Build with
Run Claude Code in Nix Agents
Claude Code is one of the CLI agents Nix Agents is built around. It runs in a real terminal pane, in a worktree the app creates for the task, with your existing login.
How it works
You install and authenticate Claude Code the way you already do. Nix Agents finds it on your PATH, starts a session in a fresh worktree of the repository you picked, and gives it a real terminal. From there it is Claude Code — same binary, same configuration, same subscription.
The workspace adds the parts that a single terminal cannot: the worktree is named after the task, the diff is one click away, and the session keeps its place on the board when you start two more.
- One worktree per session, so parallel tasks never share a checkout.
- Real PTY terminals, including on Windows.
- Per-worktree diffs for review before merge.
- Loop mode: re-run the task until your exit condition passes.
A workflow that holds up
Start with one task per worktree and resist the urge to run everything at once. Three sessions is where most people stop being able to read the diffs faster than they are produced.
When a session comes back, review the diff in the app rather than in your editor. The point of the worktree is that "not yet merged" is a real state — you can send it back for another pass without touching your working copy.
Questions
- Does Nix Agents need my Anthropic API key?
- No. Claude Code authenticates itself, exactly as it does outside the app. Nix Agents runs the CLI you installed and never asks for provider credentials.
- Can I keep using my own Claude Code settings and MCP servers?
- Yes. The session runs the same binary with the same configuration files, so project instructions, settings and MCP servers carry over unchanged.
