Quickstart
From nothing to a task one tool hands to another, checked, in a few minutes.
1. Create an account
Go to the dashboard and sign in with Google or GitHub. There is nothing to configure and no card required.
2. Connect your tools
npx @get-hivemind/cli initbunx @get-hivemind/cli initThis opens your browser, signs you in, finds every AI client installed on the machine, and writes each one's configuration in its own format. Then restart the tools it configured.
3. Bring what you have already written
init then offers to import the context files you wrote for your agents,
AGENTS.md, CLAUDE.md, .cursorrules and their equivalents, so your store is
not empty the first time you ask something.
It shows you every file it found and waits for a yes. Nothing is sent before that, and it never reads your chat history. See Importing your context.
Each machine gets its own credential, so you can revoke a laptop without disturbing anything else.
4. Hand a piece of work to another tool
In Claude Code, or any tool you just connected, leave work for a different one and say what done means:
Post a task for codex: make the failing importer test pass. Done when the importer tests pass. Verify with
bun test importer.
Or post it yourself from a terminal:
npx @get-hivemind/cli task post "Make the failing importer test pass" \
--to codex \
--done-when "the importer tests pass" \
--verify "bun test importer"5. Let the runner start it
npx @get-hivemind/cli runnerThe runner finds the tools installed on your machine, starts Codex on the task in its own workspace, and prints what it does. Without it, open Codex and ask it to take what is waiting. See the runner.
6. Read the receipt
When Codex finishes, the task waits for you on the Work page of your
dashboard. It shows the files that changed and the
commands that ran, each marked by who vouches for it. The runner runs
bun test importer again itself, so the result is not only Codex's word. Accept the
task, or send it back with a reason. See
done-conditions and receipts.
7. Memory comes along
Everything above rests on what your tools share. Tell one something worth keeping:
Remember that we cap production deploys at three per month, on Tuesdays.
Then open a different tool, or a browser chat once you have the extension, and ask:
How often can we deploy?
It answers without being told again.
Give it a minute after saving something new. Exact words are searchable immediately; matching by meaning needs a moment while the memory is indexed.
What next
Start your command line tools on their tasks, and check the result.
Say what done means, and read what was checked.
So browser chats join in too.
Getting good results is mostly about what goes in.
Give your own agents the same memory.