How Litco protects the evidence in your matters.

Litco encrypts every matter with a separate key, enforces access control at the database level, and runs the Matter Agent’s code in a closed environment with no network connection.

Litco never uses your documents or matter data to train its AI. Litco’s cloud providers operate under zero-data-retention terms and keep nothing after processing a request. A firm can also keep a matter off the cloud entirely and run it on Litco’s own hardware.

One instance per firm, one encryption key per matter.

One firm’s deployment
Matter A key A

Litco seals every document stored for the matter with key A.

Matter B key B

Litco binds every file to a single matter. Lawyers who hold the key for one matter cannot use that key to open files in another.

Underneath both: Postgres row-level security, forced on the tables that hold matter data, with the application connecting as a role the database will not let bypass it.
Key custodian

The key that wraps each matter key stays inside a managed KMS or an on-prem HSM. Litco’s servers send wrap and unwrap requests to that device but never hold the wrapping key.

Outside the deployment
Drafting sandbox

The Matter Agent writes documents to one writable folder. Litco erases that folder when the run ends.

No network

One deployment per firm

Litco deploys as a firm-controlled instance. Your firm’s documents stay in your firm’s deployment, and the Matter Agent working your cases has no path into anyone else’s.

Envelope encryption, per matter

Litco encrypts all documents in transit and at rest with AES-256-GCM. For each matter, Litco generates a separate data key and wraps that key with a firm-level key held in a KMS. Litco tags every stored object to its matter and limits each key to that matter’s files.

Row security in the database

Litco enforces Postgres row-level security on the tables that hold matter data. The application connects as a role verified unable to bypass that security, so the database itself stops any bug in application code from reaching protected rows. Privileged system paths exist for the jobs that need them, and Litco reviews those paths as an audit surface.

Matter-based access control

Litco restricts every surface, including the Matter Agent’s, by seat, role, and matter membership. Litco enforces ethical-wall separations where reviewers read documents and again where they share those documents.

The Matter Agent runs its code in a sealed environment.

When the Matter Agent needs to compute something or build a file, it writes a program and runs it inside a sandbox with no network connection at all. Before starting, the Matter Agent verifies that the sandbox is fully isolated. If verification fails, it does not run.

No network

The Matter Agent writes and runs scripts and custom programs to assemble a document or run an analysis. That code runs with no network at all. It cannot reach the internet, and it cannot reach another program running on the same computer.

No shell

Litco confines the sandbox to Python alone. The code runs as an unprivileged user against a read-only set of tools.

Zero retention

One folder is writable, and Litco erases it when the run ends, whether the run succeeded or failed.

Litco routes every task to the right model.

Litco tests the models and sends each task to the one that performs best on that kind of work. Some tasks run on Litco’s own hardware, others through a provider under Litco’s account.

Where the model runsWhose API accountWhat leaves the firm’s deploymentWhat the provider keeps
Litco’s own hardware, running an open-weight modelNone. No third party is in the request.NothingNothing. There is no provider.
A cloud provider, on Litco’s routingLitco’s, with a separate key issued for each matter and a spending cap on itThe request text, pinned request by request to a zero-data-retention endpoint that is told to collect nothingNothing once the response is returned

Litco evaluates every option before choosing which model to use for a given task. LitigationBench publishes how each model performs on litigation tasks, including the self-hosted option. Litco defaults to the highest-scoring model for each task.

Controlled sharing with co-counsel and clients.

Lawyers share files externally through LitSpace links protected by passwords or one-time email codes. LitSpace watermarks each file on the way out and records what was sent. External recipients cost nothing, so teams have no reason to route files around LitSpace.

A record of what the Matter Agent did.

Litco keeps the full step record of every Matter Agent run. The database logs matter activity and import history append-only, enforced by triggers rather than by convention. The Matter Agent’s own file tools include no delete command, and no administrator can add one.

Security at a glance.

The questions a security questionnaire asks, in the order they usually arrive.

Encryption at restAES-256-GCM on every stored object, with a separate data key for every matter.
Key custodyEach matter’s data key is wrapped by a key that stays inside a managed KMS or an on-prem HSM. A production deployment refuses to start against a key file on local disk.
Access controlSeats, roles, and matter membership at the surface, with Postgres row-level security forced on the tables that hold matter data underneath.
Model trainingNever on your documents or matter data.
Where processing happensThe firm chooses per matter: a cloud provider on Litco’s zero-data-retention routing, or Litco’s own hardware with no third party at all.
Provider retentionCloud requests Litco routes are pinned, one by one, to zero-data-retention endpoints that are told to collect nothing.
SandboxNo network interface, no shell, and one writable folder that Litco erases when the run ends.
Deletion by the agentThe agent’s file tools include no delete.
Audit trailMatter activity and import history are append-only, enforced by database triggers.

For security questionnaires and diligence reviews, reach the team at hello@litco.ai.