Terminal AI Coding (Aider + Ollama)

AI pair programming in your terminal, entirely local. Aider is an open-source terminal-based AI coding assistant (44K+ GitHub stars) that edits files directly in your repo, creates sensible git commits, and works with any LLM. Connect it to Ollama for local inference, and you get a fully private alternative to Claude Code, Cursor, or GitHub Copilot.

What you get

• AI code edits in your terminal - describe a change, Aider edits your files
• Git-native - auto-commits with sensible messages, easy to diff/undo
• Codebase-aware - maps your entire project for context-aware changes
• 100+ languages - Python, JS, Rust, Go, C++, and everything else
• Lint + test integration - auto-fix issues found by your linters
• Fully private - all inference runs on your GPU, zero data leaves your machine
• $0/month - no API costs, no per-seat licenses

Architecture

For most users, Qwen2.5 Coder 14B is a great fit on a 24GB card. For faster responses, the Qwen3 Coder 30B A3B MoE model uses only 3B active parameters per token. Recommended GPU: RTX 3090 (best value 24GB) or RTX 4090.

Prerequisites

• Python 3.9+
• Ollama installed and running
• A coding model pulled in Ollama
• Git initialized in your project

Setup

Step 1: Install Aider

python -m pip install aider-install
aider-install

Step 2: Pull a Coding Model

ollama pull qwen2.5-coder:14b

Step 3: Start Ollama with Larger Context

Ollama defaults to 2K context window, which is too small for coding. Start it with 8K+:

OLLAMA_CONTEXT_LENGTH=8192 ollama serve

Step 4: Start Coding

cd /your/project
aider --model ollama_chat/qwen2.5-coder:14b

Aider will analyze your repo structure, create a git commit if needed, and present you with a chat prompt. Describe what you want to build or change.

Use it

Add a Feature

> Add a rate limiter to the API endpoint using a decorator pattern

Aider reads your codebase, finds the right files, and makes the edit. It then shows you the diff and asks if you want to commit.

Fix a Bug

> The login endpoint throws a 500 when the email field is empty. Fix it.

Aider finds the handler, adds validation, and commits the fix.

Refactor Code

> Extract the database logic from views.py into a separate repository layer

Aider creates the new file, updates imports, and moves the code.

Write Tests

> Write pytest tests for the user registration flow

Aider generates tests based on your existing code patterns.

Cost vs cloud

For a solo dev, the 3090 pays for itself in about 8 months vs Cursor Pro. For teams under IP/compliance restrictions, it's the only option that works.

Troubleshooting

• Aider can't connect to Ollama → Make sure ollama serve is running. Set OLLAMA_API_BASE=http://127.0.0.1:11434.
• Context too small → Always start Ollama with OLLAMA_CONTEXT_LENGTH=8192 or higher. Aider will also warn if the window is too small.
• Slow responses → Use a smaller quant (Q4_K_M) or a coding-specific model. The MoE Qwen3 Coder 30B A3B is fastest for its size.
• Model warnings → Aider shows warnings for unfamiliar models. You can suppress them by adding an .aider.model.settings.yml file.
• Aider won't install → Make sure pip is up to date: python -m pip install --upgrade pip.

Swap components

• Larger model → Qwen3 Coder 30B A3B on an RTX 4090 for sharper reasoning.
• Smaller/8GB GPU → Qwen2.5 Coder 7B or Llama 3.1 8B.
• VS Code GUI → Use Continue instead of Aider for an in-editor experience.
• Terminal agent → Try Open Interpreter for a more agentic approach.