AgentXRay — read the session logs your AI coding agents already write

AgentXRay is a local-first web dashboard that reads and visualizes the session logs your AI coding agents already write to disk, for developers who want to see what those agents actually did. It is a single Node.js + Express server plus this React UI, pointed at the log directories your existing agent CLIs already use. It is a reader, not a tracing SDK: there is nothing to instrument, no API key, no hosted backend, and no data leaves the machine it runs on.

Agent log formats it reads

Seven on-disk session-log formats have a registered adapter and are normalized into a single model, so one interface covers every agent you run:

Six of the seven store JSONL; Hermes stores SQLite. The authoritative list is the PLATFORMS registry in lib/platforms/index.js, and adding a format takes one adapter file plus one line in that table.

The per-turn ledger

Each session is replayed as a turn-by-turn transcript. Tool calls are paired with the results they returned, so any call expands to show its arguments and its output. Token usage and cost are summed per user turn, and a turn's elapsed time is split into model inference versus tool execution — so a slow turn tells you whether the model was thinking or a tool was blocking. Around that sit aggregate analytics over tool calls, error clusters and token spend, the real human prompts recovered from the logs (including ones from sessions an agent's own cleanup has already deleted), full-text search across every platform at once, and a prompt library that installs the keepers as native slash commands.

Install

Run it from npm without installing anything, then open the port it prints:

npx @alloevil/agent-xray
# then open http://localhost:3800

Requires Node.js 22.13 or newer. MIT licensed. This GitHub Pages page is a demo build running against synthetic sample logs committed in the repository — it contains no real sessions, so treat it as a UI tour. Run the command above to point it at your own.