kb Tools
← all tools

Native desktopLocal-firstOpen sourceNo account

Tracelight

See what your coding agents are doing, live, on your own machine.

In design: Being designed. No download yet.

Linux · Windows · macOS

tracelight · 3 sessions
# 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

etsy/studio · timeline
12:04:10  read     products/work-order/product.toml
12:04:12  bash     scripts/fitcheck.sh            2.1 s
12:04:19  edit     brand/forms.typ               +6 -2
12:04:31  bash     uv run studio build all       41 s   ⚠ slow
Mockup of the planned interface, not a screenshot.

The tour

What it does

01

Every session at a glance

Running, waiting for you, or finished: one list for all your agent sessions, with the project, the last action and how long it has been idle.

02

A timeline of tool calls

Each read, edit and shell command in order, with duration and the files it touched. Click one to see the exact input and output.

03

Tokens and cost

Tokens in and out per session and per day, with an estimated cost, so a runaway loop shows up before the bill does.

04

Stuck and waiting alerts

A desktop notification when an agent waits for approval or repeats the same failing step.

05

Replay a session

Scrub through a finished session step by step to understand what happened and why it went wrong.

06

OpenTelemetry export

Export sessions as OpenTelemetry GenAI spans, so the same data can go into Jaeger, Grafana or your own tools.

How it works

Who does what

  1. 01 You Start Tracelight once. It finds the agents' local session logs; nothing to configure.
  2. 02 Agent Your agents work as usual in their terminals and editors.
  3. 03 Tracelight Tracelight reads the session logs as they grow and builds the timeline.
  4. 04 You Glance at the window or the tray icon: who is running, who waits, what it costs.
  5. 05 You When something goes wrong, open the session and replay it step by step.

Privacy

Stays on your machine

Reads local files only

It reads the session logs your agents already write on your disk. It does not intercept traffic or need your API keys.

Sends nothing

No telemetry, no account, no network calls. Export happens only when you ask for it.

Open source

The code will be public so you can check every claim on this page.

FAQ

Questions

Is it available?

Not yet. Tracelight is in design. This page describes what we are building; follow the repository to see it take shape.

Which agents will it support?

First Claude Code and OpenAI Codex CLI, which keep local session logs. Others when their logs are readable.

Why native and not a browser tab?

It should sit in the corner of your screen all day using almost no memory, and work offline. It will be written in Rust.

Is it affiliated with Anthropic or OpenAI?

No. It is an independent project.

Why does this exist?

Abdo is starting research on observability for multi-agent AI systems. Tracelight is the practical side of that work: the tool we want on our own desks.