Frist commit

backend/llm_client.py

@@ -0,0 +1,93 @@
+"""Local Ollama client for mail analysis."""
+
+from __future__ import annotations
+
+import json
+import logging
+import re
+from typing import Any
+
+import httpx
+
+from config import LLM_BASE_URL, LLM_MODEL, LLM_TIMEOUT_SECONDS
+
+logger = logging.getLogger("mail_summary")
+
+_JSON_BLOCK_RE = re.compile(r"\{.*\}", re.DOTALL)
+
+
+class LlmError(RuntimeError):
+    pass
+
+
+def extract_json_object(text: str) -> dict[str, Any]:
+    text = (text or "").strip()
+    if not text:
+        raise LlmError("empty_llm_response")
+    try:
+        data = json.loads(text)
+        if isinstance(data, dict):
+            return data
+    except json.JSONDecodeError:
+        pass
+    match = _JSON_BLOCK_RE.search(text)
+    if not match:
+        raise LlmError("llm_response_not_json")
+    data = json.loads(match.group(0))
+    if not isinstance(data, dict):
+        raise LlmError("llm_response_not_object")
+    return data
+
+
+async def chat_json(
+    *,
+    system: str,
+    user: str,
+    model: str | None = None,
+    temperature: float = 0.2,
+) -> dict[str, Any]:
+    selected = (model or LLM_MODEL).strip() or LLM_MODEL
+    payload = {
+        "model": selected,
+        "stream": False,
+        "format": "json",
+        "options": {"temperature": temperature},
+        "messages": [
+            {"role": "system", "content": system},
+            {"role": "user", "content": user},
+        ],
+    }
+    url = f"{LLM_BASE_URL}/api/chat"
+    try:
+        async with httpx.AsyncClient(timeout=LLM_TIMEOUT_SECONDS) as client:
+            response = await client.post(url, json=payload)
+            response.raise_for_status()
+            body = response.json()
+    except httpx.HTTPError as exc:
+        logger.exception("Ollama request failed")
+        raise LlmError(f"ollama_request_failed: {exc}") from exc
+
+    content = (
+        (body.get("message") or {}).get("content")
+        or body.get("response")
+        or ""
+    )
+    return extract_json_object(content)
+
+
+async def ping_llm(model: str | None = None) -> dict[str, Any]:
+    selected = (model or LLM_MODEL).strip() or LLM_MODEL
+    try:
+        async with httpx.AsyncClient(timeout=3.0) as client:
+            tags = await client.get(f"{LLM_BASE_URL}/api/tags")
+            tags.raise_for_status()
+            data = tags.json()
+    except Exception as exc:  # noqa: BLE001
+        return {"ok": False, "error": str(exc), "model": selected}
+    names = [m.get("name", "") for m in data.get("models", [])]
+    return {
+        "ok": True,
+        "model": selected,
+        "model_present": any(selected in name or name in selected for name in names),
+        "models": names[:20],
+    }