email-classifier

FastAPI service that classifies email using a configurable LLM backend.

What changed

The classifier no longer hardcodes a single Ollama + OpenAI-compatible endpoint. It now supports:

• OpenAI-compatible APIs
• Anthropic-compatible APIs
• per-request overrides for provider, model, endpoint, and temperature
• global defaults through environment variables

This makes it suitable for local Ollama, hosted OpenAI-compatible services, and MiniMax's recommended Anthropic-compatible API.

Environment configuration

Defaults are loaded from environment variables:

export LLM_PROVIDER=openai
export LLM_BASE_URL=http://ollama.internal.henryhosted.com:9292/v1
export LLM_API_KEY=none
export LLM_MODEL=qwen2.5-7b-instruct.q4_k_m
export LLM_TEMPERATURE=0.1
export LLM_TIMEOUT_SECONDS=60
export LLM_MAX_RETRIES=3

MiniMax example

MiniMax recommends Anthropic-compatible integration.

export LLM_PROVIDER=anthropic
export LLM_BASE_URL=https://api.minimax.io/anthropic
export LLM_API_KEY=your_minimax_key
export LLM_MODEL=MiniMax-M2.7

API

POST /classify

Request body:

{
  "email_data": {
    "subject": "Can you review this by Friday?",
    "body": "Hi Daniel, please review the attached budget proposal."
  },
  "provider": "anthropic",
  "base_url": "https://api.minimax.io/anthropic",
  "model": "MiniMax-M2.7",
  "temperature": 0.1
}

All override fields are optional. If omitted, the service uses the global env config.

Response shape:

{
  "needs_action": true,
  "category": "question",
  "priority": "high",
  "task_description": "Review the budget proposal and respond by Friday",
  "reasoning": "Direct request with a deadline requires follow-up",
  "confidence": 0.91
}

Architecture

• app/config.py: global and per-request LLM settings
• app/llm_adapters.py: provider adapters
• app/classifier.py: classification orchestration, retries, normalization
• app/prompts.py: system prompt
• app/routers/classify_email.py: thin API route

Notes

• OpenAI-compatible providers use the OpenAI SDK.
• Anthropic-compatible providers use the Anthropic SDK.
• Per-request api_key is supported, but excluded from response serialization.
• The service normalizes malformed model output and falls back safely after retry exhaustion.