Grainulator
A local, portable evidence ledger for research and engineering work done with an AI model, tracking claims, sources and open gaps across sessions and models so an answer can be checked rather than taken on faith.
Grainulator keeps typed claims, their sources, an evidence tier and provenance information attached to a task as it moves between different models and sessions, so the record survives edits and exports rather than living only in a chat transcript. It flags conflicts between claims and weak-support signals, and produces two lists: what an agent can safely continue automatically, and what needs a person's input before continuing.
It works as a plugin for Claude Code and Codex, and separately provides a command-line tool and an MCP connection so other agents can use it; a local research playground can also run models directly using a provider account configured on the machine. The project describes its output as staying local and not sending a user's questions to a model on its own.
It is written in JavaScript, distributed as a GitHub release rather than an npm package, and requires Node.js 24 or newer.