ForgeAI is in public beta

ForgeAI is now open for real projects. This first public build is a beta: the platform works end to end, and we are hardening it in the open ahead of general availability. You can create an account today — everything is free while billing is being finalized.

What ForgeAI is

AI coding agents are powerful but opaque: you prompt, you wait, you hope. ForgeAI makes AI-driven development visible, guided, and structured. It pairs two halves:

• The cockpit — the web app where your team runs methodology-guided workflows, reviews the artifacts they produce, and watches agents execute tasks on a live Kanban board.
• The forge CLI — the execution engine that runs on your machine, published on npm under the Apache-2.0 license. It spawns the agents in local Docker containers, manages git branches and worktrees for them, and streams their progress back to the cockpit over a WebSocket.

The split matters: your repository and your AI credentials never leave your machine. The cockpit orchestrates and displays; the work happens on your hardware, with your own AI account — Claude Code, Codex, OpenCode, or Kilo. ForgeAI adds no per-token AI cost on top.

From idea to shipped code

A project in ForgeAI follows a structured path:

1. Plan with a BMAD workflow. A guided wizard walks you through the BMAD methodology step by step — an agent drafts each deliverable (brief, PRD, architecture), you review, request changes, or approve. Conversations are encrypted at rest with a per-workflow AES-256-GCM key.
2. Generate tasks. A completed workflow turns its plan into grouped tasks with dependencies, ready to execute.
3. Watch agents work. Connected CLIs pick up tasks, run agents in Docker on the developers' machines, and the board updates live — statuses, activity, and the agent's session streaming in real time. When an agent needs a human decision, it asks and waits.
4. Ship a release. Workflow branches are merged back and a versioned release tags the result.

What the beta includes

• BMAD workflows with the step-by-step wizard, artifact review, and encrypted conversations
• Automatic task generation with groups and dependencies
• The live board with real-time agent streaming and human-in-the-loop responses
• Releases with agent-driven merge-back
• Organizations, roles, and email invitations

Agents run on the engine of your choice: Claude Code (the default and most battle-tested), Codex, OpenCode, or Kilo Code with its 500+ model gateway — pick one per device with forge config set agent-type or from the cockpit's device settings, each authenticated with your own subscription or API key.

And the honest list of current limits: the CLI is a prerelease (npm install -g @getforgeai/cli@beta), the agent image targets Apple Silicon first, and billing is inactive — the beta is free. The documentation states, page by page, exactly what works and what is still being wired.

Getting started

The Quickstart takes you from a fresh account to your first workflow: sign up, create a project, install the CLI, authenticate it with a device code, connect your machine, and go. The prerequisites are a ForgeAI account, Node.js and pnpm, git, Docker, credentials for your agent engine (a Claude Code login by default), and a git repository for your project.

We publish every change to the platform in the changelog. If something breaks, the troubleshooting guide covers the failure modes we know about — and the feedback button in the cockpit reaches us directly.