Drop claude, codex, gemini or opencode into real workspaces and run them in parallel — across git worktrees and remote SSH hosts. Watch every diff live. Track work with a Kanban board that's just markdown on disk, which agents read and update over MCP.
Linux x86_64 · no build toolchain · prebuilt from the latest GitHub Release
One infinite canvas per repo. Every tile is live — agent terminals, diffs that update as the agent edits, a file tree, an editor, an issues board, even a real browser.
A bare terminal throws away everything around the work — which task, what changed, the session when you close the window. hivemind makes the project the unit of work: one canvas per repo, several agents in parallel, each scoped to its own directory, branch and issues.
Full PTY tiles for claude, codex, gemini, opencode — not chat boxes. Sessions survive the window and resume after a reboot.
Bind a frame to a local repo, a git worktree, or a remote SSH host. Run agents side by side, each isolated — no conflicts.
A diff tile next to each terminal updates as the agent edits — split or unified, with line comments you send straight back.
catIssues, acceptance criteria, cycles and an activity log as plain markdown under .hivemind/. No database, no API.
A small MCP server lets an agent pick up an issue, flip its state, tick acceptance criteria, and comment its own progress.
Everything on your machine. No telemetry, no cloud account, no SDK lock-in — agents use your existing CLI login.
Real screens from the app — board, diff, explorer, and issue capture.




Yes — hivemind is an infinite canvas for Claude Code and other AI coding agents. Each agent runs in a real terminal tile, alongside diffs, a file tree, an editor, an issues board and a browser, on one zoomable canvas per project.
Bind each frame to a local repo, a git worktree, or a remote SSH host, then spawn an agent per frame. They work side by side — each scoped to its own directory, branch and issues — and you watch every diff update live.
Same idea, many agents on one screen, but with the structure agents need: a diff next to each terminal, a markdown issues board they update over MCP, worktrees as nested frames, and sessions that resume after a reboot.
claude, codex, gemini and opencode out of the box; any terminal-native coding agent can be added via the registry.
Yes. Everything runs locally, the data is files on your disk, there's no telemetry, and agents use your existing CLI login — no extra API keys.