A chat that does the work.
Streaming responses, slash commands, and tool calls you can watch happen. The agent loop plans, executes, and reports back — it answers questions, and it also finishes tasks.
Chat and agents, a vibe-coding studio, parallel job orchestration, deep research, local model serving, email, documents. One workspace you own — in your own private cell in our cloud, on your desktop, or on your servers. Your models, your hardware, your data.
$ download for Windows →Open Fork puts an entire AI-native toolkit behind one sign-in. Every part shares the same models, memory, tools, and crew — and every part runs where you decide.
open fork · gallery — one of fifteen built-in apps
Streaming responses, slash commands, and tool calls you can watch happen. The agent loop plans, executes, and reports back — it answers questions, and it also finishes tasks.
Create named crew members, each with their own persona, provider, and toolset. Put them to work in chat, inside the Builder, on scheduled tasks — even over WhatsApp.
A vibe-coding studio: say what you want and the agent scaffolds it, iterates with you, and renders a live preview beside a real terminal. Publish to an isolated hostname in one click — and roll back just as fast.
A conversational orchestrator that fans work out to parallel agents — each in its own isolated git worktree with a live terminal and a hard budget cap. Every merge is held for your approval. Always.
Kick off a multi-step research run and keep working. Open Fork gathers sources in the background, reads them, and delivers a cited, visual report — not a wall of links.
Cookbook scans your machine and recommends models that actually fit your VRAM — from a catalogue of 270+ — then serves them in one click with vLLM or llama.cpp. On this box, or on a remote GPU over SSH.
Everything below ships in the box — no add-ons, no marketplace tiers.
Living documents you write with the model — then hit Run to preview sandboxed HTML apps and games inline.
One prompt, several models, blind voting. Let the answers compete before you commit.
Private knowledge containers for RAG, persistent memory across conversations, and skills the assistant writes for itself.
Your real inbox over IMAP/SMTP, with AI triage and drafts in your voice.
Recurring agent runs, reminders, and notes that file themselves.
Built-in tools plus any MCP server you connect. Toggle each one, per chat.
Generate images, then finish them in a full editor — no round-trips.
Generate video, then let AI edit it: transcribe, cut, subtitle.
CalDAV and CardDAV, beside your email, visible to your agents.
Themes, density, layout — the workspace bends to you, not the reverse. This page is wearing the default One-Dark theme; it isn't the only one.
macOS and Linux run via self-host today — instructions on the download page.
Connect OpenAI, Anthropic, xAI Grok, and Z.ai through a credential-isolating broker — your keys stay inside the boundary you run. Route to OpenRouter or Groq directly. Or skip APIs entirely and serve vLLM, llama.cpp, or Ollama on your own silicon. Mix and match per chat, per crew member, per task.
Sign in if you have a cell. Ask us for one if you don't. Or take the desktop build and run it tonight.
$ download for Windows →