For people who work with AI agents
Desktop tools
for agent work
Coding agents are fast, but they are hard to watch, feed and steer. We are building small native tools that make that work visible and calm.
Tracelight
See what your coding agents are doing, live, on your own machine.
Read about Tracelight →# mockup of the planned interface SESSION STATE TOOL CALLS TOKENS COST LAST etsy/studio ● running 41 182k $1.92 bash: uv run studio build phd/01-literature ○ waiting 12 38k $0.41 needs your approval (Edit) landing-page ✓ done 27 96k $0.88 finished 4 min ago
All tools
The workbench
Tracelight
A small native desktop app that watches your Claude Code and Codex sessions and shows every session, tool call, token and wait on one timeline. Nothing leaves your computer.
Native desktop · Local-first · Open source · No account
Contextpack
Drag in files and folders, see the token count of each one, drop the noise, and export one clean context bundle with a manifest. For long tasks, reviews and handoffs between agents.
Native desktop · Local-first · Open source
Agentbook
Your CLAUDE.md, AGENTS.md, skills and prompt snippets live in dozens of repositories and drift apart. Agentbook shows them side by side, diffs them and keeps shared rules in sync.
Native desktop · Local-first · Open source
Approve
When several agents run at once, approval prompts hide in different terminals. Approve collects them in a tray inbox: read the request, approve or deny from the keyboard, and keep an audit log.
Tray app · Keyboard-first · Local-first · Open source
How we build
Four rules
Local-first
Your prompts, code and sessions stay on your machine. No account, no telemetry.
Small and native
Desktop apps that start instantly and idle at almost nothing. No Electron.
Open source
Every tool is public, so you can read what it does before you trust it.
Honest pages
Each page says what exists today. Ideas are labelled ideas; mockups are labelled mockups.