Skip to content

Aider.md

Latest commit

 

History

History
81 lines (59 loc) · 2.15 KB

Aider.md

File metadata and controls

81 lines (59 loc) · 2.15 KB

tinyMem Aider Integration Guide

Aider is a popular AI pair programmer in the terminal. It works excellently with tinyMem via Proxy Mode.

Prerequisites

  1. Aider installed (pip install aider-chat).
  2. tinyMem installed and configured.
  3. A backend LLM running (Ollama, LM Studio) OR an API key for a cloud provider.

Configuration

1. Configure tinyMem Backend

Ensure .tinyMem/config.toml points to your actual model provider.

Example (Ollama):

[proxy]
port = 8080
base_url = "http://localhost:11434/v1"

2. Start tinyMem Proxy

tinymem proxy

Running Aider

Aider needs to be told to talk to localhost:8080 instead of the real API.

Option A: Command Line Flags (Recommended)

aider \
  --openai-api-base http://localhost:8080/v1 \
  --openai-api-key dummy \
  --model openai/rnj-1  # Prefix 'openai/' tells Aider to use generic client

Critical: You MUST use the openai/ prefix for the model name (e.g., openai/qwen2.5-coder or openai/gpt-4). This forces Aider to use its generic OpenAI client, which respects the custom API base. If you just say --model gpt-4, it might try to hit the official OpenAI API directly.

Option B: Environment Variables

export OPENAI_API_BASE=http://localhost:8080/v1
export OPENAI_API_KEY=dummy

aider --model openai/your-model-name

Configuration Reference

For full configuration options, see Configuration.md.

Troubleshooting

"Unknown context window size"

Aider might not know the context limit of a local model proxied through tinyMem. Create a .aider.model.metadata.json file in your project root:

{
    "openai/qwen2.5-coder": {
        "max_tokens": 32768,
        "input_cost_per_token": 0.0,
        "output_cost_per_token": 0.0,
        "litellm_provider": "openai",
        "mode": "chat"
    }
}

Then run aider with:

aider --model-metadata-file .aider.model.metadata.json --model openai/qwen2.5-coder ...

Connection Error

  • Ensure tinymem proxy is running.
  • Check tinymem doctor.
  • Try using 127.0.0.1 instead of localhost if on Windows/WSL.