vibe-bill: a Claude Code skill that estimates AI costs — Claude reads, code counts
Strong_Turnip8429 · reddit · 2026-11-13
The authors built vibe-bill, a Claude Code skill that estimates an app's AI running costs from 100 to 100,000 users, and share the engineering lessons that make its output trustworthy:
- Claude reads, code counts: instead of letting the LLM do arithmetic, Claude describes each AI call in the codebase (feature, model, trigger, I/O size) while a plain script computes costs against a price list — same code, same numbers every run.
- Dated price snapshots from the OpenRouter list, with each price on the receipt showing its date.
- Never reads .env — only env variable names like ANTHROPICAPIKEY, never values.
- Asks when unsure: up to 4 clarifying questions (e.g., which model production actually uses); provider can also be pinned in config.
On one of their apps (AI grading of student submissions) at 10k monthly users: GPT-6 Luna costs $174/mo vs Claude Sonnet 5.5 at $2,901/mo — nearly all from the grading feature. The takeaway isn't "avoid Sonnet" but decide per feature which model it really needs. MIT-licensed and free.
More from coding & agent
- Open-source getmcp-cli turns 1,280-endpoint OpenAPI specs into 25-tool MCP servers, cutting tool tokens from 263k to 6k — Worried-Barracuda845 · 2026-11-13
- Formal verification is the Rust of software correctness — LLMs are making people write specs — arpit_bhayani · 2026-11-13
- whats-in-my-claude: open-source local visualizer for Claude Code's .claude folder, runs fully offline via npx — Srqi · 2026-11-13
- AuralKit 3.0: on-device Swift transcription of video audio tracks with timestamps — rudrank · 2026-11-13
- Delulu Bio adds an MCP server to its free link-in-bio tool, letting Claude manage your page by voice — delulu_intern · 2026-11-13
- Custom motion-director skill turns any voiceover into 1080x1920 motion-graphics animations with Gemini Spark — mijozakinho_2 · 2026-11-13