Skip to main content

Aider

Aider pairs with you in the terminal to discuss and edit code. Its OpenAI-compatible provider can send requests directly to Aurora.

Select Inference Endpoint
Choose preferred region.

Install and prepare​

Install with the upstream installer, or follow the platform-specific instructions:

python -m pip install aider-install
aider-install
export AURORA_API_TOKEN="<your-token>"

Create the token using Getting started.

Configure and launch​

In the shell where you will run Aider:

export OPENAI_API_BASE=""
export OPENAI_API_KEY="$AURORA_API_TOKEN"
aider --model openai/deepseek/deepseek-v4-flash

Run the command from your project directory. The leading openai/ is Aider's LiteLLM routing prefix; deepseek/deepseek-v4-flash remains the Aurora model ID. Do not use the bare Aurora ID here, because provider auto-detection may choose the direct DeepSeek API instead.

To persist the model choice, merge this into .aider.conf.yml in the project:

model: openai/deepseek/deepseek-v4-flash

Keep the base URL and token in the launching environment. The URL already includes /v1 and Aider adds the chat path.

Verify file context​

In Aider, add a small existing file as read-only context and ask about it:

/read-only README.md
/ask Summarize README.md in three bullets.

Expect a streamed summary of the file. Substitute another file if needed. This tests file context; Aider's editing workflow differs from agents that use native file-reading tool calls. Use /add when you want Aider to edit files.

Troubleshooting​

  • Unknown model / pricing warning: Aider may lack metadata for a custom model. It is not proof of an API error. Use Aurora's catalog for limits and Portal for billing; configure metadata using Aider's advanced model settings if needed.
  • Wrong provider: Keep openai/ before the full Aurora model ID and set OPENAI_API_BASE (not OPENAI_BASE_URL).
  • Edits fail despite working chat: Check Aider's selected edit format and model suitability. Chat compatibility does not guarantee reliable patch output.
  • Helper calls fail: Review any configured weak/editor models; they need valid routes and model IDs too.

See shared troubleshooting, Aider's OpenAI-compatible setup, and advanced model settings.