Litatoli

The evidence layer for AI-written code.

Litatoli ("testimony / evidence" in Lingala) answers one question: what did a model actually produce, against which requirement, verified how, approved by whom, and sealed when? It produces a chain that verifies offline, with no server in the loop — so the artefact you ship is the artefact an auditor reads.

Litatoli proves code it did not generate. It is not tied to any code generator: point it at output from Copilot, Cursor, Claude Code, an in-house agent, or a human. That independence is what makes the evidence worth anything — a tool that both produces an artefact and vouches for it is worth less to an auditor than one that hands the record to something else.

Where to start

• Getting Started — from an empty shell to a first sealed-and-verified chain, in a few minutes.
• Reference — every command, flag and evidence type.
• Evidence format — the on-disk shape, so a third party can verify a chain without running our code.
• The first chain — a real chain produced over real work, kept so it can be read by someone who was not there.

What a chain gives you, exactly

A chain is tamper-evident, not tamper-proof: any edit, deletion, insertion or reordering breaks it, and verification stops at the first invalid link and says which. That is the guarantee — detection, not prevention.

Verification is offline and third-party: it checks the BLAKE3 chain and the Ed25519 signature against a public key you pin yourself. Nothing calls home, and nothing needs our servers to be running — a year from now, or after we are gone.

The evidence is an input an auditor can use. It is not a certification, and Litatoli does not claim conformance to any standard on your behalf.