fix(llm): make request timeout configurable

.env.example

@@ -29,6 +29,8 @@
 EVEROS_LLM__MODEL=openai/gpt-4.1-mini
 EVEROS_LLM__API_KEY=
 EVEROS_LLM__BASE_URL=https://openrouter.ai/api/v1
+# Per-request timeout in seconds (default 60):
+# EVEROS_LLM__TIMEOUT_SECONDS=60
 
 
 # ─── Multimodal LLM (independent from [llm]; vision/audio capable) ────

config.example.toml

@@ -24,6 +24,7 @@
 model    = "gpt-4.1-mini"
 api_key  = "sk-..."
 base_url = "https://api.openai.com/v1"
+timeout_seconds = 60.0
 
 # ── Embedding ─────────────────────────────────────────
 [embedding]

docs/configuration.md

@@ -106,6 +106,7 @@ everos init --root /data/everos
 | `model` | string | `"gpt-4.1-mini"` | No | LLM model identifier. |
 | `api_key` | string | — | **Yes** | API key for the LLM provider. |
 | `base_url` | string | — | No | Custom endpoint URL (OpenAI-compatible). |
+| `timeout_seconds` | float | `60.0` | No | Per-request timeout in seconds. Must be greater than zero. |
 
 ### `[multimodal]`
 

src/everos/component/llm/client.py

@@ -51,6 +51,7 @@ def get_llm_client() -> LLMClient:
             model=llm_cfg.model,
             api_key=api_key,
             base_url=llm_cfg.base_url,
+            timeout=llm_cfg.timeout_seconds,
         )
     )
     logger.info("llm_client_built", model=llm_cfg.model)

src/everos/component/llm/factory.py

@@ -42,4 +42,5 @@ def build_llm_provider(settings: LLMSettings) -> LLMClient:
         model=settings.model,
         api_key=settings.api_key.get_secret_value(),
         base_url=settings.base_url,
+        timeout=settings.timeout_seconds,
     )

src/everos/config/default.toml

@@ -52,11 +52,13 @@ cache_size_kb = 2048
 
 [llm]
 # Provider-agnostic OpenAI-protocol client config. Override via env:
-#   EVEROS_LLM__MODEL, EVEROS_LLM__API_KEY, EVEROS_LLM__BASE_URL
+#   EVEROS_LLM__MODEL, EVEROS_LLM__API_KEY, EVEROS_LLM__BASE_URL,
+#   EVEROS_LLM__TIMEOUT_SECONDS
 # Or set the field directly in this file (<root>/everos.toml).
 model = "openai/gpt-4.1-mini"
 api_key = ""
 base_url = "https://openrouter.ai/api/v1"
+timeout_seconds = 60.0
 
 [multimodal]
 # Independent LLM for multimodal parsing (everalgo-parser); must accept

src/everos/config/settings.py

@@ -127,11 +127,13 @@ class LLMSettings(BaseModel):
         EVEROS_LLM__MODEL
         EVEROS_LLM__API_KEY
         EVEROS_LLM__BASE_URL
+        EVEROS_LLM__TIMEOUT_SECONDS
     """
 
     model: str = "gpt-4.1-mini"
     api_key: SecretStr | None = None
     base_url: str | None = None
+    timeout_seconds: float = Field(default=60.0, gt=0)
 
 
 class MultimodalSettings(BaseModel):

src/everos/templates/env.template

@@ -29,6 +29,8 @@
 EVEROS_LLM__MODEL=openai/gpt-4.1-mini
 EVEROS_LLM__API_KEY=
 EVEROS_LLM__BASE_URL=https://openrouter.ai/api/v1
+# Per-request timeout in seconds (default 60):
+# EVEROS_LLM__TIMEOUT_SECONDS=60
 
 
 # ─── Multimodal LLM (independent from [llm]; vision/audio capable) ────

tests/unit/test_component/test_llm/test_client.py

@@ -5,6 +5,7 @@
 import importlib
 
 import pytest
+from everalgo.llm.config import LLMConfig
 from pydantic import SecretStr
 
 from everos.component.llm import LLMNotConfiguredError
@@ -23,13 +24,15 @@ def _patch_settings(
     *,
     api_key: str | None,
     base_url: str | None,
+    timeout_seconds: float = 60.0,
 ) -> None:
     """Stub the ``load_settings`` reference bound inside the client module."""
     cfg = Settings(
         llm=LLMSettings(
             model="gpt-4.1-mini",
             api_key=SecretStr(api_key) if api_key is not None else None,
             base_url=base_url,
+            timeout_seconds=timeout_seconds,
         )
     )
     monkeypatch.setattr(_client_mod, "load_settings", lambda: cfg)
@@ -62,3 +65,28 @@ def test_returns_singleton_when_configured(monkeypatch: pytest.MonkeyPatch) -> N
 
     assert first is sentinel
     assert first is second
+
+
+def test_passes_configured_timeout_to_everalgo_client(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    _reset_singleton(monkeypatch)
+    _patch_settings(
+        monkeypatch,
+        api_key="sk-test",
+        base_url="https://example.test",
+        timeout_seconds=135.0,
+    )
+    sentinel = object()
+    received_timeouts: list[float] = []
+
+    def fake_build_client(cfg: LLMConfig) -> object:
+        received_timeouts.append(cfg.timeout)
+        return sentinel
+
+    monkeypatch.setattr(_client_mod, "build_client", fake_build_client)
+
+    client = _client_mod.get_llm_client()
+
+    assert client is sentinel
+    assert received_timeouts == [135.0]

tests/unit/test_component/test_llm/test_factory.py

@@ -2,13 +2,17 @@
 
 from __future__ import annotations
 
+import importlib
+
 import pytest
 from pydantic import SecretStr
 
 from everos.component.llm import build_llm_provider
 from everos.component.llm.openai_provider import OpenAIProvider
 from everos.config.settings import LLMSettings
 
+_factory_mod = importlib.import_module("everos.component.llm.factory")
+
 
 def test_raises_when_api_key_missing() -> None:
     s = LLMSettings(model="m", api_key=None, base_url="https://x")
@@ -26,3 +30,27 @@ def test_builds_openai_provider() -> None:
     s = LLMSettings(model="m", api_key=SecretStr("k"), base_url="https://x")
     p = build_llm_provider(s)
     assert isinstance(p, OpenAIProvider)
+
+
+def test_passes_configured_timeout_to_provider(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    captured: dict[str, object] = {}
+    sentinel = object()
+
+    def fake_provider(**kwargs: object) -> object:
+        captured.update(kwargs)
+        return sentinel
+
+    monkeypatch.setattr(_factory_mod, "OpenAIProvider", fake_provider)
+    settings = LLMSettings(
+        model="m",
+        api_key=SecretStr("k"),
+        base_url="https://x",
+        timeout_seconds=135.0,
+    )
+
+    provider = _factory_mod.build_llm_provider(settings)
+
+    assert provider is sentinel
+    assert captured["timeout"] == 135.0

tests/unit/test_config/test_settings.py

@@ -76,6 +76,21 @@ def test_env_overrides_toml(monkeypatch: pytest.MonkeyPatch) -> None:
     assert s.sqlite.synchronous == "NORMAL"
 
 
+def test_llm_timeout_env_overrides_everos_toml(
+    monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+    root = tmp_path / "myroot"
+    root.mkdir()
+    (root / "everos.toml").write_text(
+        "[llm]\ntimeout_seconds = 90.0\n", encoding="utf-8"
+    )
+    monkeypatch.setenv("EVEROS_LLM__TIMEOUT_SECONDS", "135.0")
+
+    settings = Settings(_everos_root=root)
+
+    assert settings.llm.timeout_seconds == 135.0
+
+
 def test_init_args_override_env(monkeypatch: pytest.MonkeyPatch) -> None:
     monkeypatch.setenv("EVEROS_SQLITE__BUSY_TIMEOUT_MS", "10000")
     from everos.config.settings import SqliteSettings
@@ -98,6 +113,13 @@ def test_negative_busy_timeout_rejected() -> None:
         Settings.model_validate({"sqlite": {"busy_timeout_ms": -1}})
 
 
+def test_non_positive_llm_timeout_rejected() -> None:
+    from pydantic import ValidationError
+
+    with pytest.raises(ValidationError):
+        Settings.model_validate({"llm": {"timeout_seconds": 0}})
+
+
 def test_load_settings_is_cached() -> None:
     a = load_settings()
     b = load_settings()
@@ -118,6 +140,7 @@ def test_embedding_rerank_defaults() -> None:
     assert s.rerank.api_key.get_secret_value() == ""
     assert s.rerank.timeout_seconds == 30.0
     assert s.llm.api_key.get_secret_value() == ""
+    assert s.llm.timeout_seconds == 60.0
 
 
 def test_resolve_root_default(monkeypatch: pytest.MonkeyPatch) -> None: