Launch — first 50 buyers
£79 £119
- 1 developer license
- Unlimited projects, commercial use included
- Full source — no obfuscation, no phone-home
- Zip download + private repo invite for updates
Ship on open-weight models this weekend.
A production-grade JavaScript starter kit: multi-provider by default, cost-visible from request one. Fresh clone + one API key → working chat in under 5 minutes.
- 4 providers wired: Novita, DeepInfra, Hyperbolic + any OpenAI-compatible base URL
- <2s failover: ordered fallback + exponential-backoff retry when a provider drops
- ±2% cost meter: every request priced from versioned rates;
/costsreturns session totals
What’s in the kit
-
Provider adapters
Novita, DeepInfra, Hyperbolic, plus any OpenAI-compatible base URL. Streaming and JSON response modes. Tool calling passes through behind a per-adapter capability flag.
-
Fallback + backoff
Ordered fallback across providers on failure. Automatic retry with exponential backoff. Timeout policy is one config value, not scattered try/catch blocks.
-
Cost meter
Token usage per request priced in USD from versioned prices.json (per-model in/out rates). /costs endpoint returns session totals. Matches manual token math within ±2% on the fixture set.
-
Eval harness
npm run eval runs prompt fixtures against any configured provider. String, regex, and JSON-shape graders. Prints a table and exits nonzero on regression.
The cost meter, on one screen
| Provider | $/M in | $/M out | Session cost | Verified |
|---|---|---|---|---|
| NovitaLlama-3.3-70B | $0.26 | $0.55 | $0.0118 | 2025-08-23 |
| DeepInfraLlama-3.3-70B | $0.34 | $0.72 | $0.0154 | 2025-08-23 |
| HyperbolicLlama-3.3-70B | $0.62 | $0.79 | $0.0230 | 2025-08-23 |
Pricing
Standard
£119
- Same license, same source
- Kicks in automatically after the first 50 sales
- Launch buyers never pay a difference later
FAQ
Why not just use LiteLLM?
LiteLLM is proxy infrastructure — a server you deploy to route traffic between providers. This kit is app-level starter code: chat UI, eval harness, and a cost meter wired into the request path, shipped as source you own and edit. The README states this comparison honestly; pick the proxy if routing is your product.
What is the refund policy?
14 days, no questions. Email the address on your receipt and the refund goes back to the original payment method. Launch buyers are covered identically.
How is the kit delivered and licensed?
Zip download immediately at purchase, plus an invite to the private GitHub repo for updates. License: one developer, unlimited projects, full source, commercial use included.
What do I need to run it?
Node 22+ and one OpenAI-compatible API key — Novita, DeepInfra, and Hyperbolic adapters ship in the box, and any other OpenAI-compatible base URL works through the generic adapter. No build step, no database.
Do you sell inference or hosting?
No. We sell code. You keep your provider relationship and your keys; nothing in the kit phones home.