FAQ

What kind of app is Collins?

Call it an Agent-First IDE, an AI-Native Workspace, or an Agent Orchestrator — Collins is an opinionated version of what a desktop workspace built around a coding agent should look like: the agent front and center, with your sessions organized around it.

Why did you remove Cursor support?

Can't support the new owner.

Will Collins support other agents?

No. Collins is a tool for Claude specifically, and there are no plans to introduce support for other agents.

Is Collins affiliated with Anthropic?

No. Collins is an independent, unofficial community tool — not affiliated with or endorsed by Anthropic (or any other agent vendor). It's a desktop companion for the claude CLI you already use.

Will it modify my Claude Code sessions?

No. Transcripts under ~/.claude/projects/ are read-only to Collins — names, favorites, emoji, and every other bit of app state live in ~/.config/collins/. The only exceptions are the explicit Move to trash and Delete permanently actions, both behind a confirmation. See How It Works for details.

Does anything leave my machine?

Collins talks to no third parties. Everything goes through the claude CLI and the login you already have: resuming sessions runs claude in a terminal; auto-titling and icon generation run headless claude -p jobs (pre-existing sessions are titled locally, and both have toggles in Preferences); and three things call Anthropic directly with the CLI's own stored token — read-only, never written (when it's found expired, at launch or mid-run, Collins runs one throwaway headless claude -p so the CLI itself refreshes its login): the usage panel, the model pickers (which list the models your login can use, asked for about once a day and cached in ~/.cache/collins in between), and the archive mirror, which archives a session's claude.ai sibling when you archive it here (Archive on claude.ai too, on by default). Pull request features go through the GitHub CLI (gh) and your login there.

What can break when Claude Code updates?

The parts of Collins that read the CLI's undocumented internals — its transcript format, its credentials file, the usage and session-archive endpoints, its agents --json listing, and its trust records. Anthropic can change any of these without notice, and when one moves the feature built on it stops working until Collins catches up; the app is written so that's a blank panel or a skipped step, not a crash. The list is kept in How It Works.

Do I need an API key?

No. If claude works in your terminal, Collins works. Session titles and the usage panel reuse the CLI's existing login — no extra credentials.

Why Linux only?

Because I wasn't happy with any of the existing claude tools for linux, while other platforms get native claude app support.

Why "Collins"?

My wife keeps referring to Claude as Collins by mistake. So now when she asks me if I'm talking to Collins, I can say yes.

See What is Collins? for the fork's full origin story.