Cursor is the AI editor you write in. codehere is where delegatedagent work lives. One important honesty up front: codehere doesn't run Cursor — it orchestrates CLI agents. The two compose; they don't compete.
An AI editor. Inline completions, chat, and its own agent mode — all inside the editor. The work and its history live in your IDE.
A home for delegated work. You hand a goal to a CLI agent (Claude Code, Codex, Aider, OpenCode, Gemini), watch it run in a local room, and get an honest report. The unit of work is the goal.
Cursor's agent runs inside Cursor, with Cursor's models. Powerful — and contained to the one tool.
codehere orchestrates five independent CLI agents and lets them take turns on the same goal — Claude does a pass, Codex reviews, Aider fixes — all sharing one room and one project memory.
The code, plus your editor's chat and run history — reviewable inside Cursor.
The code, plus the room: every step recorded, every file change listed, a hash-chained local audit you can verify from the CLI, and a Final Report machine-checked against the git-recorded changes. The run itself is a verifiable artifact.
Cursor is where you write. Nothing about codehere asks you to leave it — the daemon and room run alongside any editor.
codehere also ships an MCP server exposing its four trust tools — and Cursor connects to it (verified live, 2026-06-07): paste the codehere serve --show-config block into Cursor's MCP settings. Local-first, so the daemon must be running.
Local-first, open source, your keys. Invite-only while it's early — request one and we'll set you up.
Request an invite