id

Quick Start

Install the ID Agents Dashboard macOS app or the open-source core, then deploy your first team

Quick Start#

ID Agents has two entry points. Choose the one that fits you; both drive the same system (see Introduction).

  • ID Agents Dashboard: the proprietary macOS desktop GUI and the easiest way in.
  • Open-source core: the id-agents manager + CLI. Runs everywhere, scriptable, and is what the Dashboard drives underneath.

Option A: ID Agents Dashboard (macOS)#

  1. Download the proprietary ID Agents Dashboard macOS app from /download.
  2. Open the app on macOS.
  3. Click Connect at the bottom of the sidebar and copy the generated prompt into Claude Code, Codex, or another coding harness that supports skills. That agent installs the admin-control skill and becomes your manager.
  4. Ask the manager to create teams and manage agents. The manager is what deploys and drives them.
  5. Watch the work in the graphical views: Agents, Tasks, Calendar, Heartbeats, News, Library, and Agent profiles.

The Dashboard does not deploy teams itself. You create and manage teams by talking to the manager, and the app shows you what the fleet is doing. The app owns the manager daemon it starts, so you never launch the daemon separately; the coding agent you connect drives that daemon.

The Dashboard runs the core underneath, so everything below (configs, skills, agents) applies whether you drive it from the app or the CLI.

Option B: Open-source core (CLI)#

Prefer the terminal, or not on macOS? Run the core directly.

You can let an AI coding agent do the setup for you: copy the prompt below into Claude Code, Codex CLI, or Cursor CLI:

Find https://github.com/idchain-world/id-agents.git then read and follow the QUICKSTART.md file in the repo.

Manual install (advanced)#

If you'd rather run each step yourself, follow the path below. This is the fresh-install path written out for humans; the prompt above decides whether to refresh an existing checkout or clone a new one.

Prerequisites#

  • Node.js 22 or later
  • Claude Code CLI: install from claude.ai/code and run claude login, and/or
  • OpenAI Codex CLI: install from github.com/openai/codex and run codex login, and/or
  • Cursor CLI: install from cursor.com with curl https://cursor.com/install -fsS | bash and run cursor-agent login

1. Install#

git clone https://github.com/idchain-world/id-agents.git
cd id-agents
npm install

2. Add the Admin Control Skill#

Copy the admin-control skill to your Claude Code project so you can manage agents programmatically:

cp -r /path/to/id-agents/skills/admin-control /your/project/.claude/skills/

Replace the paths with your actual directories.

3. Start the Manager#

cd /path/to/id-agents
mkdir -p ./workspace/{teams,manager,agents,logs}
export AGENT_MANAGER_WORKDIR="$(pwd)/workspace"
export ID_WORKSPACE_DIR="$(pwd)/workspace"
npm run id-agents

This starts the manager daemon on port 4100 with the interactive CLI attached to the same process. The CLI does not bind its own network port. Wait until you see the prompt before continuing.

4. Deploy a Team#

Deploy a team using the admin-control skill or the /remote endpoint on the manager daemon. Ship configs:

  • default: 2-agent Claude starter (coder + researcher). Use this first.
  • idchain: 13-agent team covering the full ID Chain system.
  • apps: app team.
  • personal: personal team.

Deploy default:

curl -s -X POST http://localhost:4100/remote \
  -H "Content-Type: application/json" \
  -d '{"command":"/deploy default"}'

Mixed Claude + Codex + Cursor runtime is not a separate config: the detect-runtimes script at setup decides which runtime each agent uses based on what CLIs you have logged in, so default (or any config) can run mixed if the CLIs you need are installed and authenticated.

Dispatch returns immediately with a query ID:

{ "ok": true, "result": { "queryId": "q_1a2b3c" } }

Poll the reply with ?wait=30 to long-poll up to 30 seconds for a result:

curl -s "http://localhost:4100/query/q_1a2b3c?wait=30"

5. Talk to Your Agents#

List agents:

curl -s -X POST http://localhost:4100/remote \
  -H "Content-Type: application/json" \
  -d '{"command":"/agents"}'

Ask an agent a question:

curl -s -X POST http://localhost:4100/remote \
  -H "Content-Type: application/json" \
  -d '{"command":"/ask coder Introduce yourself and tell me what you can do."}'

6. Use the CLI Directly#

For a richer experience, launch the interactive CLI in your terminal:

cd /path/to/id-agents
npm run id-agents

Type /help to see all available commands. Some useful ones:

/agents                        List all agents
/ask coder <message>           Talk to an agent
/news coder                    Check recent messages
/deploy <config>               Deploy agents from YAML config

7. Create Your Own Team#

Create a YAML config in configs/. The YAML is infrastructure-only; agent personality and role instructions go in template files (see Configuration):

team: my-team

defaults:
  runtime: claude-code-cli
  skills: [identity, inter-agent, catalog, task-discipline]

agents:
  - name: frontend
    workingDirectory: /path/to/frontend-project
    heartbeat: 86400

  - name: backend
    workingDirectory: /path/to/backend-project
    heartbeat: 86400

Then create the team with /deploy my-config in the CLI. /deploy only creates: pointed at a team that already exists it refuses, and there is no override flag.

Editing the config file afterwards does not change the running team. The database is the source of truth, so change the team directly with /model <agent-name> <model> or /delete <agent>, use /diff <team> <config> to see how a team and a file differ, and /export <team> to write a config back out from the live team. See Configuration.

Tip: The task-discipline skill ensures agents use the /tasks lifecycle for all non-trivial work, making team activity auditable. The Dashboard's Tasks view reads this same lifecycle.

Next Steps#