Lavern
An agentic law firm with 67 AI agents and evidence-backed debate.
67 specialist AI agents that review documents through evidence-backed debate.
Lavern coordinates 59 specialists and 7 orchestrators through a structured debate protocol to review legal documents. It features mandatory human gates and a 10-pass verification loop for critical decisions. Built for legal professionals and developers looking to test multi-agent architectures.
What it is
The system reads documents, posts findings with cited evidence, runs three layers of verification, and pauses at a human gate before critical decisions land. It includes 21 MCP tools for debate, scoring, and verification, plus a persistent precedent board across engagements.
Who it's for
Suits legal tech developers, firm founders, and engineers experimenting with multi-agent coordination frameworks. It does not suit practitioners seeking an out-of-the-box commercial legal product with guaranteed output quality.
Key features
67 Agent Prompts
59 specialists, 7 orchestrators, and 1 generic base orchestrator coordinating workflows.
Debate Protocol
Agents must cite specific text from the parsed document to enter findings on the board.
Three-Layer Verification
Evaluator gate, adversarial debate, and a 10-pass verification pipeline fail closed independently.
Precedent Board
Persistent memory across engagements that reinforces recurring findings and decays stale ones.
Autonomous Clawern Mode
Processes new documents on a 30-minute heartbeat and pushes findings to Telegram or email.
Up and running in 3 steps
Our honest verdict
What's great
- Runs against Anthropic Claude, Mistral AI in the EU, or locally via Ollama
- Mechanical grounding verifier cross-checks every cited quote via string matching
- Includes 5 seeded legal datasets like CUAD, MAUD, and LEDGAR for testing
- Features an editorial React dashboard with live activity feeds and audit bundles
Where it falls short
- Legal-quality claims are an open empirical hypothesis without independent validation
- Requires API keys for Anthropic or Mistral for real engagements, plus running a local server
- One feature, the Lavern Challenge, still directly instantiates the Anthropic client in EU mode
- Requires managing Node.js dependencies for both backend and frontend visualization components
Pricing & license
Free under the Apache 2.0 license. The only direct cost is running the infrastructure yourself and paying for the underlying LLM API tokens (Anthropic or Mistral) unless operating locally via Ollama.
Open remote AI-training roles
767 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Is Lavern ready to use as a commercial product?
No. The author explicitly states this is not a product, but rather a collection of ideas and source of inspiration.
Can I run this entirely locally without sending data to US or EU cloud providers?
Yes, you can run the system locally via Ollama by configuring the provider setting.
What happens in EU mode?
Setting the provider to Mistral routes the orchestrators, agents, debate, and verification through Mistral AI in Paris.
