AI Providers

Configure Claude, OpenAI, Gemini, or Ollama for langsight investigate

langsight investigate uses an LLM to analyse health evidence and produce root cause reports. You can choose from four providers.

Quick setup

Add an investigate block to .langsight.yaml:

investigate:
  provider: gemini           # anthropic | openai | gemini | ollama
  model: gemini-2.0-flash    # optional — sensible defaults per provider

Comparison

	Claude	OpenAI	Gemini	Ollama
Free tier	No	No	✅ 1,500/day	✅ Unlimited
Data privacy	Anthropic	OpenAI	Google	On your machine
Context window	200K	128K	1M	Varies
RCA quality	⭐⭐⭐⭐⭐	⭐⭐⭐⭐	⭐⭐⭐⭐	⭐⭐⭐
Setup time	2 min	2 min	2 min	5 min

Environment variables

Provider	Variable	Where to get it
Claude	ANTHROPIC_API_KEY	console.anthropic.com
OpenAI	OPENAI_API_KEY	platform.openai.com/api-keys
Gemini	GEMINI_API_KEY	aistudio.google.com/app/apikey
Ollama	(none required)	—

Never put API keys directly in .langsight.yaml. Use environment variables. The api_key field is only for CI/CD secrets injection.

Rule-based fallback

If no provider is configured or the API call fails, langsight investigate automatically falls back to deterministic heuristics — no LLM needed.

Claude

Best quality RCA with adaptive thinking.

OpenAI

GPT-4o for teams already on OpenAI.

Gemini

Free tier, 1M context window.

Ollama

Local, free, air-gapped.