The Atlas Lavern's documentation, bound to its code
111 documents

First hour with the repo

The orientation path: what Lavern is, how to run it, where everything lives, and how to navigate from here.

Pick a provider

The setup prompts you to choose one:

Provider What it is Cost Needs
Anthropic Cloud (default, recommended) Claude via Anthropic API. Best capability. The agents, debate, and verification loop were tuned for Claude. Paid per call sk-ant-... key from console.anthropic.com
Local (Ollama) On-device inference via gemma3:4b (or any model you pull). Nothing leaves the host. Quality drops noticeably vs Claude — use for offline tinkering and privacy demos, not serious matters. $0 Ollama installed; setup will offer to brew install (macOS) or curl | sh (Linux). ~3 GB model pull.
Mistral (EU Sovereign) EU-hosted inference for GDPR-bound matters. Paid per call Key from console.mistral.ai

To change provider later: edit .env or re-run npm run setup.