See what your AI work costs.Find what needs a closer look.

Track estimated spending, growing context, and tool failures across coding sessions. Capture analytics with the desktop app, or connect your SDK to track model-call costs and performance. Compare the data over time in your dashboard.

Already invited to the private alpha?

Open your dashboard
See what you can track
trace-flow.dev/app/agents
Illustrative data
Agent analytics
Where agent work turns into spend
Estimated cost over time
$164.20
-8.4%
Aug 04Aug 11Aug 18Sep 01
Tokens processed
83.4M
Conversations
148
Projected 30d
$182
Per-turn context size
See pressure build before a session runs away
p50 / p90
early turnsconversation depthlate turns
Notable changes
Context above 140k7 sessions
Edit failures11.2%
Cost per review$4.18
What you can use today

Understand where your time and tokens go.

Compare model costs and response times. See which coding sessions accumulate context or run into tool failures. Use the dashboard and analytics tools to investigate what changed.

Coding-agent analytics

Find the sessions worth investigating

The desktop app captures analytics from Claude Code and Codex CLI, plus Cursor on macOS. Sessions are parsed locally, with redacted excerpts for investigation. Full conversation archiving is in development.

Cost and context by conversation depth
See how context and estimated cost grow through a session.
Tool reliability and notable changes
Compare failure rates and shifts by source, model, or repository.
Review and file attention
Connect agent spend to the code and review units it touched.
Desktop app
macOS and Windows. Cursor capture on macOS.
LLM requests

Track spending and performance over time

Capture estimated costs, token usage, latency, and errors as model responses stream. Compare usage over time and inspect captured requests when something goes wrong.

ProviderTokensCost
Anthropic18.2k$0.42
OpenAI9.7k$0.18
Google22.4k$0.11
OpenRouter31.8k$0.09
Illustrative request data
How it works

Keep working. Keep collecting.

Connect supported tools once, then collect data as you use them. Open Trace Flow when you want to understand the cost or investigate a problem.

1

Connect your AI tools

Install the desktop app for coding-agent analytics, or point your existing SDK at Trace Flow for model API capture.

2

Collect while you work

Model responses keep streaming. The desktop app syncs coding-session analytics in the background so you can keep working.

3

Compare your analytics

Filter by provider, model, source, or repository. Compare costs and performance across the work you have captured.

Desktop capture

Keep your coding tools. Add the desktop app.

Capture coding-session analytics without changing your workflow. The desktop app runs in your menu bar or system tray.

Install and sign in

Connect the Trace Flow desktop app to your account. Available for macOS and Windows.

Start capturing

Review the detected sources, then choose Start syncing. Capture analytics from Claude Code and Codex CLI, plus Cursor on macOS.

Leave it running

The app syncs in the background as you work. Open your dashboard to investigate costs and performance, or pause syncing from the tray.

Download and set up the desktop app
API capture

Keep your SDK. Change the route.

The proxy path starts with a base URL and one Trace Flow header.

Before
import OpenAI from 'openai';

const client = new OpenAI({
  apiKey: process.env.OPENAI_API_KEY,
});
After
import OpenAI from 'openai';

const client = new OpenAI({
  apiKey: process.env.OPENAI_API_KEY,
  baseURL: 'https://gateway.trace-flow.dev/openai/v1',
  defaultHeaders: {
    'X-Trace-Flow-Api-Key': process.env.TRACE_FLOW_API_KEY,
  },
});
Today

Analytics today.

Trace Flow keeps analytics from coding sessions and model calls so you can investigate costs and performance over time.

Available today

Cost and performance history

Keep cost and usage trends you can revisit as your tools and habits change. Monthly model usage totals are retained for five years; coding-agent analytics for one year. Individual model traces have shorter, plan-based access windows.

Filter analytics in the dashboard or query them through MCP. Coding-session capture includes redacted excerpts for investigation; it does not upload full transcripts.

Read the retention policy