Teampilot

TeamPilot

[简体中文](README.zh.md) · [Development guide](docs/DEVELOPMENT.md) · Architecture & AI: [AGENTS.md](AGENTS.md)

**TeamPilot** is a desktop client based on terminal AI Agent Team. Its centerpiece is **team capabilities**: assign **a model — and even a different CLI — per member** for tiered collaboration (save tokens, implement fast, review accurately), plus roles, prompts, skills, and plugins in the GUI—then launch **one embedded terminal per member** that talks to agents through Claude Code, Codex, opencode, cursor, or flashskyai, locally or over SSH. **Workspaces** and **sessions** attach that team (or a solo personal identity) to a repo folder and conversation.

 

Two ways to work

Mode	When	What you get
Simple mode (personal)	You just want one agent in a repo	Skip the team roster—launch a single CLI and start chatting. No member list to build.
Team mode	Multi-agent / tiered or mixed-CLI collaboration	Define a team once, then run one terminal per member in parallel. This is TeamPilot's centerpiece (below).

**Simple mode is not stripped-down.** A personal workspace still gives you the full config surface—you just skip the multi-member roster:

• Multiple CLIs and models, switchable. Each CLI (Claude Code, Codex, opencode, cursor, flashskyai) keeps its own provider + model + reasoning-effort on your personal launch identity, so you can configure several models per tool and switch the active CLI/model without retuning globally—the same tiering benefit as teams, for a single agent.
• CLI presets: Save frequent CLI + model combos as named presets and switch with one click.
• Per-identity agent plus per-workspace skills, plugins, MCP, and extensions—the personal identity carries prompt and model tiering; capability bindings mount on the workspace. Skills / MCP / plugins are installed once in the