Developers · Free preview

Build with ATLAS Code.

The ATLAS Code preview runs on an OpenAI-compatible /v1 API. We’re fine-tuning our own version, and it ships under the same name once it passes our tests. Use the OpenAI SDK or any tool that lets you set a base URL. Call the API from servers, CLIs, and editors. Browser cross-origin calls aren’t supported.

The three values every tool needs

Base URL, your ATLAS API key, and the model id. Preview accounts see atlas-code-1-preview in GET /v1/models. If it isn’t listed, your account isn’t in the preview yet.

Base URL: https://elderai.cloud/v1
Model: atlas-code-1-preview
Auth: Authorization: Bearer $ATLAS_API_KEY
Endpoints: GET /v1/models · POST /v1/chat/completions · POST /v1/responses (streaming and non-streaming)
Tool calling: OpenAI tools / tool_calls

Quickstart

Your first request

Create a key in your account and export it as ATLAS_API_KEY. Keep it in an environment variable and never commit it. Idempotency-Key is optional; send one when you want safe retries.

curl

curl https://elderai.cloud/v1/chat/completions \
  -H "Authorization: Bearer $ATLAS_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $(uuidgen)" \
  --max-time 300 \
  -d '{
    "model": "atlas-code-1-preview",
    "messages": [
      {"role": "system", "content": "You are a senior engineer. Answer with code."},
      {"role": "user", "content": "Write a Python function that parses ISO-8601 durations."}
    ],
    "max_tokens": 1024
  }'

Python · openai SDK

import os
from openai import OpenAI

client = OpenAI(base_url="https://elderai.cloud/v1", api_key=os.environ["ATLAS_API_KEY"], timeout=300)
stream = client.chat.completions.create(
    model="atlas-code-1-preview",
    messages=[{"role": "user", "content": "Refactor this function to async: ..."}],
    max_tokens=2048,
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")

TypeScript · openai SDK

import OpenAI from 'openai';

const client = new OpenAI({ baseURL: 'https://elderai.cloud/v1', apiKey: process.env.ATLAS_API_KEY, timeout: 300_000 });
const res = await client.chat.completions.create({
  model: 'atlas-code-1-preview',
  messages: [{ role: 'user', content: 'Write a Vitest test for a debounce() helper.' }],
  max_tokens: 1024,
});
console.log(res.choices[0].message.content);

Editors and agents

Use it in Cursor, Cline, Aider, or Codex

Cursor

Open Settings → Models → API Keys and paste your ATLAS key into OpenAI API Key.
Turn on Override OpenAI Base URL and enter https://elderai.cloud/v1.
Add the model atlas-code-1-preview, enable it, and pick it in Chat.

Cursor sends custom-model requests from its own servers. Features that use only Cursor’s own models, like Tab completion, won’t use ATLAS Code.

Cline (VS Code)

Set API Provider to OpenAI Compatible.
Set Base URL to https://elderai.cloud/v1, add your key, and set Model ID to atlas-code-1-preview.
Set Context Window to 24576 and Max Output Tokens to 8192. Leave image support off.

Aider

shell

export OPENAI_API_BASE=https://elderai.cloud/v1
export OPENAI_API_KEY=$ATLAS_API_KEY
aider --model openai/atlas-code-1-preview --timeout 300

Codex CLI

Codex uses the Responses API. ATLAS serves POST /v1/responses to every ATLAS Code account. Add this to ~/.codex/config.toml:

~/.codex/config.toml

model = "atlas-code-1-preview"
model_provider = "atlas"

[model_providers.atlas]
name = "ElderAI"
base_url = "https://elderai.cloud/v1"
env_key = "ATLAS_API_KEY"
wire_api = "responses"

Limits during the preview

Verified email required. Sign up and verify your email; ATLAS Code unlocks automatically with a free 7-day trial of 1,000 credits (one trial per person). No invite code needed.
Cold starts. The first request after a quiet period waits while a GPU worker starts: usually 1–2 minutes, up to 3. After that, answers start streaming in about a second. Use streaming: while a request waits, the stream gets a : keep-alive comment every 15 seconds, which OpenAI SDKs and Codex ignore. Set client timeouts to at least 300 seconds.
Shared capacity. Under load, requests may queue or return 429/503. Retry with backoff.
Hard caps. When your credits run out, requests stop. There are no overage charges.
No training on your code. We don’t use your prompts, code, or completions to train any model. See the Privacy Policy.