ChatGPT workspace agents
Repeatable team workflows in ChatGPT and Slack, with admin controls.
Group: Work and business agents
OpenAI documents repeatable workflows in ChatGPT and Slack, with connected tools, schedules, API triggers and workspace access controls.
The decision in plain English
Verify plan eligibility and admin enablement before recommending; do not conflate this with historical ChatGPT agent mode.
Best fit
Repeatable team workflows with configured access and triggers.
Poor fit
Assuming every ChatGPT subscription includes workspace agents.
Documented capabilities
Vendor statements, not verified task outcomes.
Permissions and human control
- End-user authentication runs through each person's account; agent-owned authentication shares a connection. Use a service account with minimum access.
- Write actions default to Always ask; Never ask or Custom settings may be available by app. Review risky send/edit/delete workflows.
- Connector action constraints narrow action inputs, not the data returned.
- Review private, invited and workspace sharing audiences before publication. Slack requires shared app authentication.
Documented controls do not prove that an agent always follows them.
Limits and open questions
- API trigger queues a run with 202 Accepted; the JSON body carries conversation_url (which can be null), and a beta-header flow adds agent_trigger_run_id for status polling. Retrieving the final response through that API remains limited.
- Connector constraints limit requested actions, not returned data.
- Slack requires shared authentication for every app connection; personal shared accounts can expose owner data/actions to other users.
- A separate workspace-agents security overview was referenced, but not established in this review. No product-specific retention or independent certification claim is made.
Privacy and security
Detailed privacy, retention and security controls were not established by the sources in this overview. Verify the current vendor policy before connecting sensitive data. No certification or audit is claimed.
Feature-by-feature: what is documented and what is not
Official sources checked 22 October 2026. A documented feature is a vendor statement, not proof of success.
| Feature | Evidence status | Detail and limit |
|---|---|---|
| Reusable workflows | Documented | Create from a template or builder, choose tools/model, preview with a sample task, then publish. Workspace eligibility and administrator controls apply. Source |
| Tools and apps | Documented | Google Calendar, Drive, Slack, SharePoint, custom MCPs, web search and image generation are documented. Availability depends on workspace enablement. Source |
| End-user versus shared authentication | Documented | End-user connections run as each person; agent-owned connections are shared. A personal shared connection can expose its owner's data and actions to others. OpenAI recommends service accounts and least access. Source |
| Write approvals | Configurable | Apps/connectors default to Always ask. Depending on the app, Never ask or Custom controls may be available. Do not claim every write always asks. Source |
| Action constraints | Not a data filter | Constraints limit what the agent asks a connector to do. They do not filter what the connector returns, so an allowed search may still return sensitive data. Source |
| Schedules and API triggers | Documented with limits | Schedules and API channels are supported. API queues a run with 202 Accepted, no response body or run ID; the agent response cannot currently be retrieved through that API. Source |
| Slack channels | Shared connections required | All app connections must use shared authentication for Slack. Choose every-message or mention-only responses; review the channel audience and credentials before deployment. Source |
| Business usage and credits | Plan-dependent | Current Business billing docs describe included seat usage followed by workspace credits for eligible additional usage and spend controls. No universal per-agent run price is claimed. Source |
How to start with ChatGPT workspace agents narrowly
- Confirm Agents is available and enabled in the intended workspace.
- Build one narrow workflow and keep it private while previewing a sample task.
- Choose end-user or service-account shared authentication deliberately; keep write approval on for risky actions.
- Review outputs and access before publishing, adding schedules or Slack. Confirm eligible usage and spend controls with the workspace owner.
Official product/setup source. These are documentation-based steps, not a hands-on setup test.
Common questions about ChatGPT workspace agents
What is ChatGPT workspace agents for?
Repeatable team workflows with configured access and triggers. The vendor describes it as: OpenAI documents repeatable workflows in ChatGPT and Slack, with connected tools, schedules, API triggers and workspace access controls.
Who should look elsewhere?
Assuming every ChatGPT subscription includes workspace agents.
How much does ChatGPT workspace agents cost?
OpenAI's credit rate card for Business/Enterprise flexible pricing bills Workspace Agents by tokens: GPT-5.6 at 100 credits per 1M input, 10 per 1M cached input, 500 per 1M output; OpenAI says a typical end-to-end run uses 5-25 credits. Workspace Agents draw from the same agentic usage and credit pool as Codex, ChatGPT Work, Excel and PowerPoint; standard Business seats include usage from that pool, with flexible pricing extending limits.
What can it access and what needs my approval?
End-user authentication runs through each person's account; agent-owned authentication shares a connection. Use a service account with minimum access. Write actions default to Always ask; Never ask or Custom settings may be available by app. Review risky send/edit/delete workflows. Connector action constraints narrow action inputs, not the data returned. Review private, invited and workspace sharing audiences before publication. Slack requires shared app authentication.
What happens to my data?
We did not establish a vendor statement on retention or training in the sources we read. Ask the vendor before connecting sensitive data.
Has AgentComparison tested it?
No. This record is built from the vendor's own documentation, checked 2026-10-22. Vendor claims are attributed, not confirmed.
How to evaluate ChatGPT workspace agents yourself
- Pick one low-risk task that matches its documented capabilities: schedules and api triggers.
- Write the expected output and what counts as failure before you start.
- Connect the minimum access needed, and check approval settings first.
- Run it twice and compare, then review every action it took.
- Check cost after any free allowance. Unknown price is not free.
Related
Documented use cases: Repeatable team and business workflows.
- Microsoft Copilot Studio (platform): Teams building business-data agents for several channels.
- Salesforce Agentforce (platform): Enterprise agents connected to Salesforce workflows.
- AI agent use cases by job
- Privacy and approval checklist
- Compare two agents side by side
- Latest agent updates
Sources and history
- https://help.openai.com/en/articles/20001143-chatgpt-workspace-agents-for-enterprise-and-business · official · read 22 October 2026
- https://help.openai.com/en/articles/11481834 · official · read 22 October 2026
- https://help.openai.com/en/articles/20001155 · official · read 22 October 2026
Change history
- 2026-10-10: Expanded source-linked feature evidence and narrow setup; clarified access, permissions, costs and unresolved limits. Documentation research only, not a product test.
- 2026-10-01: Initial documentation research. No hands-on testing.
Documentation overview expanded with current pricing/access sources. No hands-on results claimed.
Report a correction