A team of AI agents that takes an idea from research to launch β built on opencode, with no framework and no extra dependencies.
Each arrow is a gate: the next stage starts after your approval.
OpenCode Team is a software team you install once and use in every project. It defines:
- 14 agent roles β analyst, pm, po, ux, brand, architect, fe, be, qa, devops, security, marketing, docs, and an orchestrator that routes work between them
- 12 pipeline stages β configured in
team.yaml, not hardcoded, so you can add, remove, or reorder stages freely - 14 skills β 12 stage methodologies plus
team-memoryandteam-discipline, all local underskills/team-*; no external skill is cloned or required - 19 slash commands β run a single stage directly, or drive the whole pipeline
The orchestrator is the default agent. Open opencode anywhere, describe an idea, and the team takes it stage by stage, asking for your approval at each gate.
Requires OpenCode 2.x (tested on 2.2.5). This release targets OpenCode 2.x only. OpenCode 1 and 2 cannot coexist; OpenCode 1.x users: use Release v1.0.0 (@opencodeteam/cli@1.0.0).
Supported on macOS, Linux, and native Windows (Windows PowerShell 5.1 or PowerShell 7; verified with OpenCode 2.2.5). WSL is untested.
Full install instructions β both install methods, copy vs symlink, verify/uninstall/update, and Windows steps: docs/install.md.
Quick reference:
| Path | Target |
|---|---|
~/.config/opencode/agents |
opencode-team/agents |
~/.config/opencode/commands |
opencode-team/commands |
~/.config/opencode/skills/team-* |
opencode-team/skills/team-* |
~/.config/opencode/team.yaml |
opencode-team/team.yaml |
~/.config/opencode/team/AGENTS.md |
opencode-team/team/AGENTS.md |
The repo stays in git; the config directory holds symlinks. Edit a file in the repo and it takes effect immediately.
Neither installer modifies opencode.json. If default_agent is not set to orchestrator, it prints the recommended snippet and leaves the file alone. On Windows the default directory is %USERPROFILE%\.config\opencode; both installers honor OPENCODE_CONFIG_DIR.
Open opencode in any project and describe what you want:
cd ~/projects/my-new-app
opencodeFull usage guide, single-stage commands and pipeline control: docs/usage.md.
14 agent roles, each owning a pipeline stage and writing a .team/ artifact. Full role β agent file β artifact mapping: docs/reference/agents.md.
All agents are mode: all, reachable three ways: @agent-name to mention, Shift+Tab (or /agents) to switch, and the orchestrator can launch one as a subagent via the subagent tool.
No agent pins a model: each one runs on the session's active model, so any provider configured in OpenCode works. To pin a model for a single agent, add a model: field to its frontmatter β the filename supplies the agent id, so no name: field is needed:
---
description: "Marketing: positioning, copywriting, launch strategy, go-to-market"
model: opencode/nemotron-3-ultra-free
---Free-tier OpenCode Zen picks used while testing: deepseek-v4.1-flash-free (code, image input), beam-free (fast, text-only), nemotron-3-ultra-free (Turkish copy).
Every stage writes a numbered markdown artifact in .team/ (01-research.md β¦ 12-launch.md). The build stage also writes intermediate 07-build-fe.md and 07-build-be.md, consolidated into 07-build.md.
Full .team/, memory and runs directory trees: docs/output.md.
OpenCode Team keeps a small, project-local memory in .team/memory/ so the orchestrator and specialist agents have continuity across sessions.
Full protocol β what each agent does, commands (/memory, /remember), and rules: docs/memory.md.
The orchestrator records each run and task attempt in .team/runs/ so a compaction or restart does not repeat completed work or skip unfinished work.
Full protocol β run.json, events/, approvals/, and how the orchestrator reconciles attempts at session start: docs/runs.md.
team.yaml is configuration, not code. Change it and the pipeline changes β no code changes needed.
Full reference β profiles, stage fields, adding a stage, adding a role: docs/customization.md.
The team's working language follows you. Agent instructions, code, code comments and technical documentation are in English; everything else β PRDs, research, UX rationale, marketing copy β is written in whichever language you are using. Ask for a specific language per artifact any time and it will be honoured.
- opencode 2.x (tested on 2.2.5)
- Any provider configured in OpenCode β OpenCode Zen or your own. Agents inherit the session's active model.
OpenCode Team is a community-driven project and is not built by, affiliated with, or endorsed by OpenCode (opencode.ai) or anomalyco/opencode. "OpenCode" is a trademark of Anomaly. This project uses the OpenCode CLI as its underlying agent framework.
The team skills are local. team-build, team-backlog, team-brand and team-discipline are adapted from MIT-licensed upstream skills; see THIRD_PARTY_NOTICES.md for attribution and licence text.
Contributions are welcome! See CONTRIBUTING.md for guidelines.
β If OpenCode Team makes your AI agent workflow easier, consider starring the repo. It helps others discover the project and shows that the community finds it useful.
MIT. See LICENSE.