Cost visibility for AI-assisted engineering

Stop burning tokens.
Start shipping.

See every dollar your AI agents spend, kill the CLAUDE.md bloat quietly draining your budget — without a single line of source code leaving the machine.

tokencheat audit — ~/acme/monorepo
example output
$ tokencheat audit .
→ indexing repo via tree-sitter AST … done (1,204 files)
→ reading CLAUDE.md · .cursorrules · mcp.json
CLAUDE.md — 4,180 tokens, 48% stale [critical]
6 MCP tool schemas — 3,100 tok/turn [warning]
prompt caching off on 3 flows [warning]
retrieval targeted — 84/100
Est. avoidable spend: $412/mo
run the full audit → writes optimized CLAUDE.md
0/100
Good — 3 fixes reclaim wasted spend
Config hygiene61
Retrieval efficiency84
Cache opportunity70
Tool overhead73

Runs in your browser — nothing you paste is uploaded.

Works with the tools you already ship with
Claude CodeAnthropicGitHub CopilotCursorWindsurf
The problem

Nobody knows where the money goes.

AI coding bills arrive as one opaque number. No breakdown by repo, model, or developer. No idea which prompts are efficient and which are dragging a stale thousand-token config into every single turn.

By the time finance asks, the tokens are long gone.

100 real AI coding configs, analyzed
97%
of the detectable waste sat in just 10 files.
964
tokens — median config size (n=100)
24%
of configs ran over 200 lines
2,965
tokens in the largest session-log section found in one CLAUDE.md
Read the full analysis →
What it does

Visibility, hygiene, and control — in one CLI.

Cost attribution

Every dollar broken down by model, repo, and developer. Route flagship-priced work to cheaper tiers and watch the bill drop.

Context hygiene

Audit CLAUDE.md, cursorrules, and MCP schemas for bloat. One command rewrites them lean — you review the diff.

Local-first by design

A Rust CLI that indexes via AST on your machine. Zero source code uploaded — only anonymized metrics, if you opt in.

The dashboard

Spend, watched like production.

Team workspace preview
Monthly AI spend
$12,480
↑ 214% vs last mo
Context efficiency
72/100
↑ 6 this week
Cached ratio
38%
target 70%
Avoidable waste
$412/mo
3 fixes available
Token spend over time
Opus Sonnet Haiku
Spend by model
Claude Opus$7,120
Sonnet$3,940
GPT-4o$980
Haiku$440
Reroute flagship models on simple tasks → dollar-valued routing tips per repo.
How it works

Three commands to a smaller bill.

01
$ tokencheat collect

Get the CLI (early access)

Request access to the Rust CLI. It scans your local agent transcripts and configs — nothing to wire up, nothing sent anywhere.

02
$ tokencheat audit --claude-md ./CLAUDE.md

Audit & score

Get a context score, a ranked list of what's wasting tokens, and an optimized instruction file you review as a diff.

03
$ tokencheat sync --report report.json

Track & govern

Optionally sync scores to the team dashboard to watch spend and context health across repos over time.

Your source code never leaves the machine.

The browser tools run entirely client-side and the Rust CLI indexes locally. TokenCheat sees token counts and structure — never your source.

0 bytes of source code uploaded
audits run in your browser or on your machine
Pricing

Start free. Scale when it pays for itself.

Free
$0
Web tools + CLI, forever.
✓ Browser audit & optimizer
✓ Rust CLI — local audits
✓ 3 saved audits/month
✓ Model price comparison
Run a free audit
Done for you
Done-for-you Audit
from $29 one-time · $79 founding price
We audit your stack and hand back the fixes.
Full config & MCP review
Optimized files, ready to commit
Prioritized fixes with est. savings
Delivered in 2 business days
Book an audit
Team
$299/mo
Shared dashboards & governance.
✓ 25 seats
✓ Team aggregate dashboard
✓ Spend tracking & alerts
✓ Cost per PR metrics
See Team plan

See what you're really spending.

Run your first audit in under two minutes. Free, local, no credit card.

Frequently asked questions

How TokenCheat works, what happens to your data, and what the numbers mean.

Get early access

Product updates and TokenCheat release notes—no spam.