The systems inside Litco.
Core systems on the platform include a per-matter memory, a verification stack, a sealed drafting sandbox, an architecture for long autonomous work, per-firm deployment with per-matter keys, and public measurement of the models behind the platform.
Matter memory
The Matter Agent keeps a durable memory per matter. It records observations as it works, consolidates them overnight, and starts the next session already knowing the case. It learns the firm’s drafting voice from attorney edits. No client document ever trains a model.
The verification stack
Forty-one distinct software checks run against model output, from byte-level quote comparison to independent confirmation of negative treatment. Details →
The drafting sandbox
When the Matter Agent writes a document, the code that produces it runs in a sandbox with no network access. Nothing leaves the sandbox until the document’s quotations pass verification.
Durable autonomy
The Matter Agent runs a standing objective as a chain of bounded turns, and between turns a second model decides whether to continue, finish, or hand the work to a person. It survives restarts, waits for your approvals, and ends by delivering whatever it completed.
Deployment
Each firm runs on its own instance, and each matter carries its own encryption key. Firms choose, per matter, whether inference happens on cloud models under zero-retention terms or on Litco’s own hardware.
Measurement
Litco benchmarks the models it uses through the production system, posts every number, and routes work accordingly. LitigationBench →