_provider_resolution.py

  Show the diff

  • cli.py

    @@ -2710,6 +2710,15 @@ def _ensure_runtime_credentials(self) -> bool:
             self.api_key = [redacted:secret-assignment]
             self.base_url = base_url
     
    +        # When a custom_provider entry carries an explicit `model` field,
    +        # use it as the effective model name.  Without this, running
    +        # `hermes chat --model <provider-name>` sends the provider name
    +        # (e.g. "my-provider") as the model string to the API instead of
    +        # the configured model (e.g. "qwen3.6-plus"), causing 400 errors.
    +        runtime_model = runtime.get("model")
    +        if runtime_model and isinstance(runtime_model, str):
    +            self.model = runtime_model
    +
             # Normalize model for the resolved provider (e.g. swap non-Codex
             # models when provider is openai-codex).  Fixes #651.
             model_changed = self._normalize_model_for_provider(resolved_provider)

  • hermes_cli/config.py

    @@ -1497,7 +1497,7 @@ def check_config_version() -> Tuple[int, int]:
     
     # Valid fields inside a custom_providers list entry
     _VALID_CUSTOM_PROVIDER_FIELDS = {
    -    "name", "base_url", "api_key", "api_mode", "models",
    +    "name", "base_url", "api_key", "api_mode", "model", "models",
         "context_length", "rate_limit_delay",
     }
     

  • hermes_cli/runtime_provider.py

    @@ -304,6 +304,9 @@ def _get_named_custom_provider(requested_provider: str) -> Optional[Dict[str, An
             api_mode = _parse_api_mode(entry.get("api_mode"))
             if api_mode:
                 result["api_mode"] = api_mode
    +        model_name = str(entry.get("model", "") or "").strip()
    +        if model_name:
    +            result["model"] = model_name
             return result
     
         return None
    @@ -329,6 +332,11 @@ def _resolve_named_custom_runtime(
         # Check if a credential pool exists for this custom endpoint
         pool_result = _try_resolve_from_custom_pool(base_url, "custom", custom_provider.get("api_mode"))
         if pool_result:
    +        # Propagate the model name even when using pooled credentials —
    +        # the pool doesn't know about the custom_providers model field.
    +        model_name = custom_provider.get("model")
    +        if model_name:
    +            pool_result["model"] = model_name
             return pool_result
     
         api_key_candidates = [
    @@ -339,7 +347,7 @@ def _resolve_named_custom_runtime(
         ]
         api_key = [redacted:secret-assignment] for candidate in api_key_candidates if has_usable_secret(candidate)), "")
     
    -    return {
    +    result = {
             "provider": "custom",
             "api_mode": custom_provider.get("api_mode")
             or _detect_api_mode_for_url(base_url)
    @@ -348,6 +356,11 @@ def _resolve_named_custom_runtime(
             "api_key": api_key or "no-key-required",
             "source": f"custom_provider:{custom_provider.get('name', requested_provider)}",
         }
    +    # Propagate the model name so callers can override self.model when the
    +    # provider name differs from the actual model string the API expects.
    +    if custom_provider.get("model"):
    +        result["model"] = custom_provider["model"]
    +    return result
     
     
     def _resolve_openrouter_runtime(

  • tests/hermes_cli/test_runtime_provider_resolution.py

    @@ -1214,3 +1214,115 @@ def test_openrouter_provider_not_affected_by_custom_fix(monkeypatch):
     
         resolved = rp.resolve_runtime_provider(requested="openrouter")
         assert resolved["provider"] == "openrouter"
    +
    +
    +# ------------------------------------------------------------------
    +# fix #7828 — custom_providers model field must propagate to runtime
    +# ------------------------------------------------------------------
    +
    +
    +def test_get_named_custom_provider_includes_model(monkeypatch):
    +    """_get_named_custom_provider should include the model field from config."""
    +    monkeypatch.setattr(rp, "load_config", lambda: {
    +        "custom_providers": [{
    +            "name": "my-dashscope",
    +            "base_url": "https://dashscope.aliyuncs.com/compatible-mode/v1",
    +            "api_key": "test-key",
    +            "api_mode": "chat_completions",
    +            "model": "qwen3.6-plus",
    +        }],
    +    })
    +
    +    result = rp._get_named_custom_provider("my-dashscope")
    +    assert result is not None
    +    assert result["model"] == "qwen3.6-plus"
    +
    +
    +def test_get_named_custom_provider_excludes_empty_model(monkeypatch):
    +    """Empty or whitespace-only model field should not appear in result."""
    +    for model_val in ["", "   ", None]:
    +        entry = {
    +            "name": "test-ep",
    +            "base_url": "https://example.com/v1",
    +            "api_key": "key",
    +        }
    +        if model_val is not None:
    +            entry["model"] = model_val
    +
    +        monkeypatch.setattr(rp, "load_config", lambda e=entry: {
    +            "custom_providers": [e],
    +        })
    +
    +        result = rp._get_named_custom_provider("test-ep")
    +        assert result is not None
    +        assert "model" not in result, (
    +            f"model field {model_val!r} should not be included in result"
    +        )
    +
    +
    +def test_named_custom_runtime_propagates_model_direct_path(monkeypatch):
    +    """Model should propagate through the direct (non-pool) resolution path."""
    +    monkeypatch.setattr(rp, "resolve_provider", lambda *a, **k: "my-server")
    +    monkeypatch.setattr(
    +        rp, "_get_named_custom_provider",
    +        lambda p: {
    +            "name": "my-server",
    +            "base_url": "http://localhost:8000/v1",
    +            "api_key": "test-key",
    +            "model": "qwen3.6-plus",
    +        },
    +    )
    +    # Ensure pool doesn't intercept
    +    monkeypatch.setattr(rp, "_try_resolve_from_custom_pool", lambda *a, **k: None)
    +
    +    resolved = rp.resolve_runtime_provider(requested="my-server")
    +    assert resolved["model"] == "qwen3.6-plus"
    +    assert resolved["provider"] == "custom"
    +
    +
    +def test_named_custom_runtime_propagates_model_pool_path(monkeypatch):
    +    """Model should propagate even when credential pool handles credentials."""
    +    monkeypatch.setattr(rp, "resolve_provider", lambda *a, **k: "my-server")
    +    monkeypatch.setattr(
    +        rp, "_get_named_custom_provider",
    +        lambda p: {
    +            "name": "my-server",
    +            "base_url": "http://localhost:8000/v1",
    +            "api_key": "test-key",
    +            "model": "qwen3.6-plus",
    +        },
    +    )
    +    # Pool returns a result (intercepting the normal path)
    +    monkeypatch.setattr(
    +        rp, "_try_resolve_from_custom_pool",
    +        lambda *a, **k: {
    +            "provider": "custom",
    +            "api_mode": "chat_completions",
    +            "base_url": "http://localhost:8000/v1",
    +            "api_key": "pool-key",
    +            "source": "pool:custom:my-server",
    +        },
    +    )
    +
    +    resolved = rp.resolve_runtime_provider(requested="my-server")
    +    assert resolved["model"] == "qwen3.6-plus", (
    +        "model must be injected into pool result"
    +    )
    +    assert resolved["api_key"] == "pool-key", "pool credentials should be used"
    +
    +
    +def test_named_custom_runtime_no_model_when_absent(monkeypatch):
    +    """When custom_providers entry has no model field, runtime should not either."""
    +    monkeypatch.setattr(rp, "resolve_provider", lambda *a, **k: "my-server")
    +    monkeypatch.setattr(
    +        rp, "_get_named_custom_provider",
    +        lambda p: {
    +            "name": "my-server",
    +            "base_url": "http://localhost:8000/v1",
    +            "api_key": "test-key",
    +        },
    +    )
    +    monkeypatch.setattr(rp, "_try_resolve_from_custom_pool", lambda *a, **k: None)
    +
    +    resolved = rp.resolve_runtime_provider(requested="my-server")
    +    assert "model" not in resolved

• #54093 fix(config): canonicalize model.name/model.model to model.default (#34500)

  merged to main

  files: gateway/run.py, hermes_cli/config.py, infographic/model-name-canon/infographic.png, tests/cli/test_cli_init.py

• #71144 fix(model_switch): don't send a picker prefix as the custom provider model id

  merged to main

  files: contributors/emails/[redacted:email], hermes_cli/model_switch.py, tests/hermes_cli/test_model_switch_custom_providers.py

• #115756 fix(cli): `-m openai-codex:<model>` and other provider:model switches no longer send the prefix to the API (#64787, salvage #64788)

  merged to main

  files: contributors/emails/[redacted:email], hermes_cli/model_normalize.py, tests/hermes_cli/test_model_normalize.py