Local-first
Project files, context, and team memory stay on your machine first.
A local-first AI work team
Not another chatbot. OpenTeam organizes different models, roles, and agents into a team that can challenge, review, verify, and deliver real project work together.
- MIT License
- Node 20.19+ / 22.12+
- pnpm 10.x
Why OpenTeam?
AI agents are moving from single bots to working teams.
Many products still treat a team as multiple roles taking turns in a chat. OpenTeam focuses on real project work: constraints, handoffs, reviews, QA, and local workspace context.
Free and open-source
MIT License, no seat pricing, and no platform lock-in.
Multi-model collaboration
Different agents can use different providers and models for the job.
Cross-checking
Planning, research, implementation, review, and QA can constrain each other.
Self-evolving
Preserve project rules, user preferences, workflows, and lessons learned.
Workflow
Make the team argue with the work, not just talk around it.
OpenTeam keeps the roles and evidence visible so a room can move from intent to delivery without losing the reasoning chain.
-
01
Plan
Start with a Team, a Workspace, and acceptance criteria for the room.
-
02
Research
Agents can ask, challenge, and supplement each other with @mention.
-
03
Implement
Specialists work through the real project context instead of a detached prompt.
-
04
Review and QA
Reviewers demand tests, logs, screenshots, and concrete evidence before closure.
Configuration
Teams are configured, not improvised.
Settings expose the same objects the product is built around: Provider, Agent, Team, Team Architect, and Skills.
Provider
CLI path, API Key, Base URL, and default model.
Agent
Role name, Provider, model override, System Prompt, and tags.
Team
Description, workflow prompt, collaboration rules, and acceptance criteria.
Skills
Built-in and workspace-discovered capabilities available to the room.
Roadmap
Team Memory
Preserve project rules, user preferences, and decisions.
Team Playbooks
Reusable workflows for engineering, research, writing, and distribution.
Evidence Trail
Link conclusions to files, commands, sources, and verification results.
Windows desktop
Packaging, signing, and auto-update for Windows users.
Run the AI team from your own machine.
The desktop app runs frontend and backend capabilities inside Electron. It does not expose HTTP service ports to end users, and upgrades do not overwrite local SQLite, workspaces, or Provider configuration.