pocketcoder 1.0.0

AI-powered coding assistant optimized for local LLMs

PocketCoder

AI-powered coding assistant that works with any LLM — local or cloud.

Features

• Any LLM: Works with Ollama, OpenAI, Claude, or any OpenAI-compatible API
• Session Memory: Continues where you left off
• RepoMap: Understands your project structure
• Optimized for Local Models: Works great with 7B-14B parameter models

Quick Start

pip install pocketcoder
pocketcoder

On first run, a setup wizard will help you configure your LLM provider.

Supported Providers

Provider	Models	Setup
Ollama	qwen2.5-coder, llama3, etc.	ollama serve
OpenAI	gpt-4o, gpt-4o-mini	OPENAI_API_KEY
Anthropic	claude-sonnet, claude-haiku	ANTHROPIC_API_KEY
Any OpenAI-compatible	vLLM, LM Studio, etc.	Custom base URL

Commands

/help      - Show help
/model     - Change model
/files     - List added files
/add       - Add file to context
/drop      - Remove file from context
/undo      - Undo last change
/clear     - Clear conversation
/quit      - Exit

Configuration

Config is stored in ~/.pocketcoder/config.yaml:

provider:
  type: ollama
  base_url: http://localhost:11434
  default_model: qwen2.5-coder:7b

thinking:
  mode: smart
  show_reasoning: true

Development

git clone https://github.com/Chashchin-Dmitry/pocketcoder.git
cd pocketcoder
pip install -e .
pytest

Support

If you find PocketCoder useful, consider supporting the project:

Network	Address
ETH / USDT (ERC-20)	0xdF5e04d590d44603FDAdDb9f311b9dF7E5dE797c
BTC	bc1q3q25vw4jm8v4xe2g6uezq35q2uyn5jt6e00mj9
USDT (TRC-20)	TQj3X5nFQWqPEmRUWNFPjkaRUUFLxmCdok
SOL	5s5uP66VmnLMSApjq8ro639tXvSp59XEwQittzxF64mF

License

MIT