id

Introduction

Multi-agent orchestration software supporting Claude Code CLI, OpenAI Codex, and Cursor CLI runtimes

ID Agents#

ID Agents is multi-agent orchestration software with pluggable runtimes -- currently Claude Code CLI, OpenAI Codex, and Cursor CLI. It enables autonomous AI agents to run as local processes, coordinate as teams, and optionally register onchain for verifiable identity. You can mix runtimes in a single team so some agents run on Claude, others on Codex, and others on Cursor.

Key Features#

  • Local agent processes -- Each agent runs as a local Node.js process managed by the manager
  • Agent communication -- Agents discover and message each other via HTTP
  • Teams -- Organize agents into teams with isolated workspaces
  • Multiple runtimes -- Claude Code CLI, OpenAI Codex, and Cursor CLI -- mix and match within a single team, with runtime-aware template conventions
  • Agent templates -- Define agent personality and role via .claude/agents/ (Claude), .agents/ (Codex), or .cursor/agents/ (Cursor) template directories that overlay skills, hooks, memory, and configuration at deploy time
  • Task discipline -- Always-on task lifecycle (/tasks) for non-trivial work, making every unit of work auditable across the team
  • Onchain identity -- ENS-based identity with public wallet addresses, multi-chain address management, and reverse resolution so your agent name shows up in block explorers
  • Remote API -- Programmatic management via the /remote endpoint
  • TUI monitoring dashboard -- Live terminal dashboard for watching the whole fleet, with per-agent status strip, color-coded news feed, and age-faded cooldown indicators
  • Skills system -- Extensible capabilities including inter-agent communication, admin control, and memory
  • Scheduling -- Agent-driven heartbeats (via HEARTBEAT.md checklists) and calendar events for automated recurring work
  • Org chart -- Define team structure with nested groups and tags so agents know their peers, leads, and coordination points
  • Output convention -- Agents write artifacts to {workingDir}/output/, viewable via /output and /artifact commands

Architecture#

A team has two layers below you: a management layer (a manager agent and a CTO) that sets direction, and a development layer of project-scoped worker agents that do the actual building. You stay at the top as the human reviewer — the manager fans work out, workers report back up.

You

Management layer

Manager Agent (CEO)

Claude

CTO

Codex

Development layer

Front-end

Claude

Proj: Web App

Back-end

Codex

Proj: Web App

Database

Claude

Proj: Web App

Front-end

Codex

Proj: One Page Web

Under the hood, ID Agents uses a manager-worker architecture. The Manager acts as the central coordinator, spawning and tracking agents. Each Worker is an independent local process running a Claude agent with its own REST-AP server.

                    ┌───────────────────────────┐
                    │   Manager Daemon :4100    │
                    │                           │
                    │  Interactive CLI (in      │
                    │  process, no network port)│
                    │  /remote endpoint         │
                    │  /talk-to-manager inbox   │
                    │  /.well-known/restap.json │
                    │  agent-manager-db         │
                    └─────────────┬─────────────┘
                                  │
        ┌─────────────────────────┼─────────────────────────┐
        │                         │                         │
        ▼                         ▼                         ▼
┌───────────────┐         ┌───────────────┐         ┌───────────────┐
│   Agent A     │         │   Agent B     │         │   Agent C     │
│    :4101      │         │    :4102      │         │    :4103      │
│  (CC/Codex)   │         │  (CC/Codex)   │         │  (CC/Codex)   │
└───────────────┘         └───────────────┘         └───────────────┘

Core components:

ComponentSourcePurpose
Managersrc/agent-manager-db.tsDB-backed API, agent registry, orchestration logic, /remote endpoint
Workersrc/claude-agent-server.tsREST-AP server running Claude in each agent process
Local Agent Serversrc/local-agent-server.tsSpawns and manages local agent processes

How It Works#

  1. Start the manager -- npm run id-agents launches the manager daemon on port 4100 (configurable) with the interactive CLI attached to the same process.
  2. Deploy agents -- Use /deploy with a YAML config to spin up one or more agents, or /sync to reconcile a running team with an updated config (preserving unchanged agents). Each agent gets its own port, workspace, and REST-AP endpoints. Agent templates from .claude/agents/ (Claude), .agents/ (Codex), or .cursor/agents/ (Cursor) are overlaid at deploy time.
  3. Communicate -- Send messages to agents via /ask or the REST-AP /talk endpoint. Agents process requests asynchronously and return results through /news.
  4. Coordinate -- Agents talk to each other via /talk-to on their own port (localhost:$ID_AGENT_PORT/talk-to), which is synchronous and blocks until a reply arrives. Agents use the task lifecycle (/tasks) for all non-trivial work, making coordination auditable.
  5. Register onchain -- Optionally register agents on ID Chain for permanent, verifiable ENS-based identity.

Ports and Networking#

ComponentPortDescription
Manager Daemon4100API server, agent registry, manager inbox, REST-AP catalog, and /remote endpoint. The interactive CLI is attached to this process and does not bind its own network port.
Agents4101+Dynamic sequential ports across all teams

License#

ID Agents is licensed under the MIT License. See LICENSE for the full text.

Note: ID Agents switched from GPL-3.0 to the MIT License on March 24, 2026.

Next Steps#