Calandria documentation

Calandria runs many Claude Code and Codex sessions in parallel across your projects, from one screen. Every task gets its own agent session in its own git worktree; turns run on the server and survive the browser closing.

These pages are the repo's docs/ directory rendered in place, so they always match the branch they were written on. The README is the short version, and every page here has an “Edit page” link to its Markdown source.

Get it running

• Installation — Requirements, the two-command start, and running from source.
• Self-hosting — Docker, tunnels, auth, backups, upgrades, and every environment variable.
• Windows — Native Windows and WSL2, and which one to pick.
• Troubleshooting — First-incident runbooks: DB corruption, disk fill, headless re-auth, boot failures.

What it does

• Features — Tasks, worktrees, diff review, merge and PRs, the board, schedules, runbooks.
• Agents — Claude Code and OpenAI Codex: what each driver supports and how they differ.
• Services — Long-running processes a task can start, and how they get a public URL.
• Insights — Spend, tokens, tasks shipped, and lines merged.
• Document collaboration — Editing documents with an agent instead of a codebase.

Under the hood

• Architecture — Processes, the turn lifecycle, the agent-driver seam, where data lives.
• Desktop app — The cross-platform desktop spike and what it concluded.
• Community — Where to ask, propose, report, and contribute.