Token efficiency audits

Your AI bill, with the waste itemised.

We read the traffic you already have, price the recoverable waste call site by call site, and replay your own calls on the cheaper model to prove it.

Get your free auditSee an example report
Support triage agent · live auditAfter TokenZip
Monthly cost

$454

Recoverable

$187

Share of spend

29%

Classify intent
claude-haiku-4-5Tier · measured
$83
Retrieve context
gpt-5-miniTier · estimated
$50
Draft reply
claude-opus-5 — reasoning requiredNo change
$203
Compose reply
prefix cachedCaching · estimated
$45
Summarize thread
1,140-token system promptBloat · estimated
$73

Read-only · no SDK · nothing in your request path · first audit free

§ 01How it works

Three steps. The third is the one nobody else does.

Step 01

Connect

A read-only usage key, or a trace export you already produce. Nothing installed, nothing in your request path.

OpenAIAnthropicHeliconeLangfuseLangSmithJSONL
Step 02

Detect & price

Call sites reconstructed from the prompt text, then three detectors across them. Every finding carries a dollar figure.

Model tier mismatch$147
Prompt bloat$24
Caching opportunity$16
Step 03 · the differentiator

Replay proves it

Your real calls, re-run on the cheaper model, scored by an independent judge. Below 90 across 5 samples we drop the finding rather than caveat it.

10 samples · mean95/100
{"category":"integrations",
 "priority":"p2","team":"integrations"}
§ 02Proof

No logos. No testimonials. A whole audit you can open.

6 call sites, 3,960 call records, run end to end through the same pipeline your data goes through.

tokenzip.app/r/demoOpen the example report →
Identified monthly waste

$187

29% of a $641 monthly bill, recoverable today.

Monthly spend

$641

Findings

8

Replay-proven

$103

  1. measuredclaude-opus-5 → claude-haiku-4-5 · ticket-triage$103/mo
  2. estimatedgpt-5 → gpt-5-mini · usage-API rows, nothing to replay$44/mo
  3. estimatedUncached ~4,418-token prefix on every reply-composer call$16/mo
§ 03Pricing & payback

See the number first. Decide about paying second.

Free

$0one audit

Enough to find out whether you have a problem worth paying to fix.

  • Full spend breakdown, by model
  • Total recoverable figure — the whole number
  • Your largest finding in full
  • Remaining findings listed, locked
Get your free audit
StarterFull audit

$199per month

For the workspace that has seen the number and wants the rest.

  • Every finding unlocked, with evidence and fix
  • Replay verification on your own calls
  • Automatic monthly re-analysis
  • Board-ready PDF of the complete report
Get your free audit
Is $199 worth it?

Drag your monthly AI spend. Recovery is set at 29% — the share the example audit found, not a best case.

Monthly AI spend$6,000
Recoverable per month$1,740
Less TokenZip Starter−$199
Net, month one+$1,541

Starter pays for itself above $686/mo of AI spend. You are 8.7× past that.

Start here

Find out what your AI bill is actually paying for.

Connect one data source, get an audit back with a dollar figure on it. If there is nothing to recover, the report says so.

Get your free auditAgency plans