← Seveneves Lloom
lloom.app ↗
Seveneves · Portfolio · Desktop App

Lloom

mission control for Claude Code

A desktop control surface for Claude Code power users — real terminal sessions with a structured chat view a toggle away, every config layer managed, and the Refinery that reads your session history and fixes the leaks. Free core, 14-day full trial, Windows & macOS.

● local-first · windows · macos · free core · 14-day full trial
Download at lloom.app → See it on lloom.app
The app, in miniature

One window, the whole workflow

Click a stop on the dial — Lloom's real sidebar carries this same compass dial — to walk the core screens.

Lloom ~/code/your-project
—▢✕
Refinery · vital signs last 14 days of sessions
82/ 100 health
Trajectory Health88
Collaboration Quality92
Context & Cost74
Edit Verification61
CLAUDE.md95
Project Setup80
14 missed verifications — edits applied without a follow-up check → drafts the fix .claude/rules/verify-edits.md
Sessions · one turn, two views
❯ claude "tighten the auth guard" · thinking — the guard double-reads the token; collapse it ⏺ Read src/auth/guard.ts (1–48) ⏺ Grep "verifyToken(" → 3 matches ⏺ Edit src/auth/guard.ts +6 −11 ✓ guard collapsed to a single verify; 11 lines removed
thinkingThe guard double-reads the token — one verify pass is enough.
tool · edit⏺ Edit src/auth/guard.ts  +6 −11
claudeCollapsed the guard to a single verifyToken() call and dropped the redundant branch — 11 lines lighter, same behavior.
Config · six layers, one surface
rules/verify-edits.mdglobal · ~/.claude
1
2
3
4
5
# Verify edits Always run the relevant test or a quick read-back after an Edit before reporting it as done. Never claim a fix works on the strength of the diff alone. scope: all sessions
Usage · every token, accounted for computed on this machine
30-day spend
$148
Tokens
62.4M
Cache hit
73%
Turns
1,284
Code & edits Search & read Tests Other
30-day spendlocal JSONL
Accounts · isolated, never crossed three logins, three configs
P
Personal
max plan · 5-hour + weekly caps · ~/.claude
config A
W
Work
team seat · separate rules & agents
config B
C
Client
scoped project · own prompts & hooks
config C
Toolkit · sessions that spawn sessions MCP orchestration
◆ orchestrator main session
↳ spawn · review the diff sub-agent
↳ spawn · run the test suite sub-agent
↳ spawn · draft release notes sub-agent
⌁ results stream back to the parent via the Lloom MCP server
Reads your real session history, scores six dimensions of workflow health, and drafts the fix.
Rust session parser · local-only · AI-sandboxed (text-in / text-out).
Under the hood

What's it built with?

A native desktop shell over a Rust core. The pieces that do the sensitive work — parsing your sessions, guarding the sandbox, checking your license — are open-source crates.

Tauri v2 Rust React 19 + TypeScript Tailwind v4 Zustand xterm.js Shiki
claude-session-parser↗
Parses Claude Code session JSONL into structured turns — the engine behind Sessions, Usage, and the Refinery.
lloom-guards↗
The sandbox boundary: keeps the AI analysis strictly text-in / text-out so it never touches the filesystem or network.
lloom-auth↗
The licensing layer, split out and auditable: machine fingerprint, license-API client, and a local activation cache.

Local-first by design. Session parsing, usage stats, and the Refinery all run on your machine. The only outbound calls are read-only usage and model lookups made with your own Claude credentials — nothing about your code or sessions leaves the device.

Get Lloom

Free core, with a 14-day full trial of Pro. Pro is a one-time license — three machines, all v1.x updates included. Conversion lives on lloom.app.

Download → Get Pro / Buy a license Visit lloom.app

What is Lloom?

Lloom is a desktop control surface for Claude Code power users, for Windows 10/11 and macOS. It runs real PTY terminal sessions, lets you flip each one to a structured chat view of the same turn, manages every Claude Code config layer as a scope-aware surface, tracks your token spend locally, and includes the Refinery — an engine that scores your workflow health and drafts fixes. The free core is enough to live in; a 14-day trial unlocks the full feature set.

How does it work?

The dial above echoes the compass dial in the app's sidebar. Refinery reads your last 14 days of session history and scores six dimensions — trajectory health, collaboration quality, context & cost, edit verification, CLAUDE.md, project setup — then drafts the fix where it belongs: a CLAUDE.md edit, a .claude/rules/ entry, or a hook. Sessions gives a live terminal a structured chat view of the same turn, one toggle away. Config exposes skills, rules, prompts, agents, hooks, and output styles, global or per-project. Usage computes spend, cache efficiency, and activity mix — all locally.

What's it built with?

A Tauri v2 desktop shell, a Rust core, and a React 19 + TypeScript front end (Tailwind v4, Zustand, xterm.js, Shiki). Several pieces are split out as open-source crates on crates.io: claude-session-parser (session parsing), lloom-guards (the sandbox boundary), and lloom-auth (license activation). The AI analysis is sandboxed to text-in / text-out, so it never reaches your filesystem or network.

What does it cost, and where do I get it?

The core is free. A 14-day trial unlocks the full Pro feature set; after that, Pro is a one-time license (three machines, all v1.x updates) — no subscription. This page is a portfolio summary; downloads and licensing live on lloom.app — download here or buy a license.