Skip to Content
ExamplesContract & risk review

⚖️ Cross-vendor contract and risk review

Stack — Trigger: upload / agent API · Models: Claude + GPT-4o + Gemini members, judge · Tools: clause/precedent search (optional) · Mode: debate collaboration · Output: risk memo citing the debate

Who it’s for — legal, procurement, and diligence teams where one model’s read isn’t enough. What it does — several models argue a document’s risks from different angles and a judge writes the final memo — so a missed indemnity or auto-renewal clause is caught by the model that was looking for it, not left to one opinion.

Build it

  1. Create member agents, each a different provider and each given a lens: tenant’s risk, counterparty’s leverage, compliance with governing law.

  2. Set collaboration to debate on the primary agent — members run over several rounds, each seeing the others’ arguments, so weak readings get challenged.

    The collaboration panel — debate strategy with member agents

  3. Pick a judge model that reads the full exchange and writes the final answer, citing which arguments decided it.

  4. (Optional) add a search tool for your clause library or past contracts so arguments cite precedent, not just the text.

  5. Feed it the document — upload the contract or have your DMS call the agent API.

  6. Get the memo — a structured risk summary with the reasoning, not a single-model guess.

Make it yours

  • Reuse the pattern for research synthesis, due diligence, or incident post-mortems — anywhere one model isn’t enough.
  • Reserve debate for high-stakes calls; use race or parallel for high-volume, low-stakes work.
  • Route anything above a risk score to a human via request_approval.

Uses — Legal · Debate · Multi-model

Reconcile Stripe payoutsGoverned production agent