Contract review & e-discovery, self-hosted

The pain we built this for

If you run legal ops or contract review, the AI tools all hit the same wall: “We can’t send privileged documents to a cloud model, we can’t trust an unattributed clause summary, and we need a defensible record of how the review was done.”

• Privileged and confidential documents cannot go to a third-party AI API — which rules out most legal-AI SaaS.
• An unattributed clause summary is useless (and risky) — every flag needs to cite the exact source clause a lawyer can verify.
• Obligations (party, trigger, deadline, remedy) are buried in prose when they should be a queryable structure.
• E-discovery by keyword misses responsive documents that only match by meaning.
• Defensibility requires a consistent, logged review process — not an ad-hoc one that varies by reviewer.

See it — obligations as a graph, review as an audit

One Cypher query for what’s due this quarter, one immutable table for the review trail.

-- Extract obligations from a contract into a queryable graph.
CYPHER
MATCH (c:Contract {id: $doc})-[:HAS_CLAUSE]->(cl:Clause)-[:CREATES]->(o:Obligation)
  WHERE o.deadline < date('2026-10-01')
RETURN c.counterparty, o.trigger, o.remedy, o.deadline
ORDER BY o.deadline;                 -- what's due this quarter, across the book

-- Every AI-assisted review lands on a tamper-evident audit ledger.
CREATE IMMUTABLE TABLE review_audit (
  doc_id       TEXT,
  model_version TEXT,
  flags        JSONB,          -- risky clauses cited to source
  reviewer     TEXT,
  decision     TEXT
) WITH (ENCRYPTION = 'AES256GCM');

VERIFY TABLE review_audit;

Six capabilities. One self-hosted stack.

Clause review, obligation graph, semantic e-discovery, immutable audit — over one store, on your infra.

The objections counsel raises first