Team Mode

GPT Agents and Team Mode

Team Mode coordinates two persistent AI engineers with planning, distinct task ownership, path claims, cross-review and evidence-based completion.

Coordination before parallelism

Both engineers join before planning begins. Each posts a plan, takes a distinct implementation task, and confirms the split before RepoTunnel unlocks parallel implementation.

  • One primary owner per implementation task
  • One active implementation task per agent
  • Task scoped path claims
  • Cross-review by the other engineer
  • Explicit handoff when ownership changes

Persistent team

Completing one work request does not end the team. The team remains attached to the project until the user explicitly ends it in the desktop app.

Start from GPT Agents

Select the approved project in GPT Agents, enter a product goal and concrete success criteria, and create the Team. The app supplies separate Engineer A and B kickoff prompts for their external AI chats.

Both engineers join before the planning/split barrier unlocks implementation. The desktop screen shows their workflow stage, current tasks and verification evidence.

Continue without recreating the team

A completed work request returns the persistent Team to Ready. Give the next request to an attached AI chat so it can register the new work cycle in the same Team.

Pause blocks Team work temporarily; End Team is the user-owned action that permanently detaches the two identities. MCP cannot force an arbitrary completely idle web chat to begin generating.

Start typing to search.