Open-Source LLM Evaluation Tools

llmfit helps people choosing models for their own computers compare what will fit in memory and how fast it might run. It checks your hardware before recommending models, so you can assess candidates before downloading them.

Cua supplies desktops and automation tools for developers building AI agents that operate native apps and browsers. You bring your own agent and model; Cua provides the computer environment and ways to interact with it.

LangWatch helps AI teams test agent behavior before release and investigate production failures. Teams can reproduce a failed conversation and check how an agent handles it again.

HarnessX is a Python framework for developers building AI agents whose behavior they can reuse and adapt. It separates model choice from tool use, memory and context handling, so changing an agent's role doesn't require rewriting it.

MTEB helps researchers and developers compare embedding models and retrieval systems for the tasks they need. Its benchmarks cover multiple languages and modalities.

Garak is an LLM vulnerability scanner for developers and security teams assessing models or dialogue systems. It attempts to provoke unwanted responses so teams can identify weaknesses in the systems they test.

Mini SWE Agent is a command-line AI coding assistant for developers tackling GitHub issues and everyday coding tasks. Its small Python core also suits researchers who want to inspect and modify an agent for model evaluations.

Future AGI helps teams test AI agents, investigate production failures and compare prompt changes. It brings evaluation and monitoring into the same platform so teams can connect poor responses to the steps that caused them.

AutoGen is a framework for cooperating AI agents, with local Playwright browser automation, code execution and an experimental OllamaChatCompletionClient for local models. Its orchestration can run on your machine or across servers. Standard OpenAI and Azure OpenAI service connections send model requests to remote services; compatible endpoint overrides are separate configurations.

BotSharp is a C# framework for .NET developers building AI assistants into business applications. It coordinates agents with different responsibilities and keeps track of conversation state.

NeMo Guardrails is a Python toolkit for developers building AI assistants that need rules for what they say and which actions they take. It adds checks around conversations, document-based answers and tool use.

OSWorld is a benchmark for researchers and developers testing AI agents that use computers. It measures task completion in real desktop environments, including workflows that cross application boundaries.

ChainForge helps developers test how reliably LLMs respond before choosing a prompt or model for an application. Its visual flows turn prompt variations into repeatable comparisons rather than isolated chat experiments.

Oumi helps ML researchers and teams build specialized AI models for their own tasks. It brings data preparation, training, evaluation, and deployment into one platform, with reusable recipes and components you can customize.

Hypha is a TypeScript AI agent framework for developers building applications that need controlled tool use and recoverable work. It separates an agent's planning and tool choices from the execution layer that decides which actions may proceed.

ModelScope is an AI model hub with a Python library for developers and researchers who want to run models locally or adapt them to their own data. It covers language, vision, speech, multimodal tasks and scientific computing.

BAML is a programming language for developers building AI agents, with a documented Ollama client for models running locally and tracing that runs on your machine. The OllamaClient integration is documented for the canary toolchain. The standard OpenAI connection sends model requests to OpenAI; installing BAML locally does not make that connection offline.

UltraRAG is a self-hosted framework for researchers and developers building AI systems that answer questions using their own documents. Its Python application and Docker deployment provide a browser workspace for knowledge bases, workflow design and evaluation. Generation can use local models through vLLM or Hugging Face. Embeddings can use local Sentence Transformers or a configured OpenAI service; the OpenAI cloud connection sends embedding inputs to that service.

WebVoyager is a Python AI agent that controls a browser on your machine or a Linux server to carry out tasks on live websites. It's a research project for developers studying browser automation. Model decisions depend on OpenAI's cloud API, which receives page observations and requires an API key and internet access; the local browser doesn't provide local model inference.

ASSERT is a local-first Python framework for developers testing AI agents and LLM applications against their own requirements. It turns written expectations into evaluations for quality, safety, and regression testing.

RamaLama runs AI models for local chat or a self-hosted REST API. It's for developers who want a container workflow for trying models on their own hardware and serving them in production.

Ollama Grid Search helps developers compare model responses before choosing a model and prompt for a task. It runs experiments across combinations of models, prompts and inference settings, then presents the results for visual review.

FlagEmbedding is a Python toolkit for developers building semantic search and retrieval-augmented generation (RAG). It runs embedding models to find relevant text and rerankers to refine the order of retrieved results.

PyRIT is an open-source Python framework for testing generative AI systems, with a self-hosted CoPyRIT browser interface. Its backend serves the interface locally, and Docker deployment is available. Targets and scorers use configured endpoints, including Ollama or cloud providers. The documented OpenAI configuration sends requests to OpenAI; hosting the testing interface locally does not make every model local.

MLflow helps developers and teams inspect AI applications, compare changes and track model experiments. It brings agent debugging and evaluation into a platform you can run locally or on your own servers.

FlashRAG is an open source Python toolkit for researchers and developers comparing retrieval-augmented generation (RAG) methods or building their own. It combines implemented research methods with prepared benchmark datasets and reported results, so experiments don't have to start with a separate codebase for each approach. It uses the MIT license.

LazyLLM is a Python framework for developers building applications with multiple AI agents. It brings application workflows, model training and evaluation together so teams can refine a prototype using task-specific data.

AI Infra Guard is a self-hosted AI security testing platform for developers and security teams checking agent workflows, model responses and AI infrastructure. It combines vulnerability scans and adversarial prompt tests in a web interface.

Laminar helps developers inspect AI agent runs, find recurring failures and test fixes. It connects production traces with debugging and evaluation tools.

Kiln is a desktop workbench for teams building and testing AI applications. Engineers, QA staff and subject matter experts can work with the same tasks and datasets, including contributing ratings and examples without writing code.

Ferret is a local vision-language model for researchers studying how AI understands specific parts of an image. It supports questions about selected regions and can identify image regions in its answers. Its code and data are restricted to research use.

nanochat is a toolkit for researchers and developers who want to train an LLM and inspect the pipeline themselves. Its small Python codebase covers the stages between preparing a tokenizer and chatting with the resulting model.

Chidori is an AI agent framework for developers building long-running workflows that need crash recovery or human approval. Agents use ordinary async TypeScript, with saved runs that developers can revisit for debugging and testing.

Agno is a framework and runtime for developers building customer-facing or internal AI agents. It combines agent development, service hosting and a web interface for managing runs on infrastructure you control.