New

Sandboxes, in private preview: isolated microVMs for AI agents, compatible with the E2B SDKs.Read more

Serenity Edge

What’s new

The Serenity Edge changes that matter to your business, newest first. The technical detail of each one is in the documentation and the release notes.

  1. Feature

    A prompt cache for each API key

    Each API key now has a prompt cache of its own: a prefix cached for one key is never used to answer a request made with another, as the Data Processing Agreement and the data residency page state. Prefixes repeated with the same key are still served from the cache and billed at the cached input price.

  2. Feature

    Two image demos in the playground

    The playground adds two demos that send an image to the API, with real calls. In the OCR challenge, PaddleOCR-VL 1.6 reads made-up documents that you can rotate, blur or make noisy, and the page compares what it read with the original, field by field. In Questions about an image, Orion Plus or Orion Pro answers your question about a chart, a floor plan, a shop shelf, a dashboard or an image of your own.

  3. Launch

    Sandboxes, in private preview

    Serenity Edge Sandboxes gives your AI agents isolated microVMs in which to run the code they write, analyse data, browse with a headless browser or run build and test jobs, away from your own systems. It is compatible with E2B: the official E2B SDKs work with a Serenity Edge API key. It is in private preview: access is on request.

    Try it in the playground:A sandbox reads today’s web

  4. Feature

    Playground: the API live in your browser

    The new playground shows the Serenity Edge API with real calls, no key or sign-up needed. Play Blocks and gems against Orion Compass and try four business cases: support inbox triage, expenses against a policy read from the image of each receipt, a personal data hunt and exact clause search in a contract. Every demo shows the exact call the page makes, ready to copy.

  5. Feature

    Several images in one request

    Orion Pro, Orion Plus and the open general-purpose chat models now read several images in one request, counted over the whole conversation. Compare product photos, check the pages of a scanned document together or follow a sequence of screenshots. How many each model reads is published in the model list and in the guide.

  6. Feature

    Responses API and Codex CLI

    Serenity Edge serves OpenAI’s Responses API on every chat model, with the same keys and prices as chat completions, so the OpenAI SDKs work as they are. Codex CLI works with it too: set it up with Serenity Edge to work with Orion Pro from your terminal. The API stores nothing, so every request carries the whole conversation.

  7. Model

    Decision models that read images

    Orion Compass Flash and d3 now read images together with the text, so a decision can rest on a receipt, a photo or a scanned form. Image tokens are billed as input, like the text. The playground’s expenses demo uses it: the model approves, rejects or asks for details on each receipt from its image.

  8. Model

    Decision models in preview

    Orion Compass, Orion Compass Flash and d3 answer questions about a text with a probability for every possible answer, instead of generating text. That makes them cost less than asking a chat model for the same decision. Use them to route tickets, classify documents or prioritise cases with a decision your code can act on directly. They are in public preview and only input is billed.

    Try it in the playground:Support inbox triageBlocks and gems

  9. Model

    Embeddings, rerank and redaction

    Three new building blocks for search, RAG and workflows that handle personal data, with processing kept in the EU: multilingual embeddings for semantic search, rerank to order results by relevance, and redaction to mask names, email addresses, phone numbers and account numbers before a text is processed. Only input is billed.

    Try it in the playground:Exact clause searchPersonal data hunt

  10. Model

    Document OCR with PaddleOCR-VL 1.6

    PaddleOCR-VL 1.6 joins the chat models to read document pages: from the image of a page it returns its text, tables, formulas or chart data. Use it to digitise invoices, delivery notes or forms through the API you already use.

    Try it in the playground:OCR challenge

  11. Platform

    Available on cortecs and EUrouter

    Serenity Edge is now available as a provider on cortecs and EUrouter. If you work with one of these platforms, you can choose Serenity Edge from your account there, with that platform’s key and billing. More platforms are on the way.

  12. Feature

    Image input on the chat models

    The chat models accept images alongside text: screenshots, product photos or scanned pages that the model can describe, compare or summarise. They travel in the same chat completions request and are billed as input tokens.

    Try it in the playground:Questions about an image

  13. Launch

    Serenity Edge launches

    Serenity Edge opens its OpenAI-compatible inference API, served from our own infrastructure in Spain with data in the EU. Change the base URL to use the Orion family and open-weight models, with prices published per million tokens and zero data retention.

See all release notes

Try what is new with your own code

Get an API key and make your first request, or try the demos in the playground first.