AutoCodeRover is an open-source autonomous coding agent designed to resolve GitHub issues and fix bugs in complex codebases. Developed by researchers at the National University of Singapore (AutoCodeRoverSG), the tool focuses on automated program repair (APR) and structured software engineering tasks, featuring native integration with benchmarks like SWE-bench. The agent operates as a command-line harness that ingests an issue description and a target repository. Rather than relying strictly on unstructured file browsing or raw text grep, AutoCodeRover uses Abstract Syntax Tree (AST) based search APIs to retrieve classes, methods, and code contexts. It iteratively localizes faults, constructs an understanding of the surrounding logic, formulates a patch, and verifies fixes against available test suites using LLM backends such as OpenAI GPT models. What differentiates AutoCodeRover from general-purpose CLI coding assistants is its syntax-aware fault localization architecture. By querying code structure at the method and class level rather than injecting entire files into the prompt, the agent reduces context overhead and focuses token budget directly on relevant call graphs and code units. AutoCodeRover is released as open-source software under the MIT license, making it suitable for benchmarking, academic research, and automated patch-generation workflows in isolated containerized environments.
Tags: ai agent, cli, code assistant, codebase, coding agent, OpenSource