Add optional Ollama chat fallback engine

ollama_engine.py

@@ -0,0 +1,114 @@
+"""
+ollama_engine.py — optional external Ollama fallback for LINGO.
+
+Use this when Hugging Face provider credits are depleted but you have a local
+machine/server running Ollama behind Cloudflare Tunnel or ngrok.
+"""
+
+from __future__ import annotations
+
+import logging
+from typing import Any
+
+import httpx
+
+from config import Settings
+
+logger = logging.getLogger(__name__)
+
+SYSTEM_PROMPT = """You are LINGO, a concise AI phone receptionist.
+Speak in short, natural sentences suitable for a live phone call.
+You can answer basic questions, take messages, collect appointment requests,
+and escalate when the caller asks for a human.
+If you do not know a business-specific detail, offer to take a message.
+If escalation is needed, end the reply with [ESCALATE].
+Keep replies under 55 words.
+"""
+
+
+def enabled(settings: Settings) -> bool:
+    return bool(settings.ollama_enabled and settings.ollama_chat_url and settings.ollama_model)
+
+
+def _messages_for_ollama(messages: list[dict[str, Any]], settings: Settings) -> list[dict[str, str]]:
+    out: list[dict[str, str]] = [
+        {
+            "role": "system",
+            "content": SYSTEM_PROMPT + f"\nBusiness name: {settings.business_name or 'the business'}.",
+        }
+    ]
+    for message in messages[-8:]:
+        role = str(message.get("role", "user"))
+        content = str(message.get("content", "")).strip()
+        if not content:
+            continue
+        if role not in {"user", "assistant"}:
+            continue
+        out.append({"role": role, "content": content[:1200]})
+    return out
+
+
+def _headers(settings: Settings) -> dict[str, str]:
+    headers = {"Content-Type": "application/json"}
+    if settings.ollama_auth_token:
+        headers["Authorization"] = f"Bearer {settings.ollama_auth_token}"
+    return headers
+
+
+async def chat_completion(
+    messages: list[dict[str, Any]],
+    http_client: httpx.AsyncClient,
+    settings: Settings,
+) -> dict[str, Any]:
+    if not enabled(settings):
+        return {"ok": False, "source": "ollama", "error": "Ollama fallback disabled or missing config"}
+
+    payload = {
+        "model": settings.ollama_model,
+        "messages": _messages_for_ollama(messages, settings),
+        "stream": False,
+        "options": {
+            "temperature": 0.35,
+            "num_predict": 120,
+        },
+    }
+
+    try:
+        response = await http_client.post(
+            settings.ollama_chat_url,
+            json=payload,
+            headers=_headers(settings),
+            timeout=float(settings.ollama_timeout_seconds),
+        )
+        if not response.is_success:
+            return {
+                "ok": False,
+                "source": "ollama",
+                "status_code": response.status_code,
+                "error": response.text[:500],
+            }
+        data = response.json()
+        content = data.get("message", {}).get("content") or data.get("response") or ""
+        content = str(content).strip()
+        if not content:
+            return {"ok": False, "source": "ollama", "error": "empty Ollama response"}
+        return {
+            "ok": True,
+            "source": "ollama",
+            "model": settings.ollama_model,
+            "content": content[:700],
+        }
+    except Exception as exc:  # noqa: BLE001
+        logger.error("Ollama fallback failed: %s", exc)
+        return {"ok": False, "source": "ollama", "exception": str(exc)[:500]}
+
+
+def diagnostics(settings: Settings) -> dict[str, Any]:
+    return {
+        "enabled": enabled(settings),
+        "configured": bool(settings.ollama_chat_url),
+        "chat_url_set": bool(settings.ollama_chat_url),
+        "model": settings.ollama_model,
+        "auth_token_set": bool(settings.ollama_auth_token),
+        "timeout_seconds": settings.ollama_timeout_seconds,
+    }