Atris
View as markdownEnter

One key.
Any tool.
Pay per use.

The Atris API speaks the OpenAI protocol. Paste your key into any SDK or agent and every request bills exactly what it used, receipt included.

Get your keybase_url = https://api.atris.ai/v1

Live in three steps

  1. 01

    Create a key

    Sign in and mint a key in the developer dashboard. The secret shows once, and every key can carry a monthly spend cap, so a leaked or runaway key stops itself.

  2. 02

    Load credits

    Buy credits at atris.ai/credits with a card. One credit is one cent, and credits never expire.

  3. 03

    Call the API

    The key is a standard bearer token against https://api.atris.ai/v1. If a tool speaks OpenAI, it speaks Atris.

curl
curl https://api.atris.ai/v1/chat/completions \
  -H "Authorization: Bearer $ATRIS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "atris:pro",
    "messages": [
      {"role": "user", "content": "Say hello."}
    ]
  }'
Python — official OpenAI SDK
from openai import OpenAI

client = OpenAI(
    base_url="https://api.atris.ai/v1",
    api_key="atris_...",
)

reply = client.chat.completions.create(
    model="atris:pro",
    messages=[{"role": "user", "content": "Say hello."}],
)

Three models, plain names

Use them anywhere a tool asks for a model name. Same key, same meter. These three are the ones to start with; the live catalog, including specialist lanes, is always at GET https://api.atris.ai/v1/models.

atris:autoLet Atris pick the laneRoutes to fast today
atris:fastQuick answers, high volumeCheapest per token
atris:proThe default for real workStrong reasoning, fair price
atris:maxHardest problemsFrontier model

One call each

Beyond chat, every Atris product is one HTTP call with the same key, the same error shape, and the charge printed on the reply. Slow jobs answer 202 with an id you poll for free. Full references live at github.com/atrislabs/atrisos-backend/docs.

Router
POST /v1/chat/completions{"model": "atris:auto", "messages": [...]}

One chat call. Atris picks the model lane and bills the tokens it used.

Metered per token
Answers
POST /api/answers{"question": "...", "depth": "quick"}

A research answer with the web sources it stands on. No sources, no answer.

5 credits quick, 25 deep
Extract
POST /api/extract{"url": "...", "schema": {...}}

Schema-valid data from a page or pasted text, every value with a verbatim quote.

5 credits, plus 1 per extra page
Notes
POST /api/notes{"url": "https://www.youtube.com/watch?v=..."}

Timestamped notes and exact quotes from a video transcript.

10 credits
Publish
POST /api/publish{"slug": "my-site", "html": "<!DOCTYPE html>..."}

A live site at slug.atris.ai from your HTML, same call to update it.

50 credits
Design
POST /api/design/extractions{"url": "https://example.com"}

A brand design system as JSON: real colors, fonts, surfaces, layout.

10 credits, 2 from cache
AEO
POST /api/aeo/checks{"name": "Joe's Plumbing", "city": "Albany"}

Asks five AI engines about a local business and reports who named it, who cited it, and which rivals came up.

500 credits
Desks
POST /api/desks{"business_name": "...", "instructions": "...", "customer_email": "...", "amount_dollars": 50}

A public customer chat desk for a business, safety probed, with its own checkout link.

500 credits
RL
POST /v1/rl/store, /v1/rl/learn, /v1/rl/context{"workspace": {...}, "environment_id": "...", ...}

Store experiences, run one bounded learning batch, read back what was learned.

1 credit per learn batch
the router
curl https://api.atris.ai/v1/chat/completions \
  -H "Authorization: Bearer $ATRIS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "atris:auto",
    "messages": [{"role": "user", "content": "Summarize this contract."}]
  }'
a slow job: AEO check
curl https://api.atris.ai/api/aeo/checks \
  -H "Authorization: Bearer $ATRIS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"name": "Joe'\''s Plumbing", "city": "Albany"}'

# 202 with a job id and the charge on the reply
# {"id": "...", "status": "accepted",
#  "atris": {"credits_charged": 500, "cost_usd": 5.0, ...}}

curl https://api.atris.ai/api/aeo/checks/$ID \
  -H "Authorization: Bearer $ATRIS_API_KEY"

Screens, not paragraphs

Atris works as the brain behind generative UI kits such as json-render. You define the components and actions the model may use, send that catalog as the system message, and Atris returns a screen made only of those parts.

Use atris:max for this. In our tests it named every action exactly, and the catalog prompt is cached after the first call so repeat screens cost less.

Next.js route — json-render with Atris
// app/api/ui/route.ts in your Next.js app (json-render quickstart, two values swapped)
import { OpenAI } from "openai";
import { catalog } from "@/lib/catalog"; // your defineCatalog(...)

const atris = new OpenAI({
  baseURL: "https://api.atris.ai/v1",
  apiKey: process.env.ATRIS_API_KEY,
});

export async function POST(req: Request) {
  const { prompt } = await req.json();
  const stream = await atris.chat.completions.create({
    model: "atris:max",
    stream: true,
    max_tokens: 8192,
    messages: [
      { role: "system", content: catalog.prompt() },
      { role: "user", content: prompt },
    ],
  });
  return new Response(stream.toReadableStream());
}

The meter is the price

Every request is billed from actual model token usage at Atris rates. One credit is one cent, and each request costs at least one credit.

Cached input bills at the cheaper cache rate, so repeated context costs less. Spend caps and a per-request ceiling mean a runaway loop cannot drain a wallet.

Ask the meter directly

GET/api/atris2/billing/quote

Price estimate before you spend

GET/api/atris2/billing/usage

Balance, today and per-model spend

GET/api/atris2/billing/receipts

Per-request receipts with exact tokens

Built to be learned by agents

The whole API fits in one markdown file any model can read. Hand it to an agent with a key and it can price, call, and reconcile its own work.

one line for any agent
curl https://atris.ai/developers.md

Common questions

Updated 2026-09-18

What is the Atris API?
The Atris API is an OpenAI-compatible chat completions API at https://api.atris.ai/v1. An Atris API key works as the bearer token in any OpenAI SDK or tool that accepts a custom provider, including agent tools like Devin and Cursor.
How much does the Atris API cost?
Atris API usage is metered from actual model token usage. One credit is one cent, every request costs at least one credit, and each request shows its exact token counts on its receipt. Credits are prepaid at atris.ai/credits and keys can carry a monthly spend cap.
What can an agent call besides chat?
Nine one-call APIs share the same key and billing: the model router (atris:auto), Answers, Extract, Notes, Publish, Design, AEO checks, Desks, and RL. Each is one POST with the charge on the reply; slow jobs return 202 and a free GET to poll.
Does the Atris API work with OpenAI SDKs?
Yes. Set base_url to https://api.atris.ai/v1 and api_key to your Atris key in the official OpenAI Python or JavaScript SDK, or any tool that accepts an OpenAI-compatible provider, and requests bill your Atris credits.

Key, credits, first call. Under a minute.

Get your key