Welcome to Hivemind
Your AI tools, working together.
Your AI tools, working together, and checking each other's work.
Ask one tool for a piece of work. Another takes it and hands back a receipt: what it did, and which parts were checked and by whom. You accept it or send it back.
Hivemind sits between the AI tools you use. It passes work from one tool to the next, with one agent holding each task at a time. It lets you say what done means and shows what was checked. The runner starts the command line tools you have installed. And it remembers what you told any of them and keeps where each piece of work stands. Which tools you use is up to you.
From nothing to a task one tool hands to another, checked.
Start your command line tools on their tasks.
Say what done means, and read what was checked.
One command configures every AI client on your machine.
Capture and injection on any chat site.
A research chat, a planner and a coding agent in a line.
How it works
Remembering is the easy half. Hivemind also decides what reaches the model, which is where a memory layer either saves you tokens or quietly costs you them.
It fits a token budget you set
Ask for 4,000 tokens of context and you get the most useful set that fits inside it, not the top ten results, whatever they happen to cost. If you pay per token in your own inference, this is the difference between a bounded and an unbounded bill.
It returns nothing when nothing is relevant
Ask about something your memory has no bearing on and you get an empty payload. A memory layer that always finds something buries the answer in noise and charges you for the privilege.
It orders context to actually be read
Models attend to the start and end of a long context far more than the middle. Retrieved memories are placed accordingly rather than in rank order.
It hands work between agents, not just facts
A handoff is assembled from what you saved, sorted into facts, narrative, decisions, priorities and warnings, so the next tool picks up your reasoning, not just your output.
Where it works
| Editors and CLIs | Claude Code, Claude Desktop, Cursor, VS Code, Windsurf, Zed, Cline, Roo Code, Codex, Gemini CLI, Amp, OpenCode, Witsy |
| Agents that run on their own | Hermes Agent, OpenClaw, and anything else that speaks MCP or HTTP |
| Browser | ChatGPT, Claude, Gemini, Grok, Perplexity, T3, and any other chat site, opt-in per origin |
| Your own code | TypeScript SDK and a REST API |
| Sources | GitHub, Notion, and uploaded documents |
Hivemind is in early access. If something is wrong or missing, say so, that is what this stage is for.