Aelius Code

Use Aelius as the model in your coding agent: Cline, Continue, Roo Code, opencode or Qwen Code. Free with any account.

Looking for the JSON API for your own software? That's the general API (ael_sk_ keys, reviewed by a person). You don't need it here.

Overview

Model

qwen3.8-27b

Context window

262,144 tokens

Protocol

OpenAI Chat Completions

Auth

Authorization: Bearer aelius_code_…

Images

Yes

Price

Free; Aelius Pro raises every limit

Getting a key

Free, instant, no review. You need to be signed in (sign up); guests can't hold a key.

1. Click your name at the bottom left

   On a phone, open the menu first.

   

2. Choose Aelius Code

   It's in the App group.

   

3. Name it and click Create a key

   

4. Click Copy

   It's shown once and can't be read back. Lost it? Revoke it and make another. Treat it like a password.

   

What your editor asks for

Qwen Code, step by step

An agent in a VS Code side panel. You need desktop VS Code and your key on the clipboard before step 9.

Install the extension

1. Open Extensions

   Ctrl Shift X (Cmd on a Mac), or the circled icon.

   

2. Search qwen code

   Pick Qwen Code Companion by qwenlm. Not the one called just "Qwen".

   

3. Click Install

   

4. Open the panel

   Click the Qwen icon at the top right. No icon? Ctrl Shift P → Developer: Reload Window.

   

Point it at Aelius

Pressing Esc on any prompt cancels; start again at step 5.

1. Type /auth and press Enter

   

2. Choose Custom Provider

   Last in the list.

   

3. Choose OpenAI Compatible

   

4. Paste the base URL

   https://code.aeliusai.com/v1

   Copy

   Ends at /v1, no trailing slash.

   

5. Paste your key

   Watch for a trailing space.

   

6. Paste the model name

   qwen3.8-27b

   Copy

   

7. Say hello

   It thinks first, then answers.

   

Then fix the context window

Qwen Code assumes 1,000k; the real window is 262,144. Without this, long sessions fail. Add this to ~/.qwen/settings.json (%USERPROFILE%\.qwen\settings.json on Windows), then restart VS Code:

{
  "modelProviders": {
    "openai": [
      {
        "id": "qwen3.8-27b",
        "baseUrl": "https://code.aeliusai.com/v1",
        "generationConfig": {
          "contextWindowSize": 262144
        }
      }
    ]
  }
}

It must sit on the provider entry, not under a top-level model key. Also turn off any request timeout: agent turns can take minutes.

If something's wrong

Qwen Code in a terminal

export OPENAI_BASE_URL="https://code.aeliusai.com/v1"
export OPENAI_API_KEY="aelius_code_YOUR_KEY_HERE"
export OPENAI_MODEL="qwen3.8-27b"

qwen

PowerShell: $env:OPENAI_BASE_URL="…" and so on. Apply the context window fix too.

Cline

1. Install Cline from the VS Code marketplace.
2. In its settings, set API Provider to OpenAI Compatible.
3. Fill in:

Base URL

https://code.aeliusai.com/v1

API Key

aelius_code_…

Model ID

qwen3.8-27b

Context window

262144

Continue

Add to ~/.continue/config.yaml:

models:
  - name: Aelius
    provider: openai
    model: qwen3.8-27b
    apiBase: https://code.aeliusai.com/v1
    apiKey: aelius_code_YOUR_KEY_HERE
    defaultCompletionOptions:
      contextLength: 262144
    roles:
      - chat
      - edit
      - apply

Don't add autocomplete: it fires on every pause and burns your allowance.

Roo Code

Same as Cline: provider OpenAI Compatible, the same values, "Use Azure" off.

opencode

Works the same in the desktop app, the terminal and opencode web.

1. Install opencode

   Desktop app: opencode.ai/download. Terminal:

   curl -fsSL https://opencode.ai/install | bash

   Copy

   Windows terminal: npm i -g opencode-ai.

2. Copy your key

   Settings → Aelius Code. No key? Get one.

3. Quit opencode, then run one line

   Paste your key when asked.

   Windows Mac and Linux

   irm https://aeliusai.com/code/opencode-setup.ps1 | iex

   Copy

   In PowerShell (Start → type powershell).

   curl -fsSL https://aeliusai.com/code/opencode-setup.sh | sh

   Copy

   In Terminal (Mac: Cmd + Space → terminal).

   It writes opencode's settings file and backs up the old one. Windows script · Mac/Linux script.

4. Open your project

   Desktop: Add project → folder → New session. Terminal: cd into it, run opencode. The model button should say Aelius Code:

   

   If not, click it and pick it (terminal: /models).

   

5. Ask it to do something

   

If something's wrong

export AELIUS_CODE_API_KEY="aelius_code_YOUR_KEY_HERE"
opencode auth login https://aeliusai.com

Anything else

Any tool that takes a base URL, key and model works. To test your key:

curl https://code.aeliusai.com/v1/chat/completions \
  -H "Authorization: Bearer aelius_code_YOUR_KEY_HERE" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.8-27b",
    "messages": [{"role": "user", "content": "Reply with the word ready."}],
    "stream": false
  }'

Endpoints

Standard OpenAI Chat Completions, with tools and streaming. Reasoning arrives as reasoning_content.

Legacy text completion.

The one model and its context length. Needs a key, so a bad key often shows up as an empty model list.

Limits

Credits are per account, across all keys. Request limits are per key.

Retry-After always has the exact wait. Your usage is in Settings → Aelius Code.

How credits work

Like Codex: a rolling five-hour window that starts at your first request, plus a weekly cap. Failed turns are free.

A typical agent turn costs about 1.14 credits, so a free window is roughly 130 turns. An agent running flat out can use that in half an hour.

The context window

262,144 tokens per request, prompt and reply together. GET /v1/models reports it, but some clients guess from the model name instead (Qwen Code shows 1,000k). If your editor shows a different number, set it by hand.

One allowance per network

While one free account on a network is at its limit, other free accounts on that network get network_quota_exceeded until it resets. Dodging limits with extra accounts, keys, VPNs or proxies breaks the Terms (§7.6). Sharing a connection with another user? Tell us.

Errors

Data and privacy

• Your code isn't stored or used for training.
• We log accounting only: time, token counts, duration, key, account, IP and user-agent. Never the prompt or answer.
• While a turn runs, its first 160 characters show on the operator dashboard, in memory only.
• Request records are deleted after 45 days.

Get a key · The text and vision API · Contact