GitHub repo token-waste auditor
Audit any public repository for uncached prompts and expensive or retired model ids; get a shareable receipt page and badge.
Enter any public GitHub repository and the static auditor scans it for uncached static prompts, hard-coded expensive or retired model ids, padded tool schemas and reasoning models on simple tasks, with file and line; precision is 100% and recall 100% on the 13-case golden corpus. The 10 most recent public audits have a median estimated waste of 7%.
· Source: pricing table 2026-10-26 and the gateway's own tokenizer and benchmark files; your inputs never leave the browser except token counts, which are not stored
At a glance
- Input
- owner/name of a public GitHub repository
- What it finds
- uncached static prompts, expensive or retired model ids, padded tool schemas, reasoning on simple tasks, missing cache keys
- Corpus precision / recall
- 100% / 100% (13 cases)
- Recent public audits
- 10 (computed)
- Median estimated waste (recent)
- 7% (computed)
- Output
- findings with file:line, estimated waste %, cost per 10k runs before/after
- Receipt page
- https://tokenzero.sh/audit/
/ - Badge
- https://tokenzero.sh/badge/
/ .svg - Runs where
- hosted for public repos; locally via npx -y @tokenzero/cli audit for private code
Most recent public audits
| Repository | Estimated waste | Findings | Audited |
|---|---|---|---|
| mendableai/firecrawl | 13% | 15 | 2026-10-26 |
| Pythagora-io/gpt-pilot | 9% | 4 | 2026-10-26 |
| assafelovic/gpt-researcher | 8% | 3 | 2026-10-26 |
| danny-avila/LibreChat | 8% | 1 | 2026-10-26 |
| Block/goose | 7% | 3 | 2026-10-26 |
| sweepai/sweep | 7% | 3 | 2026-10-25 |
| geekan/MetaGPT | 7% | 1 | 2026-10-25 |
| langchain-ai/chat-langchain | 7% | 1 | 2026-10-25 |
| mem0ai/mem0 | 7% | 1 | 2026-10-25 |
| allenai/olmocr | 4% | 15 | 2026-10-26 |
Audit a repository
Private code: npx -y @tokenzero/cli audit . runs locally and uploads nothing.
Questions people ask
Is my code uploaded?
For a public repository the hosted auditor fetches it from GitHub; for private code run the CLI, the MCP server or the GitHub Action locally and nothing leaves your machine.
How accurate are the findings?
Static heuristics scored at 100% precision and 100% recall on the 13-case golden corpus with 0% false positives on clean files; the corpus ships in the public repository. Waste percentages are estimates, labelled as such.
How do I add the badge?
Copy the Markdown snippet shown with the result: an SVG that links back to the receipt page and updates when the repository is audited again.
What does TokenZero cost?
Free up to $500 per month of routed spend; then 15% of verified savings with a 3x-ROI-or-$0 rule. Gateway overhead is 1.52 ms p50, measured.