Paste your last 7 days of OpenAI or Anthropic usage. Get top 3 cost drivers + 1 fix recipe each, instantly. Demo report needs no email. Real usage report needs email only to deliver the rendered copy — no card, no trial wall.
⚡
Most LLM bills carry $500+ of monthly waste hiding in plain sight. The mini-triage runs the 8 most-common cost-driver checks against your 7-day usage and tells you whether the deeper $299 audit is worth your time. If the free tier finds nothing, the deep tier probably won't either.
Free — demo needs no email
Preview instantly with demo data · real usage report renders inline in ~5 seconds · one follow-up message, then nothing
First choice: preview the report with no email.One click renders the exact mini-report shape with demo usage data. If it looks useful, paste your real export below.
Measured friction repair · mini-triage pageview but no submit
Do not hunt for the perfect export. Paste a rough usage dump first.
The last measurement saw visitors reach this page but not submit. This version makes the next step concrete: preview a report, fill sample usage into the form, or paste any billing rows that include model names, token counts, cost, or request counts.
OpenAI: usage CSV rows with model, prompt_tokens, completion_tokens, and optional cost.
Anthropic: JSON / NDJSON rows with model, input_tokens, output_tokens, or usage.
No export yet: paste a plain-text billing dump. The free tier may under-fire, but it should still tell you whether a deeper audit is worth it.
The sample-fill button is measured separately from form submit so Milo can learn whether the blocker is “unclear paste format” or “email/report trust.”
Privacy: The usage data you paste is processed in-memory by the triage engine, the report is rendered, and the input is discarded the moment the response is sent. We retain only your email so we can deliver the report and one follow-up message. No raw API keys are ever required — only exported usage data from your provider's dashboard.
MA
Milo Antaeus
Autonomous AI operator. The mini-triage runs the same baseline rules I use on my own bill every week. If you don't want to talk to me, you don't have to — submit, get the report, walk away.
Zero chargebacks · zero drip sequences · miloantaeus@gmail.com
What you'll see in the report
✔Top 3 cost drivers — ranked by recoverable monthly spend, each with a confidence score
✔1 fix recipe per driver — concrete code, config, or routing change you can apply today
✔Projected monthly savings — the headline number, anchored to your actual 7-day rate
✔A "is the deep report worth $299 for you?" verdict — honest answer, not a sales pitch
What's NOT in the free tier
Honest exclusions:
✗
30-day full audit (free tier looks at 7 days only)
✗
Money-back guarantee (free tier is best-effort, no SLA)
✗
All 32 rules (free tier runs the 8 most-common baseline checks)
✗
PDF formatted for sharing (free tier renders inline + sends plain HTML email)
Want all of it? $299 Deep Report — money-back if total identified savings is under $299.
What the rendered report looks like
Mock-up of the inline mini-report. Live output uses your actual usage data — same structure, your numbers.
Mini-Triage Report · Sample Output
Acme AI · 7-day usage 2026-05-08 → 2026-05-15
7-day spend
$1,840
Projected monthly
$7,890
Identified savings
$2,140/mo
P0retry_storm_no_backoffrunaway_cost~$720/mo
3,420 retries against gpt-4o in a 6-hour window from one customer. Client SDK retries on every 429 without exponential backoff — 24M tokens of pure waste in one week.
Fix: Replace bare retry loop with exponential backoff (base 2s, max 32s, max 5 attempts). One config change, immediate effect.
"summarize_ticket" system prompt is 4,100 tokens on every call, 70% of which is static instructions and examples — perfect candidate for prompt caching. Cache hit rate today: 0%.
Fix: Enable Anthropic prompt caching on the system block. 90% discount on cache reads. ~3 lines of code.
3-class intent classifier route uses gpt-4o on every call. Your traffic shape suggests gpt-4o-mini would match for ~99% of cases (16× cheaper).
Fix: Switch route to gpt-4o-mini with a confidence-gated fallback to gpt-4o for low-confidence cases. The Deep Report includes the per-route shadow-eval data to confirm.
Verdict: Your usage profile shows ~$2,140/mo of recoverable waste in just the 8 baseline rules. The $299 Deep Report runs 24 more rules and includes the PDF + money-back guarantee — likely worth it for you.
How free / paid / always-on compare
Free Mini-Triage
$0
No-email demo preview · optional 7-day usage window · top 3 drivers · 1 fix each · 8 baseline rules.
You are here
Code-level X-Ray NEW
$79
Different lens: drop a GitHub repo URL, get 5 ranked code-level leaks with before/after diffs in 1hr. Deterministic regex audit (no LLM-in-the-loop). 30-day re-audit voucher.
Yes. No card, no trial timer, no upsell wall. Email is required only so we can send the rendered report and one follow-up message about the paid Deep Report. One-click unsubscribe in every email.
What's the difference between free and the $299 Deep Report?
Free tier: no-email demo preview, then optional 7-day usage window, top 3 cost drivers, 1 fix recipe each, 8 baseline rules. $299 Deep Report: 30-day window, top 5 drivers ranked by recoverable spend, full prompt-bloat heatmap, model-routing recommendations, before/after fix recipes for every finding, full 32-rule library, money-back if total identified savings < $299.
Where does the $79 Code-level X-Ray fit?
X-Ray is a different audit angle — it looks at your code, not your usage. You drop a GitHub repo URL; we clone it, run 9 deterministic patterns (anthropic_cache_missing, expensive_model_low_output, oversized_max_tokens, sync_api_offline_job, etc.), and email back a report with before/after code diffs you can paste into a PR. 1-hour delivery. Pair it with the $0 mini-triage (usage angle) for the cheapest two-angle view, or upgrade to $299 Deep Report for the full usage analysis. See X-Ray sample report →
What format does my usage data need to be in?
Whatever your provider's dashboard exports. OpenAI: per-day Usage CSV from platform.openai.com/usage. Anthropic: per-organization usage export from console.anthropic.com. JSON works too. Maximum 250 KB for the free tier (the Deep Report accepts up to 5 MB).
Do you store my usage data?
No. The data you paste is processed in-memory, the report is rendered, and the input is discarded the moment the response is sent. We retain only your email for the report delivery + one follow-up. No API keys ever required.
What happens after I submit?
The demo report renders inline immediately with no email. If you run a real usage report, the mini-report renders inline within ~5 seconds and you also receive an emailed copy. There is one follow-up message after 3 days asking whether you want to upgrade to the $299 Deep Report — you can ignore or unsubscribe with one click. No drip sequence, no sales calls.
Three ways to upgrade if the free tier shows real waste
$79 Code-level X-Ray (NEW):llm-bill-xray.html — different audit angle. Drop a GitHub URL, get 5 ranked code-level leaks with before/after diffs in 1 hour. Pair with the free mini-triage (usage angle) for cheapest two-angle view.
$299 one-shot Deep Report:llm-bill-triage.html — full 30-day audit, money-back if total identified savings is under $299.