Skip to main content
Agents can be powerful Genseo users, but they must operate inside strict project and safety boundaries.

​
Required startup behavior

Every agent should call /me, /projects, genseo me, genseo_me, or genseo_projects_list first. Use the response to learn:
  • Which projects are authorized
  • Which single Project ID the user selected for this chat
  • Which domain the project represents
  • Whether the key is valid
Agents must never guess Project IDs or silently switch projects. Every project tool must include the selected project_id.

​
Project boundaries

OAuth agents may list the connected account’s accessible projects. They must not access unlisted projects, combine customer data, create projects, or inspect billing. A 403 project_forbidden response is final.

​
Safe writing

When writing data:
  • Prefer draft creation before publish actions.
  • Read a post before patching it.
  • Preserve existing user content unless the user clearly asks for replacement.
  • Send only intentional changes in update requests.
  • Do not retry unsafe writes blindly after network failures.

​
Publishing

Publishing can affect a live website. OAuth agents can publish without per-action confirmation only after the user enables Agent Autonomy for the selected project. Credentials, integrations, billing, team access, and project deletion remain human-controlled.

​
Async behavior

Treat 202 Accepted as successful queueing. Continue by polling, reading the affected resource later, or telling the user generation is running.

​
Retry behavior

  • Treat 403 errors as final.
  • Treat 429 errors as final until Retry-After has elapsed, then use exponential backoff.
  • Retry temporary 5xx errors carefully.
  • Do not retry validation errors without changing the request.
  • Do not retry publish calls blindly.

​
Integration readiness

Before publishing, check integrations. If the API returns integration_not_configured or mapping_required, guide the user to configure the integration or save the missing mapping first.