Decillion's developer API is now in public preview: language, image, video, speech and embedding models, Linux computers and file storage behind one key and one prepaid USD balance.
The Decillion developer API is now in public preview. It is one HTTPS API that gives your own code the infrastructure our AI agent rooms run on: models of every kind, clean Linux computers and file storage, with one key and one prepaid balance in US dollars. Anyone with a Decillion account can open the
developer console, create a key and send a first request in a few minutes.
This post covers what is in the preview, who it is for, how billing and limits work, and how to start. The
developer docs have every endpoint in full.
What is in the public preview
• Computers: Linux machines you create, start, stop and run commands on from your code, billed by the minute while they run. See
computers.
• Storage: files your code uploads, downloads and shares with its computers, billed for the gigabytes you keep. See
storage.
• The console: a playground to try any model, the model catalog with live prices, API keys, usage by model, key and day, a log of every request, billing and team management.
Who it is for
• Product and B2B teams adding AI features (chat, search, summaries, image or voice) to an app they already run, without signing up with a separate vendor for each kind of model.
• Developers and indie builders who want one key, one bill and one set of docs for the whole stack.
• Agencies and consultancies building for clients: each client can be its own organization, with its own balance, keys and people.
• Teams building agents that write and run code, and need a clean machine to run it on instead of their own servers.
• Search and data teams who need embeddings and reranking next to the language model that reads the results.
Why one API instead of five vendors
A typical AI stack today is a language model provider or two, another vendor for images or speech, a sandbox service for running code, an object store for files, and a spreadsheet to reconcile the bills. Each has its own key, SDK, dashboard, rate limits and invoice. When one provider is down, your app is down unless you have built the failover yourself.
The Decillion API puts those pieces behind one base URL. Every response reports what it cost, every call lands in the same usage view and log, and everything is paid from one balance. You add a kind of model, or a computer, by calling another endpoint, not by onboarding another company.
OpenAI-compatible, so you keep your code
The chat endpoint follows the OpenAI Chat Completions format, including streaming and images, audio and files inside messages. Any OpenAI SDK works if you change two things: the base URL, which is
https://api.decillionai.com/v1, and the key. Embeddings, image generation, speech and transcription follow the same familiar shapes, so moving an existing integration is usually a configuration change. The
quickstart has the request in curl, Python and JavaScript.
Fallback models instead of a router to maintain
Every model endpoint takes an optional list of fallback models. If the model you asked for cannot answer because it is down or overloaded, the request moves to the next one on its own, and the response says which model answered. A request refused for what it asks, such as a bad parameter, is not retried, because the others would refuse it too. See
fallback models.
Computers for the code your agents write
A computer is a clean Linux machine of your own. The standard image is the same machine our own rooms use: Debian 12 with Node.js 22, Python 3, git, common command-line tools, build tools, and Chromium with a remote desktop. You can also start from a public container image. Its files live in a persistent /workspace folder that is kept when the computer stops.
• Charged by the minute while it runs, and nothing while it is stopped.
• It stops by itself after a couple of idle minutes, so a forgotten machine does not run up a bill.
• You choose its size in CPU and memory. The current limits and prices come from the API itself, so your code never has to guess them.
• Typical uses: running code a model wrote, testing a build, scraping with a real browser, or giving your own agent a machine to work on.
Control spend before it happens
• Prepaid in US dollars: add funds first, and requests stop when the balance runs out. You are never invoiced after the fact.
• Costs to the fraction of a cent: every response reports what that call cost.
• Per-key spend caps, requests-per-minute limits and a list of the models a key may use.
• Permissions per key: All, Read only, or a Restricted level for each service.
• A request refused by a cap, a permission or a limit costs nothing.
• Usage by model, key and day, and the full request and response of every call in Logs.
Organizations for teams and clients
Everything in the console belongs to an organization. Your personal organization is your own Decillion account, and its balance is the wallet you already use in the app. Create more for a company, a product or a client: each has its own balance, keys and people, and a key only ever spends its own organization's balance. Invite teammates by their Decillion username. See
organizations.
What public preview means
• The API is open to everyone, and requests are real: they run on real models and machines and are charged to your balance.
• We may still change limits, defaults and prices during the preview. The docs, the console and the API's own options always show the current values.
• We would like to hear what is missing. Tell us through
support.
How to start in five minutes
• 2. Add funds to your organization under Billing.
• 3. Create a key under API keys. You see the secret only once, so copy it straight away.
• 4. Put it in an environment variable, such as DECILLION_API_KEY, rather than in your code.
• 5. Send a chat request with any OpenAI SDK pointed at https://api.decillionai.com/v1, or try models first in the Playground.
• 6. Open Logs to see the request, the reply and what it cost.
Is the Decillion API compatible with the OpenAI SDK?
Yes. Set the base URL to https://api.decillionai.com/v1 and use your Decillion key. Chat completions, streaming, embeddings, images, speech and transcription use the formats the OpenAI SDKs already send.
How is the Decillion API priced?
By use, from a prepaid balance in US dollars. Language models charge per million tokens in and out, other models for what they produce (images, seconds of audio or video, characters or searches), computers by the running minute, and storage by the gigabyte kept. There is no seat fee and no monthly minimum. Current rates are in the console's model catalog. The same idea runs through the whole product:
pay for the work, not another AI seat.
Do I need to use the Decillion app?
No. The API is for people who want the infrastructure without the app. You sign in with a Decillion account to manage keys and billing, and your code talks to the API host only. That host serves developer keys and nothing else, so a key can never reach the rest of your account. If you also want ready-made AI agent teams working in shared rooms, that is
the Decillion app.