seedeep
A dashboard that shows you exactly what Claude Code is doing in real time — every file it touches, every subagent it spawns, and how much each step is costing you in tokens.
🔗 Visit seedeep
Description
AI coding agents like Claude Code do a lot of work behind the scenes — reading files, calling tools, sometimes spawning their own subagents — and most of that happens invisibly in a terminal scroll you can't easily follow, especially once a session runs for a while and you start wondering where your token budget went. Seedeep was built by a developer who literally couldn't see what Claude Code was doing, so instead of squinting at logs, they built a live visual dashboard for it.
It runs entirely on your machine — read-only, no proxying your API calls through a third party, no session data leaving your computer — and shows every model call, tool execution, and subagent operation as it happens, each with its own token/cost/latency breakdown. There's a menu-bar tray client for keeping an eye on things in the background, search across past sessions, and a "waste scoring" feature meant to flag turns that burned tokens without much useful output. It's open source under MIT, actively maintained (174 commits), and ships as both an npm package and standalone binaries for macOS, Linux, and Windows.
💬 Our review
The short version: seedeep fills a real gap — Claude Code and similar coding agents give you almost no visibility into cost and behavior by default — and the fact that it's local-only and read-only is the right architectural choice for something watching your AI coding sessions, but it's a young, single-repo project without the broader observability-platform features some teams may eventually want.
Compared to just running `ccusage`-style CLI cost trackers, seedeep's live visual dashboard and subagent-level breakdown is a meaningfully richer view, especially once you're running multi-agent workflows where a single command can spawn several subagents with their own token budgets. Compared to general LLM observability platforms like LangSmith or Helicone, which are built for production applications with many users and often require routing calls through their infrastructure, seedeep is narrower in scope (specifically Claude Code) but wins on privacy and simplicity — nothing to configure server-side, nothing leaving your machine. If you're a heavy Claude Code user who's ever wondered where your tokens went, this is a low-risk, free way to find out.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Free and open source (MIT license), distributed via npm and standalone binaries.
Pros
Live, visual view of every Claude Code call and subagent
Runs entirely locally — read-only, no proxying, no data leaves the machine
Per-call cost, token, and latency breakdown
Menu-bar tray client for background monitoring
Free, open source, actively maintained
Cons
Narrowly scoped to Claude Code, not a general LLM observability platform
Young project without a large community yet
No hosted/team-shared dashboard — it's a local tool
Windows/Linux testing acknowledged as lighter than macOS