The fastest path is hosted: Agent Relay runs the integration stack for you — OAuth, sync workers, writeback — so you only run a local mount. One command signs you up and puts provider-backed files in an ordinary directory.
Want to run the whole stack yourself instead? See Run locally for the self-hosted Docker path. The source lives at github.com/AgentWorkforce/relayfile — clone it first.
One command
From the project where your agent will work, run:
npx relayfile@latestRelayfile opens Google sign-in, creates your account and a workspace named after the current directory, connects GitHub, and mounts at ./relayfile-mount. No invite code, API token, existing Cloud account, or separate agent-relay installation is required.
Keep that terminal open for synchronization. In another terminal, ./relayfile-mount is the live mirror of your workspace:
ls ./relayfile-mount/github
cat ./relayfile-mount/LAYOUT.mdFor a different provider or explicit names, use:
npx relayfile@latest setup \
--provider notion \
--workspace my-agent \
--local-dir ./relayfile-mountAdd --no-open to print the hosted login and connect URLs instead of opening a browser, which is useful inside agents and headless environments.
See Relayfile Cloud for what the hosted plane manages, and Mounting to do the same thing programmatically from a sandbox.
Read and write
Once mounted, the interface is the filesystem you already know:
# read
cat ./relayfile-mount/github/repos/acme/api/pulls/42/metadata.json
# post a pull-request review by writing the canonical path
cat > ./relayfile-mount/github/repos/acme/api/pulls/42/reviews/review.json <<'EOF'
{"body":"Looks good to me.","event":"COMMENT"}
EOF
# find pull requests across connected repositories
find ./relayfile-mount/github/repos -path '*/pulls/*/metadata.json'See Reads and writes for the full PATCH / CREATE / DELETE model.
Next steps
Mount layout
Learn the tree shape: LAYOUT.md, _index.json, canonical naming, and alias views.
Reads and writes
How file ops map to PATCH, CREATE, and DELETE — and where schemas live.
The SDK
Use RelayFileClient from any agent runtime, with the built-in read cache.
Agents
One connect() call for Vercel AI SDK, OpenAI Agents SDK, and LangChain.