Aider Installation and OmniProxy Configuration
Want a tool to help modify code but worried that the changes will be difficult to review? Aider handles tasks step by step in the terminal and uses Git to record differences, making changes easy to inspect and undo.
Aider is a tool that helps modify code from the terminal. It builds a map of the entire codebase to help the model understand larger projects. Git is a version control tool that records code changes, allowing you to inspect differences and restore previous versions. After you give Aider a task, it processes the task step by step and writes commit messages. If you do not like the changes, you can undo them with /undo.
Before You Start
There are two common installation options:
- If your computer already has Python 3.8 through 3.13, install
aider-installfirst. Python is the runtime environment required by Aider. - If you do not have a suitable Python version, use the one-line installation command. It will also set up Python 3.12 when necessary.
Mac and Linux
install.sh
- Run with sh after curl
- Installs Python if needed
- Use wget if curl is unavailable
Windows
install.ps1
- Run in PowerShell
- Reopen the terminal after installation
- Set environment variables with setx
Python already installed
aider-install
- First install this package with pip
- Then run aider-install
- It creates its own environment
Make sure OmniProxy is running. After installation, enter the project directory you plan to modify before starting Aider.
Install and Launch for the First Time
If You Already Have Python 3.8 Through 3.13
Run these commands in order:
python -m pip install aider-install
aider-install
aider-install installs Aider in a separate Python environment so that it does not interfere with other Python projects on your computer. When necessary, it also installs Python 3.12 for Aider to use.
One-Line Installation
On Mac and Linux, run:
curl -LsSf https://aider.chat/install.sh | sh
If curl is not installed, use:
wget -qO- https://aider.chat/install.sh | sh
In Windows PowerShell, run:
powershell -ExecutionPolicy ByPass -c "irm https://aider.chat/install.ps1 | iex"
These one-line commands are based on uv, a program for installing and managing Python tools and environments. You can also use the following command:
uv tool install --force --python python3.12 --with pip aider-chat@latest
You can also use pipx. When installing directly with python -m pip install aider-chat, the supported Python versions are 3.9 through 3.12. However, this is an alternative installation method, so prefer the methods described above. Aider packages from system package managers may have dependency mismatches, so they should not be your first choice.
On 2026-10-09, the one-line installation command installed Aider v0.86.2. During installation, you may see a message saying that the cache and destination are on different disks and that files will be copied instead. The installation will still complete.
After installation, enter the project directory. A compatible API is an interface that communicates with other model services using the OpenAI format. Start Aider using this pattern:
aider --model openai/<model-name>
Replace <model-name> with the full model ID copied from OmniProxy in the next section. A model ID is the name used to identify a model precisely.
Get Connection Settings from OmniProxy
You need to set the following two environment variables. Environment variables can be understood as a set of connection settings read by terminal programs:
export OPENAI_API_BASE=https://your-omniproxy-origin.example/v1
export OPENAI_API_KEY=<YOUR_OMNIPROXY_API_KEY>
OPENAI_API_BASE is the API endpoint and must include /v1. OPENAI_API_KEY is the key used to access the API.
On Windows, use setx to save the variables. You must reopen the terminal afterward:
setx OPENAI_API_BASE https://your-omniproxy-origin.example/v1
setx OPENAI_API_KEY <YOUR_OMNIPROXY_API_KEY>
Copy the entire model name from the current OmniProxy list, including any slashes in the model ID. When starting Aider, add openai/ before the full ID.
As of 2026-10-09, the two reference IDs in the provider documentation are gemini-3.8-flash and grok-4.7. If these names are not in the current list, do not copy them directly. Use a full ID that actually appears in the list.
Address
OPENAI_API_BASE
- Point it to OmniProxy
- Include /v1
- Do not put it in the repository
Key
OPENAI_API_KEY
- Use the one provided by the gateway
- Reopen Windows terminals after changing it
- Keep provider keys in the gateway
Launch
openai/model
- cd to the project directory
- Prefix the model name with openai/
- Use the name from the gateway list
Enter the Tool Configuration
You do not need to create a separate configuration file. Enter the project directory, start Aider, and replace the model ID with a name that actually exists in the current OmniProxy list:
cd /path/to/your/project
aider --model openai/gemini-3.8-flash
To use the other example model, run:
aider --model openai/grok-4.7
When Aider encounters an unfamiliar model, it displays a warning. This warning does not by itself mean that the endpoint is incorrect. Check OmniProxy’s request logs to confirm whether the request appeared.
Aider only adds files already tracked by Git to the context. The context is the code and information the model can read during the current task. If the current directory is not yet a Git repository, Aider creates one, but files that Git does not track still cannot be read. Use git add to add the files and commit them before starting Aider.
Aider can still modify files. --no-auto-commits only disables automatic commits; it does not prevent file modifications. When using Aider for the first time, choose an unimportant repository and run git diff after the task to review the actual changes.
Do not put real keys in project files or public pages. Store provider keys only in OmniProxy. If the gateway does not validate keys, you can set OPENAI_API_KEY to a placeholder string used only on your local computer.
Verify with a Real Task
First, make sure the file you want Aider to read has been committed to Git. Then run the following command from the project directory. Replace the model ID with a full ID from the current OmniProxy list:
aider --model openai/gemini-3.8-flash --message "What is the first line of this file? Do not modify the file." --no-auto-commits
--message sends a task directly to Aider, while --no-auto-commits tells Aider not to create Git commits automatically during this run.
On 2026-10-09, Aider v0.86.2 successfully read the first line of a tracked file. If the file is not tracked by Git, Aider will say that it cannot see the content. This does not mean that the API endpoint is incorrect. After completing the query, check OmniProxy’s request logs to verify the request source and model ID.
Launch
aider launches
- Command is available
- Environment variables are active
- In the project directory
Task
It inspected the file
- Ask about a specific file
- Review its explanation
- Unfamiliar models may trigger a warning
Logs
Visible in the gateway
- Source is Aider
- Model ID matches
- The key was not entered on the page
Common Errors
aidercannot be found: Reopen the terminal first. If you installed it directly with pip, trypython -m aider.- Windows still connects to the old endpoint after changing the variables: You must reopen the terminal after running
setx. Only a new window will read the updated variables. - The model name does not begin with
openai/: The compatible API requires the formatopenai/<model-name>. - An unfamiliar-model warning appears: This is a normal warning and does not mean that the connection failed. Check whether OmniProxy received the request.
- The model ID is invalid: Copy the complete ID again from the current OmniProxy list. Keep the
openai/prefix and any slashes already present in the ID. - Aider says there is no file content: Make sure the file is tracked by Git. Files in an empty repository or untracked files are not added to the context.
Official References
For installation steps, see Aider installation. For environment variables and the model prefix, see OpenAI compatible APIs. The example model IDs can be found in the Gemini model list and on the xAI models page.
Replace every placeholder with values from your own OmniProxy configuration. Never publish a real key.
A saved result or HTTP 200 alone does not prove full Agent compatibility. Check the source and model in OmniProxy's call record.