Orchestrate local and remote coding agents into automated pipelines.
Claude Code · Codex CLI · Gemini CLI · any command-line agent
中文文档 · Quick start · Concepts · Workflow format · Remote runners · API
iTeam turns the agent CLIs you already have into a team. You define agents (a role + a CLI runtime with its own model and reasoning effort, running locally or on a remote machine), compose them into a workflow (steps with dependencies), and run it. Independent steps execute in parallel, each step can use the output of the steps before it, and you watch everything happen in one place: the pipeline graph, the roadmap, and the live logs.
| Agents | Claude Code, Codex CLI, Gemini CLI, or a custom command (plus a built-in demo agent to try the product before installing anything). Per-agent model, effort (low / medium / high / max), role instructions, working directory, timeout; the form shows the exact command it will run. |
| Local or remote | Local agents run on the server. Remote agents run on any machine that starts the lightweight runner. |
| Workflows | Steps form a DAG. {{inputs.x}} and {{steps.id.output}} pass data between steps. Per-step model/effort override, retries, timeout, continue-on-error; approval steps pause the run until a person approves; a check command (e.g. npm test) decides whether a step really succeeded; a cost limit stops runaway runs. |
| Runs | Pipeline view, stage roadmap + timeline, live per-step logs with readable agent events (tool calls, results), cost and tokens per step, rendered prompt and output, the git diff each step produced, cancel, retry failed steps, rerun from any step, run again, follow-up prompts in the same Claude Code session; browser notifications and a webhook. |
| Import / export | Workflows are plain YAML or JSON files. Built-in templates, import from the UI (agents are created from the file, preview shows what they will run), export any workflow. |
| Quick task | One prompt on one agent, no workflow needed. |
| Pipeline view of a run | Stage roadmap + timeline |
|---|---|
![]() |
![]() |
| Workflow editor | Overview |
|---|---|
![]() |
![]() |
Requirements: Node.js 20+, and at least one agent CLI installed and logged in (claude, codex, or gemini).
git clone https://github.com/ULis3h/iteam.git
cd iteam
npm install
npm run devOpen http://localhost:5173.
- Agents → New agent — name it, pick the CLI, model and effort.
- Workflows → Import — paste
examples/feature-development.yaml(or build one in the editor). - Run — fill in the inputs and watch the pipeline.
Single-port production mode: npm run build && npm start serves the UI and API on port 3000. Docker: docker compose up -d (set ITEAM_TOKEN and your CLI API keys in .env). The server listens on 127.0.0.1 until ITEAM_TOKEN is set.
name: Bug fix
inputs:
- key: repo
required: true
- key: bug
required: true
agents:
- name: Investigator
provider: claude-code
model: sonnet
effort: high
- name: Fixer
provider: codex
model: gpt-5-codex
effort: medium
steps:
- id: reproduce
name: Reproduce
agent: Investigator
prompt: "Repo {{inputs.repo}}. Bug: {{inputs.bug}}. Find the root cause and add a failing test."
- id: fix
name: Fix
agent: Fixer
dependsOn: [reproduce]
retries: 1
prompt: "Repo {{inputs.repo}}. Fix the issue described here and make the test pass:\n{{steps.reproduce.output}}"More in examples/. Full reference: docs/workflow-format.md.
cd runner && npm install
ITEAM_TOKEN=<token> node bin/iteam-runner.js --server http://SERVER:3000 --name my-mac --max-jobs 2The machine appears under Agents → Runners; create an agent with location remote and pick it. See docs/runner.md.
Agents run unattended with auto-approval in the directories you choose, so the API is an admin interface: it binds to localhost until you set ITEAM_TOKEN, runners can use a separate credential, agent processes never see the server's secrets, and env values are masked in the UI and exports. Details in SECURITY.md.
server/ Express + Prisma (SQLite) + Socket.IO — API, scheduler, executors, tests
client/ React + Vite + Tailwind — the web UI
runner/ Node CLI — executes steps on remote machines
docs/ Documentation
examples/ Importable workflow files (also the in-app template gallery)
MIT © ULis3h




