Small tools for the files AI coding tools leave behind.
Every one of these runs entirely in your browser. Nothing is uploaded, nothing is stored, and there is no account — the same reason holds as for the app itself: none of this work needs a server, so none of it gets one.
Transcripts
Claude Code and Codex write one JSON Lines file per session. These read them: as records, as another format, or as something a person can actually read.
Drop a .jsonl file — or paste it — and read it as formatted, collapsible records. Broken lines are reported with their line number instead of failing the whole file.
JSONL ConverterFour conversions, all in the browser: JSONL to a JSON array, a JSON array to JSONL, JSONL to CSV, and CSV to JSONL.
Transcript to MarkdownDrop a Claude Code or Codex .jsonl session and get clean Markdown: speakers as headings, code fenced, tool calls kept or dropped as you choose.
Tokens
What a piece of text costs, and how much of a context window it takes up.
Config files
The files that decide how your agent behaves and which servers it can reach — checked before they cost you a session.
Fourteen checks over the instructions file your coding agent reads at the start of every session: contradictions, dead weight, machine-specific paths, and secrets that should never have been committed.
MCP Config ValidatorPaste .mcp.json or claude_desktop_config.json and get the JSON error with its line and column, the schema problems underneath it, and the small mistakes that fail silently.
They are offcuts. Building Inventory meant writing readers for the formats six different AI coding tools use, which meant learning where those files live, how they break, and which parts of them are worth reading. Publishing the small pieces of that work costs nothing extra and saves somebody else an afternoon.
Where each tool stores its history, and what breaks when you try to read it, is written down in full on the storage reference.