beafk is a local-first control layer for coding agents.

Run Claude Code, Codex, Grok Build and Kimi Code together without replacing them with API wrappers. The native CLIs run on your own machine, signed in with your own accounts and using the subscriptions you already pay for.

Give one agent a job and let it delegate parts of the work to the others. Follow every run from one thread, answer blockers, review results, and keep sessions running even after you close the tab.

Your code, chats and agent credentials stay on your machine. Your phone connects directly to it over your private network.

One machine. Multiple agents. One place to run the work.

Key points

• Real CLIs, not API wrappers — use the tools the providers actually ship.

• Multi-agent orchestration — let one agent delegate work to Claude Code, Codex, Grok Build or Kimi Code.

• Right agent for each job — use different agents for research, implementation and review instead of forcing one model to do everything.

• Shared MCP servers and skills — configure your tools once and make them available across different CLIs instead of setting up each agent separately.

• Mobile-first control — steer agents, review progress and answer blockers from your phone.

• Persistent sessions — close the tab or lock your phone; the work keeps running on your machine.

• Use your existing subscriptions — no separate pay-as-you-go model bill.

• Local-first — code, chats and credentials stay on infrastructure you control.

• Built for teams too — share machines and control who can access them.