Asteria Docs

Most AI projects fail in the gap between two teams. The people who understand the job cannot build software, and the people who build software do not understand the job well enough to encode it. Six months go by producing something that answers correctly and sounds wrong.

Asteria Cloud closes that gap by splitting the work where it naturally divides.

Whoever owns the job builds the agent, in the app, with no code. They give it instructions, point it at the right documents, test it against real questions, and keep correcting it until it is right. See Agents in the user guide.

Whoever owns the product plugs it in, through an API shaped like the one they already use. A few lines, not a project.

Nobody has to explain the domain to an engineer, and nobody has to explain retrieval to a marketer.

Two ways in

Agent mode calls an agent somebody built and validated in the app, with its instructions, its documents and its tools. This is the interesting one.

Gateway mode is a plain model call: every model your organisation has, behind one provider, one key, one URL.

Both are the same endpoint. The agent field decides which you get.

curl -X POST https://api.asteria-labs.com/v1/chat/completions \
  -H "Authorization: Bearer sk-ast-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "agent": "asteria",
    "messages": [{"role": "user", "content": "Summarise our Q4 board deck"}]
  }'

It is the OpenAI shape

/v1/chat/completions takes and returns what you expect, so an OpenAI-compatible SDK works by changing the base URL and the key. Streaming, system messages, multi-turn: the usual shapes.

That is a deliberate constraint rather than a shortcut. Adopting this is a config change, and so is leaving.

Where to go

Projects

What a project is, why every key belongs to one, and how spend stays separate.

From the app to production

The worked example: a chatbot built by marketing, shipped by engineering.

Gateway mode

One provider for every model, and why you should not name one.

Getting started

Your first call, with a key.

Base URL and keys

https://api.asteria-labs.com/v1

Every request carries a Bearer token, and there are two kinds:

Prefix	Scope	For
`sk-ast-…`	One project	Chat completions, listing models and agents
`sk-ast-adm-…`	The organisation	Everything under `/v1/organization/`

Project keys are created inside a project. Admin keys are created by an organisation administrator under Settings then Admin API Keys, and they can manage users, projects and keys, so treat them accordingly.

Full endpoint documentation is under Endpoints in the sidebar, generated from the API itself.

Platform & API