Lloom

Claude Code terminals across projects and accounts, in one window. Each session has two views of the same turn — the raw terminal, and a chat view that lays out tool calls, diffs, and thinking blocks readably.

Skills, rules, hooks, agents, output styles, and CLAUDE.md files, edited in one place rather than scattered directories — globally or per project. Guidance can be set per model, and applies when a session starts.

Cost and token statistics per session, project, model, and account, computed from the session files on your machine — alongside your live 5-hour and weekly limits.

Reads your session history — turns, tool calls, retries, idle gaps — and reports where sessions go wrong. When a finding has a concrete fix, a sandboxed background job drafts it and shows the diff; nothing is written until you apply it.

A local MCP server lets a running session use Lloom itself: spawn worker sessions in other projects, send follow-ups, check status, pull results. Long tasks can spread across sessions — each with its own context and its project's own configuration.

Past sessions are searchable from inside a session, so questions like "how did we solve this last time" resolve against your actual transcripts.

Sessions can be started and steered from a phone over Tailscale, then picked up at the desk.