Claude Code, Codex, Cursor and Open WebUI

Anything that takes an OpenAI- or Anthropic-compatible base URL and a key works with Aderra.

Format	Base URL	Key header
OpenAI	https://api.aderra.ai/v1	Authorization: Bearer
Anthropic	https://api.aderra.ai (no /v1)	x-api-key or Authorization: Bearer

Which key? An API key pays from the API wallet and has no plan rules. A terminal key, from the Connect page, pays from your plan’s credits (Plus and up). Both work in every tool below.

Claude Code

The easiest way is the aderra command, which installs and configures Claude Code for you, with Aderra’s /model list and image tools: see Claude Code and Codex.

To set it up by hand:

export ANTHROPIC_BASE_URL=https://api.aderra.ai
export ANTHROPIC_AUTH_TOKEN=ak_...
export ANTHROPIC_MODEL=deepseek-v4.1-flash-uncensored
export ANTHROPIC_SMALL_FAST_MODEL=deepseek-v4.1-flash-uncensored
claude

Set up by hand, Claude Code assumes a smaller context window than it does under aderra claude, so it compacts long sessions sooner. If a long session hits request_too_large, run /compact.

Codex

Use aderra codex: see Claude Code and Codex. Current Codex versions ignore OPENAI_BASE_URL, so a hand setup with environment variables does not work. Codex speaks the Responses API, which Aderra serves.

Cursor

In Cursor’s settings, under Models:

1. Turn on OpenAI API Key and paste your Aderra key.
2. Turn on Override OpenAI Base URL and enter https://api.aderra.ai/v1.
3. Add a custom model with an Aderra id, for example deepseek-v4.1-flash-uncensored.

Open WebUI and other chat apps

Add an OpenAI API connection with the base URL https://api.aderra.ai/v1 and your key. The model picker fills itself from GET /v1/models. The same settings work for LibreChat, LobeChat, Continue, Cline and other OpenAI-compatible clients.

Note

The API access box on the Connect page shows these settings with your details filled in, for the Anthropic SDK, the OpenAI SDK, Cursor, and Open WebUI & others.

Instructions on terminal keys

Requests made with a terminal key run with Aderra’s own agent instructions, written for coding tools, so the model works as Aderra in your terminal. This applies to terminal keys only; an API key gets your prompt untouched plus one safety line.

• The tool’s own system prompt (or Codex’s instructions) is kept.
• The tool’s identity line (“You are Claude Code…”, “You are Codex…”) is reworded to Aderra’s, and Claude Code’s security-refusal line is removed.
• Aderra’s instructions are added after the tool’s on every request that carries tools, and when the tool sends no system prompt. A side request with a system prompt and no tools (a title, a summary) runs with the tool’s prompt alone.
• On Chat Completions, Aderra’s instructions come first, as a system message.

Your own agents

Agent frameworks that accept an OpenAI or Anthropic client (the OpenAI Agents SDK, LangChain, LlamaIndex, the Vercel AI SDK and others) work by passing a client configured with the base URLs above. Tool calling works on all three chat endpoints. See Chat completions.

Aderra’s own coding agent runs on the same API: see Aderra Agent.