cloudwego/abcoder

deep, reliable and confidential coding-context

What it solves

ABCoder is a framework designed to improve the coding context provided to Large Language Models (LLMs), reducing hallucinations and increasing the precision of AI-assisted programming. It solves the problem of LLMs struggling to understand large, complex codebases by providing a structured, language-independent representation of code.

How it works

The framework introduces a Universal Abstract-Syntax-Tree (UniAST), a specification that converts code from various languages into a standardized, AI-friendly format. It includes a General Parser to transform code into UniAST and a General Writer to convert it back.

ABCoder operates as a Model Context Protocol (MCP) server, allowing AI agents (like Claude Code) to use a set of tools to navigate and analyze code repositories locally and precisely. It provides a multi-layer analysis chain (from repository to package, file, and specific AST nodes) to help theMs LLM understand dependencies and references.

Who it’s for

Developers building AI-assisted programming applications (such as code reviewers, optimizers, or translators) and software engineers using AI agents for codebase analysis and implementation.

Highlights

  • UniAST Specification: A language-independent, structural coding context for AI.
  • Code-RAG: Local, confidential Code-Retrieval-Augmented-Generation that supports both in-workspace and third-party libraries.
  • Claude Code Integration: Deep integration via custom slash commands (/abcoder:schedule, /abcoder:task, /abcoder:recheck) for an AST-driven workflow.
  • Multi-language Support: Supports parsing for Go, Rust, C, Python, JS/TS, and Java.
  • MCP Server Capability: Can be integrated into any AI agent that supports the Model Context Protocol.

Related

  • Project
    charlie12345/ROCmFPX

    An extension for llama.cpp that introduces AMD-optimized GGUF quantization formats and accelerated kernels to improve LLM inference speed and memory efficiency on Radeon GPUs.

  • Project
    ultracontext/ultracontext

    An open-source, real-time context infrastructure that synchronizes AI agent sessions, allowing users to share and version context across different AI tools.

  • Project
    ekimetrics/adaptive-chunking

    A framework for RAG pipelines that automatically selects the best chunking strategy for each document by evaluating it against five intrinsic quality metrics.

  • Project
    kraina-ai/srai

    A Python library for geospatial machine learning that provides tools to acquire vector data, divide areas into micro-regions, and embed them into vector spaces.