VernaOne Documentation

VernaOne is the prompt engineering platform that turns your prompts into versioned, model-agnostic API endpoints. You author a prompt once — as a named endpoint with typed {{variables}} — and call it from your code. VernaOne handles model and provider selection, automatic fallback, cost analytics, and evals, so you can change models without changing code.

These docs cover the three ways to run your prompts in production, plus how to authenticate.

Ways to call VernaOne

  • TypeScript / JavaScript SDK — a typed client for Node and the browser. run a named prompt and get a structured result.
  • HTTP API — call any prompt by name over plain HTTPS from any language.
  • MCP server — connect VernaOne to Claude, Cursor, and other MCP clients as a set of tools.
  • Desktop app — the free “Postman for AI” to author and compare prompts across models on your machine.

Core ideas

  • Prompts are endpoints. Every prompt is a named, versioned resource. Your code references the name, never a vendor SDK or a hardcoded model id.
  • Model-agnostic. A prompt runs across OpenAI, Anthropic, Google, and more. Swap the model or add fallbacks in config — no code change.
  • Governed & observable. Projects isolate prompts, models, keys, and usage. Every execution is recorded with model, cost, latency, and tokens.

Next steps

New here? The fastest way to get a feel for VernaOne is the free desktop app — test and compare prompts across models with your own keys, no signup.