Aider
Information
Aider is an AI pair programming tool that runs in the terminal and edits your code files directly. It is git-aware — it can read your repository, understand context, and commit changes automatically. Aider supports multiple LLM providers including Ollama (local), OpenAI, Anthropic, Google Gemini, and any OpenAI-compatible API.
Key features:
- Edits existing files in your codebase rather than only generating snippets.
- Architect mode separates the planning model from the coding model for better results.
- Works with local models via Ollama for fully private, no-cost operation.
- Automatic git commits with meaningful messages after each change.
Installation
pip install aider-chat
aider --version
Setup with Ollama (Local AI)
export OLLAMA_API_BASE=http://localhost:11434
export OLLAMA_MODEL=deepseek-coder:6.7b
Running Aider
aider --model ollama/deepseek-coder:6.7b
Open a specific project file:
cd your-project
aider main.py
Architect / Agent Mode
aider \
--model ollama/deepseek-coder:6.7b \
--architect \
--auto-commits \
--dirty-commits
Separate Planner and Coder Models (Recommended)
aider \
--model ollama/deepseek-coder:6.7b \
--architect-model ollama/llama3:8b \
--architect \
--auto-commits
Context Awareness
aider --map-tokens 2048
Model Configurations
Balanced
- Planner: llama3:8b
- Coder: deepseek-coder:6.7b
Lightweight
- Planner: mistral
- Coder: deepseek-coder:6.7b
High Performance
- Planner: llama3:70b
- Coder: deepseek-coder:33b
Alternative
- Planner: llama3
- Coder: codellama:13b
Full Recommended Command
aider \
--model ollama/deepseek-coder:6.7b \
--architect-model ollama/llama3:8b \
--architect \
--auto-commits \
--dirty-commits \
--map-tokens 2048
Notes
- Local models depend on your hardware.
- GPU recommended for best performance.
- Fully private, no API costs when using Ollama.
Supported Model Providers
OpenAI (GPT)
export OPENAI_API_KEY=your_key_here
aider --model gpt-4o
Models: gpt-4o, gpt-4o-mini, gpt-4.1
Anthropic (Claude)
export ANTHROPIC_API_KEY=your_key_here
aider --model claude-3-5-sonnet
Models: claude-3-5-sonnet, claude-3-opus, claude-3-haiku
Google (Gemini)
export GEMINI_API_KEY=your_key_here
aider --model gemini/gemini-1.5-pro
Models: gemini-1.5-pro, gemini-1.5-flash
OpenRouter (Multi-model gateway)
export OPENROUTER_API_KEY=your_key_here
export OPENAI_API_BASE=https://openrouter.ai/api/v1
aider --model openrouter/deepseek/deepseek-coder
Supports: DeepSeek, Mixtral, Claude, GPT, and many others.
Local via Ollama
export OLLAMA_API_BASE=http://localhost:11434
aider --model ollama/deepseek-coder:6.7b
Models: deepseek-coder, codellama, llama3, mistral
Custom / OpenAI-compatible APIs
export OPENAI_API_KEY=your_key_here
export OPENAI_API_BASE=https://your-endpoint.com/v1
aider --model your-model-name
Examples: Together.ai, Fireworks.ai, Groq, local proxies.
Multi-Model Setup
aider \
--model openrouter/deepseek/deepseek-coder \
--architect-model claude-3-5-sonnet \
--architect \
--auto-commits
Usage, tips and tricks
- Use strong reasoning models for planning (Claude, GPT, Gemini).
- Use coding-specialised models for implementation (DeepSeek, CodeLlama).
- Combine local + cloud for best performance/cost balance.
- OpenRouter provides the most flexibility across providers.
Aider is model-agnostic and works with cloud models (GPT, Claude, Gemini), local models (Ollama), and any OpenAI-compatible API.
| Model | Size | RAM (Q4) | VRAM (Q4) | Notes |
|---|---|---|---|---|
mistral |
7B | ~4–6 GB | ~4 GB | Very good lightweight |
llama3:8b |
8B | ~6–8 GB | ~6 GB | Best general-purpose |
deepseek-coder:6.7b |
6.7B | ~5–7 GB | ~5 GB | Good for coding |
codellama:13b |
13B | ~10–14 GB | ~10 GB | Decent for coding |
deepseek-coder:33b |
33B | ~24–32 GB | ~24 GB | Very powerful |
llama3:70b |
70B | ~48–64 GB | ~40+ GB | Server-level requirement |