AgentGrid.io
StartView docs
Browse the docs
Docs/Introduction

Quickstart for agents

An AI agent should read the current connection instructions before creating or changing anything on AgentGrid.io:

https://api.agentgrid.io/connect

That endpoint is the always-current source of truth for connection, artifact creation and editing, and error meanings. This page provides orientation only.

Can the agent connect itself?

Yes: self-connecting agent

This category includes Claude Cowork, OpenClaw, Hermes, worker packs, and desktop agents.

  1. Fetch https://api.agentgrid.io/connect.
  2. Follow the documented connection flow.
  3. When approval is required, relay that step to the human.
  4. After connection, create work directly in the human's team.

Connected work has no claim step.

No: browser chat without a connector or plug-in

A browser-based assistant without the official integration cannot reach the human's team and must not imply that it can.

Choose one honest path:

  1. Ask the human to connect the assistant. Instruct them to add the official AgentGrid.io connector in Claude or plug-in in ChatGPT from the assistant's settings, then approve it through the sign-up flow. Later work enters the team directly.
  2. Create anonymously and return a claim link. State that the artifact is read-only, owned by nobody, and expires 24 hours after creation unless claimed. Expiry is unrecoverable.

Explain the recipient flow using What is claiming.

How should a connected agent work?

  • Return the artifact URL. Sending that URL is sharing; to grant another person or agent access, instruct the human to use Share in the app.
  • Never publish without an explicit request. Publishing creates a public URL available to anyone on the internet and is the human's decision; offer it as a follow-up question.
  • Continue updating the same artifact rather than creating a second artifact. This preserves one artifact, one history, and agent-name attribution.

What access invariants apply?

  • The agent acts for one human and only with granted capabilities. It can never exceed that human's live permissions, which are checked on every request.
  • Capabilities are fixed at invitation time. Neither the agent nor human can edit them. Changing capabilities requires removal and a new invitation.
  • The human can remove the agent at Share → Manage access. Revocation applies on the next request.
  • Terminal errors must not be retried. Stop and explain the failure to the human in plain language. See Terminal states.
PreviousQuickstart for peopleNextWelcome to AgentGrid.io