CiteSafe

Intel Inside · API for legal AI

CiteSafe is the trust backend other legal AI plugs into

Generators write. Suites research. You should not self-grade your own cites. Ship independent verification as infrastructure — statuses, lineage, and audit hashes your customers can show a partner or a court.

Partnership / white-label

Independent gate

Model-agnostic. Works after Harvey, GPT, Claude, or human draft.

Named layers

Heuristics · corpus · CourtListener · CAP · GovInfo — no black box green checks.

White-label ready

Badge + report hash in your product. You keep the UX; we own trust.

Endpoints

FastAPI engine ships /v1/verify today. Production tokens + rate limits: set NEXT_PUBLIC_API_URL . Beta keys are issued in your workspace.

  • POST /v1/verify

    Extract + verify citations in text. Returns statuses, confidence, source layers, report hash.

  • POST /v1/upload

    Multipart brief upload (when workers live). Same report shape as verify.

  • GET /v1/report/:id

    Fetch a prior signed verification report by id.

  • GET /v1/sources

    Public methodology registry — layers, providers, legal envelope.

  • GET /health

    Engine version, corpus size, provider configuration.

Example
curl -X POST https://api.citesafe.co/v1/verify \
  -H "Authorization: Bearer cs_live_…" \
  -H "Content-Type: application/json" \
  -d '{"text":"See Brown v. Board, 347 U.S. 483 (1954)."}'

# → citations[].status: verified | needs_review | could_not_verify
# → report_hash, providers_used, legal disclaimer envelope

Who this is for

  • Legal AI startups that refuse to self-certify their own drafts
  • Practice-management / DMS vendors adding a verify step
  • In-house platforms routing every model output through one trust layer
  • White-label “CiteSafe Verified” badges inside another product

Honest beta: Existence and fabrication-risk signals ≠ good law. Integrators must surface CiteSafe disclaimers. Not a substitute for counsel judgment or licensed citators.

Methodology