Show HN: I built a tool to un-dumb Claude Code's CLI output (Local Log Viewer)

Finally see what your AI actually did — devs cheer, grumble, and call for alternatives

TLDR: A new desktop viewer exposes every move Claude Code makes by reading local logs—no setup, no API keys. Devs love the transparency but gripe about config bloat, worry it’ll break if logs change, and some say skip Claude entirely and use alternatives instead, making this a broader fight over AI “black boxes.”

Hacker News lit up after a developer dropped a desktop viewer that turns Claude Code’s blurry mystery summaries into a full, step‑by‑step playback. The app reads the logs already on your computer and reconstructs what the AI actually did — which files it touched, what it searched, how many “brain” tokens it spent — with no setup or API keys. In plain English: it’s an X‑ray for your AI assistant.

The crowd’s mood? Split right down the middle. One camp is popping popcorn at the transparency win, calling it “finally, receipts.” Another camp is fuming that this tool needs to exist at all. As khoury put it, we shouldn’t need third‑party add‑ons just to see “how much gas is left in the tank.” A separate roast came from gregoriol, who peeked at the repo and groaned about “>20 config files” for a simple viewer — the dev‑era equivalent of bringing a suitcase to a day trip.

Then came the plot twists. Skeptics like 6LLvveMx worried the whole thing could break the moment Claude’s log format changes. Meanwhile, miroljub waved the “just use something else” flag, pushing alternatives like Opencode and pi.dev. And eurekin chimed in from the corporate trenches, saying Copilot in IntelliJ hides outputs too — which makes this feel less like a Claude problem and more like an industry trend toward black‑box agents.

Bottom line: CSI: ~/.claude is here, and the community can’t decide whether to celebrate the transparency or side‑eye the entire ecosystem for hiding the dashboard in the first place.

Key Points

  • claude-devtools is a desktop app that reconstructs detailed Claude Code session activity from local logs in ~/.claude/.
  • It requires no API keys or configuration and does not modify or wrap Claude Code.
  • The tool restores visibility into file paths, tool calls, diffs, searches, and token consumption via a visual interface.
  • It supports macOS (Apple Silicon and Intel), Linux (AppImage/.deb/.rpm/.pacman), and Windows (.exe), with simple installers.
  • Visible Context Reconstruction estimates per-turn token attribution, including CLAUDE.md injections, skills, mentions, tool I/O, and user prompts.

Hottest takes

we have to install 3rd party packages/implement it ourselfs just to see how much gas is left in the tank basically — khoury
>20 config files, just for a simple tool, something is terribly in todays development — gregoriol
Why not just use something sane by default? — miroljub
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.