Not a rival — codehere runsClaude Code. The comparison is about where the work lives afterwards: a terminal session that vanishes, or a local project room that doesn't.
The session lives in your terminal. Close the tab, restart, or log out at the wrong moment and the in-progress work is gone — /resume only reaches your most recent sessions.
Every run in a real project folder lands in a local room that outlives the tab, the terminal, and the daemon. Reopen the room days later — the full thread is there, in order.
A fresh session starts from zero unless you hand-roll continuity: CLAUDE.md notes, TODO breadcrumbs, re-explaining the project state in your first prompt.
Once the project is initialized (codehere init, or an existing AGENTS.md), the room re-feeds project memory at spawn — the agent reads what the last run did before it starts. The next task continues the project instead of rediscovering it.
Claude Code's history is Claude Code's. Hand the work to Codex or Aider and you start over — different tool, different (or no) history, context in your head.
One goal, one room, any of five agents (Claude Code, Codex, Aider, OpenCode, Gemini) taking turns. Each turn shares the same thread and the same project memory.
Terminal scrollback. If you weren't watching, you're trusting the summary the agent wrote about itself.
Every tool step recorded, every file change listed, a hash-chained local audit you can verify, and a Final Report that's checked against what actually shipped.
Local-first, open source, your keys. Invite-only while it's early — request one and we'll set you up.
Request an invite