Skip to the product

Token accounting for coding agents

See inside your software factory.

Analyze your org's agent sessions to see where tokens go, how teams work with agents, and what to improve.

Get started for free Book a demo

Free for up to 5 engineers.

Works with:Claude CodeCodexCursor

Your AI bill this month (and counting)

$122,480.60
this month

VibeCheck shows you where it goes.

Engineering teams at these companies use Maestro AI

Spend by initiative

Track tokens to outcomes.

See what your agent tokens became. VibeCheck groups sessions by the work they served, so every feature, fix and migration your engineers shipped with their agents carries its own tokens, sessions and spend.

Agent use by initiative

Sessions per day · est. agent spend

  • Billing webhook retry queue

    71 sessions from Oct 12 until it shipped on Oct 25.

    $896 est. agent spend

    2 engineers · 9 PRs

    440M tokens

  • Invoice service migration

    96 sessions from Oct 20 until it shipped on Nov 1.

    $1,240 est. agent spend

    2 engineers · 11 PRs

    617M tokens

  • Self-serve trial onboarding

    182 sessions from Oct 13 until it shipped on Nov 2.

    $2,310 est. agent spend

    3 engineers · 21 PRs

    1.14B tokens

Cost per PR

Understand ROI, team by team.

Answer the board per PR: payroll plus agent spend, divided by the PRs that landed (merged and still in main 30 days later), against the $1,500 a landed PR cost before agents. See it team by team, including the team where it got worse.

Efficiency gains per PR

12 weeks · $2.96M payroll · $335k agent spend

2.0×Each landed PR now costs $757 in payroll + agent spend, vs. $1,500 before agents: same engineers, more PRs.

  1. All 14 teams

    $757 per landed PR, 2.0× as many PRs per dollar as before agents.

    4,348 PRs · $680 payroll + $77 agent spend

  2. Platform

    $521 per landed PR, 2.9× as many PRs per dollar as before agents.

    340 PRs · $483 payroll + $38 agent spend

  3. Growth

    $637 per landed PR, 2.4× as many PRs per dollar as before agents.

    500 PRs · $591 payroll + $46 agent spend

  4. Payments

    $1,881 per landed PR, more than the $1,500 before agents (0.8×).

    125 PRs · $1,577 payroll + $304 agent spend

Agent security

Know when a session crosses a line.

VibeCheck flags secrets, data leaving your sessions, destructive commands, new tools and weakened safety settings. A second, independent check must find evidence before an alert reaches you.

HighSecretsVisible to Theo and admins

An AWS access key was pasted into a prompt

  1. Rotate the key, then load it from the secrets manager.
  2. Check CloudTrail for calls made with it since 14:12.
EvidenceTheo MarshCodex14:12
Staging deploy fails auth. Here’s my key: [redacted:aws_access_key]

For developers

Improve how you work with agents.

Weekly Rewind finds the corrections you keep repeating in your own sessions and turns them into a rule you can paste, linked to the sessions it came from.

7 corrections → 1 rule

Kai · Oct 12–18

  • Tue · api-paginationI asked for the fix, not a rewrite.×3 in this session
  • Wed · null-checkLeave the rest of the file alone.×2 in this session
  • Thu · retry-backoffSmaller diff. Just the bug.×2 in this session

Rule for CLAUDE.md: Fix what I asked. Leave the rest alone.

Spend by work type

Match the model to the task.

Map model choices to real tasks: which models engineers reach for, where expensive ones dominate, and where a lighter model is worth testing.

Review

3.7B tokens · 4 weeks · $10.29/M-Tok (dollars per million tokens)

  • Claude Opus 5.5
  • Claude Sonnet 5.5
  • GPT-6 Astra
  • GPT-6 Sol
  • GPT-5.6 Sol
  • Claude Haiku 4.5

For platform teams

Change the harness. See if it helped.

Your harness is the prompts, skills, and models your agents work with. Change one, then compare corrections, rework, and cost per session before and after to see whether it made agent work better.
12 weeks
After plan-feature skill

2.4 1.5corrections / session−38%

Security and compliance

Your data stays yours.

Every session, token, and insight Maestro AI handles is protected by enterprise-grade safeguards and transparent compliance practices.

Explore our Trust Center Read our Privacy Policy
SOC 2 Type II compliant, monitored by Oneleet
  • Your data never trains AI. A contractual commitment.
  • SOC 2 Type II audited. The report is available in our Trust Center.
  • Encrypted in transit and at rest. Protection for the data Maestro AI handles.

Your bill says how much.VibeCheck shows what it was for.

Your team’s Claude Code, Codex and Cursor sessions, with the work they served and how they went. Free for up to 5 engineers.

Get started for freeOr book a demo →