DevLoop: Autonomous AI Debugging & Code Repair

Software debugging is often slow and fragmented. Developers must move between failure logs, stack traces, source files, Git history, test results, and manual code changes just to resolve a single issue. This becomes especially difficult when debugging unfamiliar or rapidly changing codebases. DevLoop addresses this by turning debugging into an autonomous, structured repair workflow. A developer submits a bug report along with a Git repository and test command. DevLoop creates an isolated workspace, analyzes failure evidence and repository files in parallel, inspects recent Git changes, forms a root-cause hypothesis, generates a candidate patch, validates it, and pauses for human approval. After approval, the patch is applied only to the isolated workspace and the tests are executed again. Failed verification can trigger another repair iteration, and the workflow ends with a structured resolution report. Our target users are software developers, engineering teams, and students who want to reduce repetitive debugging work while maintaining control over code changes. Users interact through a web dashboard where they can submit repositories and issues, watch each agent's progress in real time, inspect evidence and proposed diffs, approve or reject patches, and review final test results and resolution reports. DevLoop is built as a multi-agent workflow using LangGraph, with specialized agents for log analysis, file identification, hypothesis generation, patch generation, and reporting. It uses Ollama with Qwen locally and supports Git-based repositories as well as local Git repositories during development. What makes DevLoop distinctive is its combination of autonomous reasoning with deterministic safety checks and human oversight. The system does not blindly trust model-generated file paths or patches. Repository evidence is checked against the actual workspace, patch targets must correspond to real files, candidate patches are validated with git apply --check, and tests are run in disposable copies before approval. This creates a practical balance between automation and developer control. We also used IBM Bob during development to build and iterate on the application, including the debugging workflow, agent logic, testing, and developer dashboard. By combining agent-based reasoning, repository-aware validation, isolated execution, automated testing, and human approval, DevLoop turns a traditionally manual debugging process into a faster and more repeatable repair loop.