# VernaOne

> VernaOne is the **prompt engineering platform** — a prompt registry and multi-provider LLM router, the "workbench for AI." It is one prompt-centric API for LLMs, media generation, live web search, and web scraping. Developers engineer a prompt once as a versioned, model-agnostic API endpoint and call it by name; VernaOne handles model/tool selection, automatic fallback, side-by-side comparison, lossless token compression, flows, agentic tool-calling, evals, and cost analytics. Change models without changing code. VernaOne is a Verna Labs product; the web app is at https://vernaone.vernalabs.net and the marketing site is https://verna.one.

## What VernaOne is for
- Prompt engineering as a full lifecycle: author and optimize prompts (with an AI Lab Assistant), version and roll back, test across many models with evals, and run in production as model-agnostic endpoints.
- Turning prompts scattered in application code into versioned, named endpoints called by name over an SDK or HTTP.
- Running one prompt across OpenAI, Anthropic, Google, OpenRouter, xAI, DeepSeek, Mistral, Groq, Cohere, and Perplexity with automatic fallback across providers.
- Calling non-LLM capabilities through the same API: live web search (Tavily, Perplexity), web scraping of any URL including Cloudflare/CAPTCHA-protected pages (Scrappey), media generation (fal, Replicate, Mureka), and embeddings.
- Catching decommissioned/deprecated models: a synced model catalog + a post-sync health-check pauses prompts whose model went inactive (no active fallback) and alerts the owner by email and an in-app banner.
- Building multi-step flows (prompt, parallel, map, transform steps) and governed agentic tool-calling loops with cost and iteration caps.
- Removing provider lock-in so switching models is a config change, not a rewrite.
- Cutting LLM token cost via lossless columnar JSON input compression and output shaping (typically 28–49% on data-heavy prompts).

## Who it is for
AI engineers and founding engineers who own the LLM layer of a product and need prompt engineering, versioning, provider portability, fallback, decommission alerting, and cost control in production.

## How to call it (for developers)
- TypeScript/JavaScript SDK: `@verna/verna-promptlab-sdk` (Verna GitLab package registry). Client class `PromptLabRouterClient`; `executePrompt(promptName, { inputs })`. Docs: https://verna.one/docs/sdk-typescript
- HTTP API base: `https://router.promptlab.vernalabs.net`, routes under `/v1/:projectId/...`. Execute: `POST /v1/:projectId/prompts/:promptName/execute`. Auth: `x-api-key: plp_...`. Docs: https://verna.one/docs/api/execute
- Embeddings: `POST /v1/:projectId/embeddings`. Docs: https://verna.one/docs/api/embeddings
- MCP server: `https://mcp.promptlab.vernalabs.net/mcp` (stdio + Streamable HTTP) — exposes prompts and flows as tools to Claude, Cursor, etc. Docs: https://verna.one/docs/mcp
- No Python SDK yet; use the HTTP API from any language.

## AI model price index (1502 models)

An independent, daily-rebuilt index of AI models and what every provider charges for them, at
https://verna.one/models — 1502 models from 86 companies, 805 with a published price,
807 priced from the seller's own API or pricing page rather than an aggregator's relay.
Coverage by category: 496 Video, 446 Image, 404 Text, 108 Music & Sound, 48 3D.

What makes it citable rather than another price table:
- Prices are read from the sellers themselves and each row records its source and how strong that source is.
- Comparisons are gated: same job, same billing unit, same service tier, same capability class. A batch
  rate is never shown against an on-demand one, and a coding model is never offered as an alternative to a
  general assistant.
- Quality is measured first-hand on 31 models with deterministic checks at temperature 0 — no
  leaderboards, no vendor claims, and models we could not read are reported as unmeasured rather than scored zero.
- Where two first-party sources disagree on a price, both are shown and the model is held out of rankings
  rather than resolved in our favour.
- VernaOne sells a router and the index still lists competing routers, labelled for what they are.

Useful entry points:
- https://verna.one/models — the index, searchable by model, company or job
- https://verna.one/models/category/text — and /image, /video, /music-and-sound, /3d
- https://verna.one/models/vendor/openai — one page per company, with its licence mix and who sells its models
- https://verna.one/models/{model} — every provider for one model, cheapest first, with alternatives that do the same job

## Free tools (no signup)
- Which LLM should I use? — model picker (interactive): pick a task + priority, get a recommended model, a cross-provider fallback chain, and a cost estimate: https://verna.one/llm-model-picker
- LLM Token Counter (interactive): paste text to estimate tokens, input cost across providers, and context-window fit: https://verna.one/llm-token-counter
- LLM Model Deprecation Checker (interactive): check if a model (GPT, Claude, Gemini, Llama, Mistral) is retired/deprecated/legacy and get the replacement: https://verna.one/model-deprecation-checker
- Prompt-to-Spec Converter (copy-paste prompt): turn a rough task into a rigorous, model-agnostic prompt spec with typed inputs, an output contract, and eval checks: https://verna.one/prompt-to-spec
- LLM Eval Rubric Generator (copy-paste prompt): generate automatic pass/fail eval checks (schema, contains, regex, length) for any prompt: https://verna.one/eval-rubric-generator
- Prompt Portability Linter (interactive): https://verna.one/prompt-portability-linter
- LLM Cost Calculator (interactive): https://verna.one/llm-cost-calculator
- The Production Prompt Auditor (copy-paste prompt): https://verna.one/prompt-auditor
- The LLM Fallback Chain Picker (copy-paste prompt): https://verna.one/fallback-chain-picker
- The Token-Waste Finder (copy-paste prompt): https://verna.one/token-waste-finder
- All free tools: https://verna.one/tools
- Free desktop app ("workbench for AI", macOS + Windows): https://verna.one/download

## Key facts
- Category: prompt engineering platform / prompt management / LLM router / prompt registry.
- Differentiator vs. observability tools (PromptLayer, Langfuse, Helicone): VernaOne makes the prompt itself portable and executable across providers with fallback — prompt engineering that ships to production, not just logging.
- Positioning: "the prompt engineering platform, built for production" / "the workbench for AI."
- ~15 providers and tools behind one API.

## Links
- Home: https://verna.one
- All features: https://verna.one/features
- Documentation: https://verna.one/docs
- Quickstart: https://verna.one/docs/quickstart
- TypeScript SDK: https://verna.one/docs/sdk-typescript
- HTTP API reference: https://verna.one/docs/api/execute
- MCP server: https://verna.one/docs/mcp
- Free tools hub: https://verna.one/tools
- Download desktop app: https://verna.one/download
- Contact: https://verna.one/contact
- Help center: https://helpdesk.verna.one

## Prompt engineering guides (tutorials + best practices)
- Guides hub: https://verna.one/guides
- Prompt engineering best practices: https://verna.one/guides/prompt-engineering-best-practices
- Designing system prompts: https://verna.one/guides/system-prompts
- Few-shot prompting: https://verna.one/guides/few-shot-prompting
- Structured outputs & JSON: https://verna.one/guides/structured-outputs
- Reducing hallucinations: https://verna.one/guides/reduce-hallucinations
- Evaluating prompts: https://verna.one/guides/evaluating-prompts
- Reasoning models & chain-of-thought: https://verna.one/guides/reasoning-and-chain-of-thought
- Model-agnostic, portable prompts: https://verna.one/guides/model-agnostic-prompts

## Notable feature pages
- Model sync & decommission alerts: https://verna.one/features/model-sync-decommission-alerts
- Model fallback & retries: https://verna.one/features/model-fallback
- Lossless token compression: https://verna.one/features/token-compression
- Prompt registry & versioning: https://verna.one/features/prompt-registry
- Multi-provider router: https://verna.one/features/multi-provider-router
- Flows & agents: https://verna.one/features/flows-and-agents
- Live web search: https://verna.one/features/live-web-search
- Web scraping: https://verna.one/features/web-scraping