The procedures for running FAVA Trails live in the FAVA Trails repository, which owns every default, command and failure behavior. If you are still deciding whether FAVA Trails fits, start with fit and limits.
Open the setup instructions →
Set up a data repository and a reviewer
The setup instructions cover per-machine setup, providers, credentials and the configuration reference. The README adds client registration examples for Windows with WSL and OpenAI Codex CLI, and a maintenance procedure for empty scopes. Choosing a reviewer is a data-egress decision, because a reviewer receives the candidate before it returns a verdict. The effective configuration decides where review text goes, and fava-trails doctor shows it. Point doctor at your data by exporting the absolute FAVA_TRAILS_DATA_REPO first, because the default ~/.fava-trails may hold a different repository. The local review walkthrough records one run against a local reviewer, including the failure cases.
Give each agent its own identity
The identity is a stable role name that the operator chooses, not a host fingerprint, so a new machine does not need a new one. Processes that reuse a name share its authoring view, and agents that need private drafts need separate names and endpoints. Governed recall explains the three visibility modes, operator-set agent identity and why one shared endpoint is one identity boundary. The usage instructions describe the session protocol agents follow.
Publish and sync across machines
Publishing and syncing are separate. Manual push mode needs an explicit push from the data repository, and push_strategy: immediate lets the server push after a successful write operation, a successful sync included. The sync tool itself fetches and rebases. Read the push procedure in the setup instructions. For a shared repository that a whole team uses, see team rollout, an optional guide to the choices a team makes.
Upgrade and check what is running
The runtime and upgrade guide explains how to confirm which install your MCP client launches and why you restart a server after changing the package.
Expose a private endpoint
The Secure MCP Tunnel guide describes running the private runtime behind a tunnel for a long-lived ChatGPT connection. If ChatGPT created empty scaffolding scopes before the read-only guard existed, the README’s cleanup procedure removes them after a dry run.
Optional extensions
Lifecycle protocols such as SECOM, ACE and RLM are optional hooks that you enable one at a time in the data repository.
Where the facts live
The documentation ownership map names the document that owns each fact, and release history is in the changelog.