Skip to main content
Use this page to choose an agent integration, and then follow that agent’s guide. Every guide assumes the machine setup in Configure local capture. To enroll a whole team, follow Enroll your team.

​
Compare integrations

A Developer decision can record a human choice or an automatic result, and Sediment keeps that distinction in explicit. Each agent reports decisions in a different unit, so decision counts aren’t comparable across agents. Developer decisions by agent harness defines each agent’s events and missing signals. Edit observations pair the text that an agent applied with the file’s content at Session end. Claude Code transcript capture also records external line counts, Rejected edits, and Retry linkages. The other agents don’t. Codex decision telemetry can include patch text even when transcript capture is off. Privacy boundaries and ceilings lists what each capture path sends.

​
GitHub Copilot Chat

Sediment reads three Copilot Chat log events: copilot_chat.edit.feedback, copilot_chat.inline.done, and copilot_chat.edit.survival. It doesn’t cover Copilot CLI. sediment install doesn’t configure Copilot Chat. Configure the client, or a collector, to send those events as authenticated OpenTelemetry Protocol (OTLP) HTTP/JSON logs to POST /v1/logs. See Copilot’s OpenTelemetry configuration. Traces alone don’t carry the decision events. After telemetry flushes, check that developer_decisions grows in sediment facts.

​
Integrate another agent

Sediment has no built-in integration for Gemini CLI, OpenCode, or Factory. A compatible gateway can capture their inference calls when the client carries a Session identifier. Developer decisions, commit notes, and Edit observations need an agent integration. To build one, start with What a harness client sends and the pi extension. The agent also needs an entry in Sediment’s AgentHarness enum. The server skips unregistered agents.