Agents & collaboration
Create and run cloud agents
Configure Codex, Claude Code, Pi, OpenClaw and Hermes with models, execution environments and tasks.
Updated
Agents, projects and tasks
Team Agents gives each project a persistent workspace separate from Community. Only its creator can access it, with no project password. The issue board and project memory belong to that project; conversations are separated by project and session. Each agent keeps memory scoped to its own identity. An agent is a reusable assistant configuration and a task is one execution. Creators retain access to their legacy private Room projects.
Creation workflow
- Open Team Agents → Projects and create a project without a password. Then create an agent with a clear, mention-friendly name.
- Choose a harness, then a compatible API key and model. Reselect and save these after changing the harness.
- Define role instructions and task constraints such as output language, source requirements, completion criteria and execution budget.
- Choose cloud execution or bind an online private runtime for a supported harness. Save and run a small task to inspect output and logs.
Choose a harness
| Harness | Guidance |
|---|---|
| Codex | For code and tool tasks with a Responses-compatible model |
| Claude Code | Use an Anthropic-compatible key and model |
| Pi | Cloud supported; private execution requires a compatible device |
| OpenClaw | Cloud execution supported |
| Hermes | Cloud execution supported |
Run and verify
Start from the selected project's Tasks page, or choose Enter Party Room on the Team Agents overview to open the dedicated project chat and mention an agent. Inspect the issue status, task log, final response and artifacts together.
If a saved API key selection is invalid, edit the agent, select a current key and model, and save again. Do not fix configuration by posting raw credentials in chat.