Orchestrate AI agents
like infrastructure

A macOS desktop app for managing, scheduling, and monitoring Claude Code agents through a kanban-style task board — with recursive agent spawning, DAG pipelines, and chat integrations.

Queue
Code review: auth module
immediate · 2m ago
Refactor DB queries
scheduled · in 15m
Running
Run test suite
elapsed 1m 32s
Deploy staging
elapsed 0m 48s
Done
Update dependencies
completed · 5m ago

Built for agent orchestration

Everything you need to run, schedule, and chain AI coding agents at scale.

Agents that Spawn Agents

Any running task can create sub-tasks, build DAG pipelines, and collect results — enabling recursive, multi-agent workflows.

DAG Pipelines

Define task dependencies with automatic cascade execution. If an upstream task fails, all downstream tasks are automatically cancelled.

Flexible Scheduling

Immediate, delayed, one-time datetime, and cron-based recurring tasks. Schedule agents to run exactly when you need them.

Live Output Streaming

Real-time colorized output from Claude Code CLI, streaming NDJSON events as your agents work.

Two-process architecture

Electron spawns a Python backend. React talks REST. No WebSocket needed.

React Frontend
HTTP/JSON :9712
Python Backend
SQLite
Scheduler
Claude CLI

Recursive agent delegation

A single prompt triggers a multi-agent workflow.

User Prompt
Task A: Research
Task B: Analyze
Task D: Report
Task C: Summarize
prompt.txt
# Example: multi-agent research workflow
Research the top 5 competitors in the AI coding space.
For each competitor, spawn a sub-task to analyze their
pricing, features, and developer experience.
Collect all results and produce a final comparison report.

Control from anywhere

Trigger and monitor agents from your favorite messaging platforms.

Telegram

Bot API polling. Easy setup. Send prompts and receive results directly in Telegram.

telegram
/newtask Code review | Review the latest PR in myapp

Slack

Socket Mode. Real-time events. Create tasks and get status updates in channels.

slack
@AgentForge newtask Deploy | Deploy staging branch

Feishu / Lark

WebSocket long-connection. No public IP needed. Send any message to create a task.

feishu
Send any message to the bot to create a task

Full REST API

All endpoints at localhost:9712/api

MethodEndpointDescription
GET/api/tasksList all tasks
GET/api/tasks/:idGet a single task
GET/api/tasks/:id/runsGet run history
GET/api/tasks/:id/outputGet task output
GET/api/tasks/:id/eventsStream output events
POST/api/tasksCreate a task
POST/api/tasks/:id/cancelCancel a task
POST/api/tasks/:id/retryRetry a task
DELETE/api/tasks/:idDelete a task
GET/api/healthHealth check
curl
# Create a new task
curl -X POST http://127.0.0.1:9712/api/tasks \
  -H "Content-Type: application/json" \
  -d '{
    "title": "Review auth module",
    "prompt": "Review the authentication module for security issues",
    "working_dir": "~/projects/myapp",
    "schedule_type": "immediate"
  }'

Get started in minutes

Three steps to your first agent-powered workflow.

1

Download

Grab the latest AgentForge DMG from the releases page. Requires macOS 12+ on Apple Silicon.

2

Install

Drag AgentForge to your Applications folder. The app bundles its own Python backend — no extra setup.

3

Launch

Open AgentForge, point it at your project directory, and start creating tasks for Claude Code agents.

Build from source
# Clone and build
git clone https://github.com/hetaoBackend/agentforge.git
cd agentforge

# Install dependencies
uv sync
cd taskboard-electron && npm install && cd ..

# Start the dev server
cd taskboard-electron && npm start
Copied to clipboard