Run rp agent configure in a terminal, choose OpenCode or Codex, and
follow its sign-in prompts. rp remembers your choice in a private local profile
without starting or pairing the runner. When you're ready to connect, run
rp agent start.
rp agent configure [--engine opencode|codex] [--profile <directory>]
Omit the flags for interactive setup with the default profile. Use --engine
to skip the harness picker (the harness's own sign-in may still need input).
If you specify --profile, use the same path when starting the runner.
Stop a runner using this profile before configuring it. For OpenCode V2,
rp runs OpenCode's terminal sign-in
in the private profile. Choose a provider and follow its prompts; you don't
need to open the full OpenCode interface. To override the initial model, set
model in this profile's OpenCode configuration or start the runner with
--model provider/model.
Credentials stay in that local profile; pairing does not supply a provider
account or promise compatibility with every subscription or OAuth provider.
Use the same --profile when starting the runner.
To prepare Codex credentials instead, use rp agent configure --engine codex --profile <directory>. This runs codex login with a private CODEX_HOME inside
the selected runner profile. Sign in with ChatGPT once; the ordinary personal
Codex login and its sessions are not imported. Credentials remain on the local
runner, not in Repopack's database. Start this same isolated profile with
rp agent start --profile <directory>. Codex is its own native
engine, not an OpenCode provider alias. Keep the profile and workspace to
preserve native threads across restarts; do not copy your personal Codex home.
Next: start the runner or follow local agent setup.