Now in private beta

Parallel agents. Real PRs. Your review.

Tell AutoForge what to build. It splits the work across multiple AI coding agents running in parallel — each one opening a GitHub PR you review and merge.

See how it works
autoforge — run
$ autoforge "Build a SaaS dashboard with auth, billing, and analytics"
✓ Planning... 3 work bundles, 6 tasks
→ Spawning agents in parallel...
┌ bundle/core
│ ├─ agent/authPR #201 ✓ done
│ │ wrote: middleware, login flow, session mgmt
│ └─ agent/databasePR #202 ✓ done
│ wrote: schema, migrations, seed data
│
├ bundle/features
│ ├─ agent/billingPR #203 █████░░░ building
│ │ → wiring Stripe webhooks...
│ └─ agent/analyticsPR #204 ██████░░ testing
│ → running integration tests...
│
└ bundle/frontend
├─ agent/dashboardPR #205 ███░░░░░ building
│ → scaffolding React components...
└─ agent/api-routes(queued, waiting on: bundle/core)
$
How it works

From English to merged in three steps.

No prompt engineering degree required. No babysitting agents. No mystery black box.

1

Describe your objective

Write what you want in plain English — a feature, a refactor, a bugfix, or an entire module — and AutoForge breaks it into a parallel task plan, assigning each piece to a specialized AI coding agent.

2

Agents execute in parallel

Multiple AI agents work simultaneously on separate branches, each tackling its assigned task independently — collapsing days of serial work into hours.

3

Review PRs on GitHub

By default, every completed unit of work opens a GitHub pull request — you review diffs, test results, and a plain-English summary before anything ships. Or skip GitHub entirely and let AutoForge build fully autonomously.

Features

Built for how production software actually ships.

Every design decision traces back to one constraint: real code, real PRs, real repositories — not sandboxed demos.

Agents working in parallel

Multiple AI coding agents work on your codebase simultaneously, each on its own branch — so implementation speed is no longer your bottleneck.

GitHub-native from day one

Every agent writes to a real branch in your real repository — branches, commits, pull requests, and CI checks all live in your existing GitHub workflow, with no proprietary storage or lock-in.

Bring your own API keys

Connect your own OpenAI or Anthropic key and AutoForge calls it directly — no markup, no re-billing, every token visible in your provider dashboard.

Multi-tenant by default

Serve your entire team — or your customers — from one AutoForge instance, with workspaces fully isolated at the data and execution layer.

Claude Code + ChatGPT Codex

AutoForge has first-class integration with Claude Code and ChatGPT Codex — the two AI agents that consistently produce reviewable, mergeable output on real codebases.

Predictable task execution

The task plan is computed before any agent runs, stored as a structured plan, and executed in the right order — you can inspect, modify, and re-run any part without starting the whole objective over.

Bring your own keys

Your keys. Your usage. Zero markup.

Connect your own API key — OpenAI for Codex, Anthropic for Claude — and AutoForge uses it directly at your provider's published rates. We charge for orchestration, not compute.

Your API keys are stored encrypted at rest and are never transmitted to AutoForge servers during inference — requests go directly from the agent runtime to the model provider.

No AutoForge account is required to call model APIs. If your key works in the playground, it works in AutoForge.

autoforge.config.ts
// AutoForge config — your keys, your calls
export default {
  providers: {
    anthropic: {
      apiKey: process.env.ANTHROPIC_API_KEY,
      model: "claude-opus-4-5",
    },
    openai: {
      apiKey: process.env.OPENAI_API_KEY,
      model: "codex-mini",
    },
  },
  // Tokens go directly to the provider.
  // AutoForge never intercepts inference.
}
Anthropic (Claude Code)
Supported
OpenAI (ChatGPT Codex)
Supported
Google (Gemini)
Coming soon
Mistral
Coming soon

Integrations

Works with the tools already in your stack.

GitHub
Claude Code
ChatGPT Codex
Vercel
Cloudflare
GitHub Actions
Pricing

Pay for orchestration, not for inference.

Model costs go directly to your provider at their published rates. AutoForge charges for the orchestration layer — the planner, the agent runner, the GitHub integration, and the review workflow.

Starter

Free

No credit card required

For solo developers kicking the tires. Full access to core orchestration on public and private repositories.

  • Up to 5 active runs per month
  • 2 parallel agents per run
  • GitHub PR integration
  • Claude Code + ChatGPT Codex support
  • Community support (Discord)
Join the waitlist
Recommended

Pro

$49/mo

Billed monthly, cancel anytime

For indie hackers and small teams shipping continuously. Unlimited runs, higher parallelism, priority execution.

  • Unlimited runs
  • Up to 10 parallel agents per run
  • Priority agent scheduling
  • Task plan inspection and replay
  • Email + priority Discord support
Join the waitlist

Enterprise

Contact us

For engineering teams that need SSO, audit logs, custom model routing, and SLA-backed uptime.

  • SSO / SAML
  • Audit logs and compliance exports
  • Custom model routing and quotas
  • Dedicated support and onboarding
Get in touch

All plans include BYOK (bring your own API keys). Model inference costs are always billed directly by your provider — AutoForge never adds a margin.

FAQ

Questions developers actually ask.

If something's still unclear after this, find us on Discord.

Ready to ship faster?

Join the waitlist and be first to run parallel AI agents on your real GitHub repositories.

No credit card required. Free tier available at launch.