Skip to main content
Aider is a terminal AI pair-programming tool built around a git diff workflow. You can connect Aider to the ScitiX Model Inference API through its OpenAI-compatible integration.

​
Configure

​
1. Install Aider

​
2. Configure the model

Create an API Key on the API Keys page, then add the ScitiX configuration to ~/.aider.conf.yml. Aider reads this file on every launch and in every terminal.
Keep the following in mind:
  • This guide uses glm-5.2 as the example model. Reasoning models such as glm-5.2 may output reasoning content before the final response, which can feel slower. Switch to a non-reasoning model if speed matters more.
  • The openai/ prefix routes requests through Aider’s OpenAI-compatible integration. There is no SCITIX_-prefixed variant.
  • As a session-only alternative, you can set OPENAI_API_BASE and OPENAI_API_KEY, then run aider --model openai/glm-5.2. A new terminal session will lose these exports.
  • Instead of ~/.aider.conf.yml, OPENAI_API_BASE and OPENAI_API_KEY can also live in a .env file. Aider searches in the home directory, the git repo root, and the current directory. Later locations take precedence.
Launch Aider from your project directory:
Commit messages and chat-history summaries are generated by a separate weak model. Add --weak-model openai/DeepSeek-V4-Flash if you want to use DeepSeek-V4-Flash for this role.

​
3. Clear the unknown model warning

On first launch, Aider may show a warning similar to:
This is expected when Aider’s model database does not yet include ScitiX model IDs. Aider falls back to defaults and may suggest a similar model name. Answer N to the documentation prompt and ignore the model-name suggestion. To clear the warning, add a .aider.model.metadata.json file in your project root:
  • The litellm_provider field must match the prefix in the model name. Aider looks for this file in your home directory, the git repo root, and the current directory. Later locations take precedence. You can also pass --model-metadata-file <path> explicitly.
  • Adjust max_input_tokens in .aider.model.metadata.json to match the model’s actual context window.

​
Verify

In interactive mode, ask Aider to make a small change, such as adding a line to the README, and confirm that it generates and applies a diff. You can also verify the connection non-interactively:

​
Other Modes

The same ScitiX configuration also works with Aider’s other modes:
  • One-shot: --message "<task>" sends a single instruction, applies the result, and exits. This is useful for scripting.
  • Browser UI: aider --browser runs Aider in your web browser. This mode is experimental.
  • Watch mode: aider --watch-files watches your repo for AI comments. For example, write // add a factorial() function ai! in a file and Aider picks it up. AI! requests a change, while AI? asks a question.