# Sediment

> The open-source, self-hosted evidence store for coding agents. Evaluate agent work, reuse evidence as context, and build training datasets.

- [Sediment](https://docs.sediment.so/index.md): Sediment is an open-source, self-hosted evidence store for coding agents.
- [Quickstart](https://docs.sediment.so/quickstart.md): Run a local server and verify synthetic capture in a scratch repository.
- [Agent integrations](https://docs.sediment.so/agents/overview.md): Compare agent capture capabilities and choose an integration guide.
- [Capture Claude Code work](https://docs.sediment.so/agents/claude-code.md): Capture Claude Code decisions, Attribution, inference calls, and optional edit evidence.
- [Capture Codex work](https://docs.sediment.so/agents/codex.md): Capture Codex decisions, Attribution, inference calls, and optional edit evidence.
- [Capture Cursor work](https://docs.sediment.so/agents/cursor.md): Capture Cursor commit Attribution and implicit accepts from local desktop work.
- [Capture pi work](https://docs.sediment.so/agents/pi.md): Capture pi implicit accepts, Attribution, inference calls, and optional edit evidence.
- [How Sediment works](https://docs.sediment.so/how-sediment-works.md): One evidence store for evaluating agent work, reusing selected context, and building training datasets.
- [Architecture](https://docs.sediment.so/architecture.md): Sediment's components, data flow, persistence boundaries, and network boundaries.
- [Why Sediment?](https://docs.sediment.so/operational-value.md): Questions about model outcomes, retained code, rework, and cost.
- [Deploy Sediment on EC2](https://docs.sediment.so/operate/deploy-ec2.md): Install a published Sediment version on one EC2 instance, with PostgreSQL and Traefik HTTPS.
- [Deploy Sediment on your own host](https://docs.sediment.so/operate/deploy.md): Install a published Sediment version on a host where you already run PostgreSQL and HTTPS.
- [Enroll your team](https://docs.sediment.so/operate/run-pilot.md): Create capture tokens, connect repositories and developer machines, and verify each agent's capture.
- [Measure agent work](https://docs.sediment.so/operate/measure-agent-work.md): Compare model outcomes, find accepted work that didn't ship, and measure retention through merge.
- [Continue a task with captured evidence](https://docs.sediment.so/operate/resume-with-evidence.md): Give an agent exact captured parts from a previous Session, through retrieval tools or a CLI evidence packet.
- [Maintain a deployment](https://docs.sediment.so/operate/maintain.md): Upgrade, back up, rotate credentials, quarantine data, troubleshoot, and tear down a deployment.
- [Secure a deployment](https://docs.sediment.so/operate/security.md): Review stored data, credential authority, and network exposure, and verify a release.
- [Rehearse the single-host Compose deployment](https://docs.sediment.so/operate/rehearse-compose.md): For contributors: build and qualify the repository's PostgreSQL, API, LiteLLM, and Traefik containers on one host.
- [How capture works](https://docs.sediment.so/capture/how-capture-works.md): The five capture signals, their data paths, attribution, edit survival, and privacy boundaries.
- [Local capture](https://docs.sediment.so/capture/local-capture.md): Connect one developer machine, its coding agents, and its repositories to Sediment.
- [Managed capture](https://docs.sediment.so/capture/managed-capture.md): Connect GitHub webhooks, private repository mirrors, a model gateway, and a managed developer fleet.
- [How Derivation works](https://docs.sediment.so/derive/how-derivation-works.md): Facts, snapshots, resolved policy, cohort selection, and the canonical derived bundle.
- [Attribution](https://docs.sediment.so/derive/attribution.md): Git notes and jaccard fallback join commits to inference calls.
- [Run derivations](https://docs.sediment.so/derive/run-derivations.md): Run, scope, inspect, recompute, and export a derivation.
- [Choose a training export](https://docs.sediment.so/exports/training-exports.md): Choose a training objective, prepare its evidence, and audit the output.
- [Export for a consumer](https://docs.sediment.so/exports/consumer-compatibility.md): Export for pinned downstream consumers with explicit configuration and auditable evidence.
- [Export DPO pairs](https://docs.sediment.so/exports/dpo.md): Pair chosen and rejected responses from matching prompt buckets.
- [Export SFT and diff-SFT rows](https://docs.sediment.so/exports/sft.md): Export eligible responses or exact attributed-file patches.
- [Export Recovery rows](https://docs.sediment.so/exports/recovery.md): Export red-to-green CI transitions with their fixing diffs.
- [Export RLVR tasks and trajectories](https://docs.sediment.so/exports/rlvr-export.md): Verifiable task specifications and reward-labeled trajectories: tasks.jsonl, rollouts.jsonl, and the taskset manifest.
- [CLI reference](https://docs.sediment.so/reference/cli.md)
- [HTTP API reference](https://docs.sediment.so/reference/api.md)
- [Schema reference](https://docs.sediment.so/reference/schema.md)
- [Consumer compatibility reference](https://docs.sediment.so/reference/compatibility.md)