Test Agent

Report Abuse

Basic Information

Test-Agent is an open source project that provides an intelligent assistant for software testing by combining large language models and engineering techniques from the quality domain. The repository packages a domain-tuned model named TestGPT-7B together with an engineering scaffold that lets teams run the model locally, expose a chatbot UI, and integrate model capabilities into testing workflows. It targets generation and augmentation of test artifacts rather than general chat, focusing on automated multi-language test case synthesis and automatic assertion completion to raise test quality. The project emphasizes private, on-premise deployment so models interact with local code and data without external leakage. The README documents model performance benchmarks, hardware and software prerequisites, and simple commands to start controller, model worker, and a web UI for interactive use.

Links

Categorization

App Details

Features
The repo ships the TestGPT-7B model fine-tuned from a CodeLlama-7B base for testing tasks and reports comparative metrics for pass@1 and scenario coverage across Java, Python and JavaScript. It supports two primary capabilities: multi-language test case generation and automatic Assert completion (currently for Java). An engineering framework contains a ChatBot web UI, a controller, a model worker for different devices, and instructions for local/private deployment. The QuickStart lists prerequisites such as Python 3.8+ and transformers==4.33.2 and indicates ~14GB GPU memory is needed. Device options and examples show how to run on mps, xpu, npu or cpu and how to start the web service to exercise sample test generation flows. The project is built on top of FastChat components.
Use Cases
Test-Agent helps development and QA teams automate repetitive testing tasks by generating readable multi-language unit tests and by filling missing assertions in existing tests, which increases coverage and defect detection without manual effort. Local deployment protects proprietary code and data while enabling an interactive web UI for rapid experimentation and iteration. The included engineering scripts simplify launching a controller, model worker and web service so teams can validate model outputs against code bases and integrate the assistant into test pipelines. Benchmark results give a baseline for expected model performance when choosing to adopt the tool. The repository and examples speed up adoption by providing device-specific options and minimal environment requirements so teams can evaluate and iterate quickly.

Please fill the required fields*