Now tracking Claude Code & Codex sessions

Know what your AI agents actually cost.

Agent Burndown turns Claude Code, Codex, and OTEL agent telemetry into a burn-rate dashboard — sessions, repos, users, models, tools, tokens, and dollars, in one place. Stop guessing where the spend and the waste are coming from.

25k free events / mo
2 min collector install
No raw prompt capture by default
token_spend.burndown live
Day 1 Day 30
−42% waste eliminated
Actual spend
Ideal burndown
Features

Every agent session, every dollar, accounted for.

Everything a local agent dashboard gives one developer, aggregated across your whole org.

Agent Burndown dashboard showing events, tokens, cost benchmark, live sessions, and collector status

Usage by user, repo & model

Slice every session by who ran it, which repo it touched, and which model handled it — down to a single tool call.

Usage breakdown table toggling between user, repo, and model views with events, tokens, and cost per repo

Token & cost breakdowns

Input, output, and cache tokens rolled up into real dollar figures per session, repo, project, and org.

Projects view with active repos, total tokens, cost benchmark, sessions, and a daily project signals chart

Collector status & health

See which machines are reporting, when they last synced, and get alerted the moment a collector goes dark.

Collectors page with ingest endpoint, collector key management, and per-device status and last-used times

An executive report, ready to send

Cost benchmark, adoption trend, top projects, and data-quality coverage in one brief — print it or save as PDF.

Executive report with cost coverage methodology, cost benchmark, tokens, sessions, active projects, and adoption trend

Claude Code & Codex, out of the box

A lightweight local collector reads OTLP and JSONL agent events and forwards normalized, attributed batches — no central OTEL collector to run.

Metadata by default

We capture counts, durations, tokens, and cost — not your prompts. Raw payload capture is opt-in and redacted, never the default.

Data freshness by plan

Every 4 hours on Free, hourly on Pro and Team, down to 15-minute refresh on Team Plus, and near-real-time by contract for Enterprise.

How it works

Install once. See the whole operation.

Connect your agents, establish a baseline, and turn expensive patterns into a visible burndown.

01

Connect

Add the lightweight collector and forward agent events without shipping source code or prompt contents.

02

Measure

Understand cost, token consumption, runs, and reliability by repo, model, and teammate.

03

Improve

Find expensive patterns, compare models, and catch failures before they compound.

Claude Code
Codex
OTLP / OTEL native
Metadata-first collection

Stop finding out about token spend at the end of the month.

Install the collector, start your agentic coding agent, and watch stats appear in real time. Projects are detected automatically from each session's Git repository or working directory.