Skip to content
ULis3hPublic

About

No description, website, or topics provided.

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

iTeam

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

CI MIT Node 20+ v1.2.0


What it does

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.

Screenshots

Pipeline view of a run Stage roadmap + timeline
Run pipeline Run roadmap
Workflow editor Overview
Workflow editor Overview

Quick start

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 dev

Open http://localhost:5173.

  1. Agents → New agent — name it, pick the CLI, model and effort.
  2. Workflows → Import — paste examples/feature-development.yaml (or build one in the editor).
  3. 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.

A workflow file

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.

Remote machines

cd runner && npm install
ITEAM_TOKEN=<token> node bin/iteam-runner.js --server http://SERVER:3000 --name my-mac --max-jobs 2

The machine appears under Agents → Runners; create an agent with location remote and pick it. See docs/runner.md.

Safety

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.

Project layout

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)

License

MIT © ULis3h

About

No description, website, or topics provided.

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages