Sediment
A self-hosted evidence store for coding agents that captures model calls, code changes, developer decisions and check results, so teams can compare models, reuse prior session context, and build training datasets.
Sediment records what a coding agent did on a team's own infrastructure: model calls, the code changes that resulted, developer decisions and the results of automated checks. That evidence can be used to evaluate agent work by comparing models on recorded decisions, code retention and check outcomes, to reuse selected messages, tool calls and results from a previous session, or to export examples for fine-tuning, preference training or reinforcement learning.
It integrates with Claude Code, Codex, Cursor, the Pi agent and Copilot Chat, though the signals available vary by agent. Installation is a shell script on macOS with Homebrew, Debian or Ubuntu, followed by a local server that a team can also deploy to EC2 or their own host if they already run PostgreSQL and HTTPS.
It is licensed under AGPL-3.0, with the harness shims under the MIT license.