Documentation
Three ways to use G6: through the MCP server from your agent, as a Python library, or from a worked example. The code is on GitHub.
- Get startedInstall G6 on your own computer and connect it to your coding agent.
- Developer quickstartThe MCP server, the Python library and the command line.
- Tune G6 to your workThe meta-harness: proposes changes, tests them on held-out tasks, and changes nothing until you approve.
- ComponentsWhat each part of G6 does, and how mature it is.
- Worked examplesA restricted subset of G6's components applied to one problem, ready to download.
- Build a workflowTurn a task you repeat into a G6 workflow.
- ResearchEvery pre-registered study, including the ones that found nothing.
- The free guideFailure and harness engineering for AI agents: why agent work fails, and the patterns that catch it.
- Full documentationThe code on GitHub, and a map of every file, class and function in it.