Forkbench

Per-job threads with plans, notes, and vaulted keys for Claude Code, Codex, and Gemini CLI.

Last verified:

Visit Forkbench

What is Forkbench?

Forkbench is a macOS desktop application that orchestrates local AI coding agents (Claude, Codex, Gemini CLI) and manages collaborative development tasks through a shared board interface, keeping all code, repositories, and terminal history local on your machine while providing remote access and team collaboration features.

Forkbench pricing

Pricing model: Freemium

30 days free (no card required); app included when trial ends. Full pricing details available at forkbench.com/pricing

Forkbench pros

  • All code, repos, and terminal history remain local on your Mac—no cloud sync of sensitive data
  • Vault system keeps API keys and secrets hidden from agents; never exposed in prompts, commands, or transcripts
  • Agents cannot verify or close their own work—tasks require manual acceptance for accountability
  • Shared board persists across app restarts, reboots, and can sync across multiple Macs with team members
  • Remote 'Talk' feature allows checking on threads and unblocking agents from phone or browser

Forkbench cons

  • macOS only—no Windows or Linux support mentioned
  • Free 30-day trial requires pre-existing AI service subscriptions (Claude, Codex, Gemini)
  • All agents run on your machine's resources, not cloud-based—scales with local hardware

Frequently asked questions about Forkbench

Where does my code and data stay?

Everything—repos, terminal history, and code—remains on your Mac. Only raw tool output and scrollback pass through Forkbench's servers when using remote 'Talk' access, then deleted after a few minutes.

How do agents get access to API keys without seeing them?

Keys are stored in the Vault once and injected directly into commands at runtime. They never appear in prompts, command lines, or transcripts. You can require Touch ID on each use or grant standing access.

Can agents close their own tasks?

No. Agents report work finished, but tasks only count as done when you manually accept them. This prevents agents from verifying their own output.

Can I collaborate with teammates?

Yes. Invite colleagues onto a Thread; they work from their own Mac on their own agents at permission levels you set. All use the same board and notes.

Categories

Use cases

Browse all AI tools on NeedAnAI