What is knabit-coding-agent?

Knabit Coding Agent is an AI powered coding assistant that runs locally on your machine and helps you read, write, refactor, debug, and execute code within your project. Its standout feature is a smart context management system designed to keep long conversations affordable. The problem it addresses is that AI language models have a fixed context window, a limit on how much text they can process at once. If a coding assistant naively sends the entire conversation history on every query, token usage and cost grow with every exchange until the limit is hit and the agent breaks. Knabit solves this by keeping only a configurable number of recent conversation turns in full detail, controlled by a WINDOW_SIZE setting, and using the language model itself to summarize all older turns into a compact block. Because important decisions, file names, and code changes are compressed rather than discarded, the agent retains meaningful history without escalating cost. No external memory system such as a vector database is required. The architecture has two parts: agent.py acts as the brain, managing conversation history and routing messages through a LangGraph state machine with tools attached, server.py acts as the hands, running as a FastMCP server that exposes tools for reading and writing files, running shell commands, searching code, and performing git operations. When you type a query, the agent forwards it to the Gemini model, which can then call the MCP server to take real actions on your filesystem. It is built with LangGraph and LangChain and requires a Google API key for Gemini along with the uv Python package manager to install dependencies.