AI cost and margin intelligence for SaaS founders. See exactly which users cost you more than they pay, per LLM call, zero added latency.
Drop the SDK into your app, get a dashboard that shows cost per user / feature / model and recommends cheaper swaps. Set per-plan spending caps the SDK enforces before the LLM call.
| Package | Install | Source | |
|---|---|---|---|
| TypeScript / Node | @weckr/sdk |
npm install @weckr/sdk |
typescript/ |
| Python | weckr-sdk |
pip install weckr-sdk |
python/ |
| Claude / Cursor (MCP) | @weckr/mcp |
npx -y @weckr/mcp |
mcp/ |
| Claude Code (Skills) | weckr plugin |
/plugin install weckr@weckr |
skills/ |
Each subfolder has full setup docs, examples, and supported-model lists.
This repo is also an installable Claude Code plugin that bundles four skills, so your AI assistant knows real model prices and the exact Weckr syntax instead of inventing code:
/plugin marketplace add Ghiles3232/weckr-sdks
/plugin install weckr@weckr
| Skill | What Claude learns |
|---|---|
weckr-integration |
Wire the SDK in with the correct two line pattern, errors, and provider notes |
weckr-model-pricing |
Current per token prices for OpenAI, Anthropic, Gemini, and Kimi |
weckr-cost-estimator |
Project what an AI feature costs per call, per user, per month before shipping |
weckr-margin-audit |
Flag which pricing plans go underwater once LLM cost is counted |
The pricing skills read the live feed at useweckr.com/pricing.json first (generated from the same table Weckr bills with), and a weekly pricing watcher opens a reviewed PR whenever a provider changes rates, so the numbers track reality instead of drifting. Full explainer at useweckr.com/skills.
Click around with seeded data for a fictional SaaS. No signup required.
Clone a full AI SaaS with Weckr already wired in and deploy in minutes. Both are MIT licensed.
- weckr-nextjs-starter: Next.js 14, Supabase auth, Stripe billing, an OpenAI endpoint, and Weckr tracking cost and margin per user.
- weckr-fastapi-starter: FastAPI, a Claude endpoint, SQLite, and Weckr tracking cost and margin per user.
Step by step integrations for the stacks people actually ship on, each with the exact code and where the wrap goes:
- Next.js: route handler pattern,
flush()for serverless, streaming notes. - FastAPI: endpoint pattern with the authenticated user, zero added dependencies.
- LangChain: wrap the client at the boundary, not the callbacks.
- CrewAI: price every agent turn, catch the stuck crew.
- You wrap your LLM calls with
wk.chat(client, opts). - The SDK forwards the call to OpenAI / Anthropic / Gemini and returns the original result, with no added latency on the request path.
- After it resolves, the SDK fire-and-forgets a log to the Weckr API with
(userId, feature, model, tokens, latency, plan). - The backend computes cost server-side from public per-token pricing (clients can't forge cost) and stores
(cost, revenue, margin)per request. - The dashboard rolls it up, per user, per feature, per model, and surfaces unprofitable users + cheaper-model recommendations.
Only the call metadata above: model name, token counts, latency, plan, and your userId string. No prompt text or completion text is ever sent.
app.useweckr.com. Free tier, no credit card.
Have a question about integration or a feature request? Open a discussion: https://github.com/Ghiles3232/weckr-sdks/discussions
We respond to every question.
MIT. See LICENSE.