Handle slow AI backend timeouts

README.md

@@ -38,7 +38,8 @@ Copiez `config.json.template` vers `config.json` puis adaptez :
   "model": "llama3.1",
   "api_key": "",
   "server_url": "http://localhost:11434",
-  "hotkey": "ctrl+alt+a"
+  "hotkey": "ctrl+alt+a",
+  "request_timeout_seconds": 300
 }
 ```
 
@@ -63,6 +64,10 @@ Copiez `config.json.template` vers `config.json` puis adaptez :
 }
 ```
 
+### Timeout IA
+
+`request_timeout_seconds` vaut `300` par défaut. Si Ollama charge un gros modèle ou répond lentement, augmentez cette valeur. Mettez `0` pour désactiver le timeout côté application.
+
 ## Compilation
 
 ### Windows

config.json.template

@@ -4,7 +4,7 @@
   "api_key": "",
   "server_url": "http://localhost:11434",
   "hotkey": "ctrl+alt+a",
-  "request_timeout_seconds": 120,
+  "request_timeout_seconds": 300,
   "type_delay_seconds": 0,
   "system_prompt": "Réponds directement et de manière ultra-concise. Aucune phrase d'introduction, aucune salutation, aucun formatage superflu. Uniquement la réponse brute."
 }

src/ai_typewriter/ai_client.py

@@ -34,6 +34,13 @@ def _ask_ollama(prompt: str, config: AppConfig) -> str:
         response = requests.post(url, json=payload, timeout=config.request_timeout_seconds)
         response.raise_for_status()
         data = response.json()
+    except requests.Timeout as exc:
+        timeout_label = "désactivé" if config.request_timeout_seconds is None else f"{config.request_timeout_seconds:.0f} s"
+        raise AIClientError(
+            "Ollama n'a pas répondu avant le délai configuré "
+            f"({timeout_label}). Le modèle est peut-être en chargement ou trop lent; "
+            "augmentez request_timeout_seconds dans config.json, ou mettez 0 pour désactiver le timeout."
+        ) from exc
     except requests.RequestException as exc:
         raise AIClientError(f"Erreur Ollama: {exc}") from exc
     except ValueError as exc:
@@ -65,6 +72,13 @@ def _ask_gemini(prompt: str, config: AppConfig) -> str:
         )
         response.raise_for_status()
         data = response.json()
+    except requests.Timeout as exc:
+        timeout_label = "désactivé" if config.request_timeout_seconds is None else f"{config.request_timeout_seconds:.0f} s"
+        raise AIClientError(
+            "Gemini n'a pas répondu avant le délai configuré "
+            f"({timeout_label}). Augmentez request_timeout_seconds dans config.json, "
+            "ou mettez 0 pour désactiver le timeout."
+        ) from exc
     except requests.RequestException as exc:
         raise AIClientError(f"Erreur Gemini: {exc}") from exc
     except ValueError as exc:

src/ai_typewriter/config.py

@@ -15,7 +15,7 @@ class AppConfig:
     api_key: str = ""
     server_url: str = "http://localhost:11434"
     hotkey: str = "ctrl+alt+a"
-    request_timeout_seconds: float = 120.0
+    request_timeout_seconds: float | None = 300.0
     copy_wait_seconds: float = 1.0
     type_delay_seconds: float = 0.0
     restore_clipboard: bool = True
@@ -33,6 +33,13 @@ def _coerce_provider(value: Any) -> Provider:
     return provider  # type: ignore[return-value]
 
 
+def _coerce_timeout(value: Any) -> float | None:
+    timeout = float(value)
+    if timeout <= 0:
+        return None
+    return timeout
+
+
 def load_config(path: str | Path = "config.json") -> AppConfig:
     cfg_path = Path(path)
     if not cfg_path.exists():
@@ -46,7 +53,7 @@ def load_config(path: str | Path = "config.json") -> AppConfig:
         api_key=str(data.get("api_key", "")),
         server_url=str(data.get("server_url", "http://localhost:11434")).rstrip("/"),
         hotkey=str(data.get("hotkey", "ctrl+alt+a")).lower(),
-        request_timeout_seconds=float(data.get("request_timeout_seconds", 120)),
+        request_timeout_seconds=_coerce_timeout(data.get("request_timeout_seconds", 300)),
         copy_wait_seconds=float(data.get("copy_wait_seconds", 1)),
         type_delay_seconds=float(data.get("type_delay_seconds", 0)),
         restore_clipboard=bool(data.get("restore_clipboard", True)),

tests/test_ai_client.py

@@ -1,4 +1,7 @@
-from ai_typewriter.ai_client import ask_ai
+import pytest
+import requests
+
+from ai_typewriter.ai_client import AIClientError, ask_ai
 from ai_typewriter.config import AppConfig
 
 
@@ -27,6 +30,19 @@ def test_ollama_payload_contains_strict_system_prompt(monkeypatch):
     assert "Uniquement la réponse brute" in seen["json"]["messages"][0]["content"]
 
 
+def test_ollama_timeout_message_suggests_config_change(monkeypatch):
+    def fake_post(url, json, timeout, **kwargs):
+        raise requests.Timeout("too slow")
+    monkeypatch.setattr("ai_typewriter.ai_client.requests.post", fake_post)
+
+    with pytest.raises(AIClientError) as exc:
+        ask_ai("texte", AppConfig(provider="ollama", model="m", request_timeout_seconds=300))
+
+    assert "300 s" in str(exc.value)
+    assert "request_timeout_seconds" in str(exc.value)
+    assert "0 pour désactiver" in str(exc.value)
+
+
 def test_gemini_payload(monkeypatch):
     seen = {}
     def fake_post(url, params, json, timeout):

tests/test_config.py

@@ -16,6 +16,15 @@ def test_load_default_config(tmp_path):
     assert "Réponds directement" in cfg.system_prompt
 
 
+def test_timeout_zero_disables_timeout(tmp_path):
+    path = tmp_path / "config.json"
+    path.write_text(json.dumps({"provider": "ollama", "request_timeout_seconds": 0}), encoding="utf-8")
+
+    cfg = load_config(path)
+
+    assert cfg.request_timeout_seconds is None
+
+
 def test_reject_invalid_provider(tmp_path):
     path = tmp_path / "config.json"
     path.write_text(json.dumps({"provider": "bad"}), encoding="utf-8")