AI Agent Frameworks, MCP Servers and SDKs

Agent S is an open-source AI agent framework that controls ordinary desktop and web applications through the screen, mouse and keyboard. It runs on macOS, Windows and Linux and is aimed at developers, computer-use researchers and people building automation for their own desktops. You describe the task in natural language; the agent clicks, types and scrolls without requiring an API integration or a separate script for each application.

LLM is an Apache-2.0 command-line tool and Python library for sending prompts to local models and remote APIs. Local model support comes through plugins; cloud providers require their own API access. It can also connect to an arbitrary OpenAI-compatible Chat Completions endpoint, including LM Studio.

AG2 is an open-source Python framework for developers and researchers building systems where AI agents share work. The code uses the Apache 2.0 license.

Safetensors is a file format and library for developers who store, share, or load AI model weights on their own hardware or servers. It avoids the arbitrary code execution risk of PyTorch's pickle-based files while supporting fast access to tensor data. The project is open source under Apache 2.0, with a Rust implementation and Python support.

LangChain is an MIT-licensed open-source framework for developers building AI agents and applications powered by LLMs. It provides a shared interface for models, tools and data connections, so developers can change providers or test workflows without rebuilding the whole application.

Spring AI is an open-source Java framework for developers adding AI to Spring applications. It connects application data and APIs to models through a common interface, with Ollama support for local LLM use and integrations with cloud providers such as OpenAI, Anthropic and Amazon Bedrock. The framework runs within your application; your choice of model provider determines whether model requests stay local or go to a cloud service.

Crawl4AI is a self-hosted web crawler and scraper for developers building AI agents, retrieval-augmented generation (RAG) systems and data pipelines. It turns web pages into Markdown or structured JSON and runs as a Python library or a Docker server on your own hardware. The open-source code uses the Apache 2.0 license.

MCP Inspector is a locally run developer tool for testing Model Context Protocol (MCP) servers. It's for developers building or checking the servers that connect AI applications to tools and data. You can inspect a local server or connect to a remote HTTP endpoint.

Ollama JavaScript connects Node.js and browser applications to models running through Ollama. It's for developers building chat interfaces, AI agents or other apps that need a local LLM backend. The library is open source under the MIT license, with TypeScript types and an API that follows Ollama's REST interface.

Qwen Code is an Apache 2.0 licensed AI coding agent for developers who want help working through a codebase, changing code and checking the result. It runs on macOS, Windows and Linux, with a terminal interface and a desktop app. It builds on Google Gemini CLI and has developed into an agent that can use Qwen models alongside other model providers.

Google ADK is an open-source AI agent framework for developers building applications that carry out multi-step tasks. You can run agents locally in Docker or on your own infrastructure, and connect them to locally running models through adapters. The framework is optimized for Gemini but supports other models and providers. Its Python repository uses the Apache 2.0 license.

Wren AI is a self-hosted data agent for teams and agent builders who need answers based on agreed business definitions. It turns plain-language questions into SQL and interactive dashboards, using the same definitions when someone asks directly or through Claude, ChatGPT or Gemini over MCP.

llama-cpp-python brings llama.cpp model inference into Python applications and exposes it through a self-hosted OpenAI-compatible server. It's for developers building local AI applications or connecting existing API clients to models on their own hardware. The package is open source under the MIT license.

Transformers.js is a JavaScript library for developers building web apps that run AI models on the user's device. Inference happens in the browser, so an app doesn't need a separate model server to process its inputs. The library is open source under Apache 2.0.

aichat brings Ollama and cloud AI services into the same terminal interface for developers and people who work at the command line. It runs locally on macOS, Linux and Windows, with Android support through Termux. Model processing happens through the backend you choose: Ollama supports local models, while providers such as OpenAI, Claude and Gemini process requests in the cloud.

Guardrails AI is an open-source Python framework for developers who need to check what goes into an LLM and what comes back. It runs within your application or as a self-hosted service. The framework uses the Apache 2.0 license and helps address risks such as policy violations, hallucinations, and data leakage before outputs reach users.

Semantic Kernel is an MIT-licensed SDK for developers adding AI agents to their applications. It supports local models through Ollama, LMStudio and ONNX, alongside cloud services such as OpenAI and Azure OpenAI. You choose the model backend. The SDK runs on Windows, macOS and Linux and supports C#, Python and Java.

Inspect AI is a Python framework for researchers and developers testing language models and AI agents. Developed by the UK AI Security Institute and Meridian Labs, it evaluates coding, reasoning, knowledge, behavior and multimodal understanding, including tasks where agents must take actions to succeed.

Stagehand is an open-source browser automation SDK for developers building AI agents that interact with websites and extract structured data. It can run with Chrome on your own machine or use Browserbase's cloud browsers. Local runs require Chrome. The project uses the MIT license and supports TypeScript, Python, and Go.

Guidance is an MIT-licensed, open source Python library for developers who need language model output to follow a defined format. It works with local LLM backends including Transformers and llama.cpp, as well as OpenAI's cloud service. It's for application code.

mistral.rs is an open source inference engine for running models on your own computer or self-hosted server. It's for developers building AI applications and people who want local chat, multimodal models and agent tools in the same runtime. The Rust project uses the MIT license.

DocsGPT is an MIT-licensed, open-source platform for teams that want AI search, assistants and agents over their own documents. It can run on your servers with local models, including fully air-gapped deployments where documents and questions stay inside your network. Answers include the source title and page number so readers can check the evidence.

PDFMathTranslate translates scientific PDFs while keeping their page layout, formulas, charts, contents pages and annotations. It's for researchers, students and others who need to read papers in another language without losing the relationship between the text and its figures. It produces both translated PDFs and bilingual documents for comparison with the original.

PrivateGPT is a self-hosted API layer for developers building AI applications around local models. It adds document retrieval, database access and agent tools to an existing model server. Local workflows can work offline and keep data within your environment; web search and connections to online providers need internet access.

Letta is an open-source platform for AI agents that retain memory across conversations and use experience to improve over time. Formerly called MemGPT, it's for people building or using assistants that need continuity beyond a single chat. Agents can run locally or on a self-hosted server.

Local Deep Research is a self-hosted AI research assistant for people who need cited answers drawn from academic papers, the web and their own documents. It can produce a quick summary or pursue a complex question through repeated searches, then assemble a structured report. It's open source under MIT.

txtai is a Python framework for developers building search applications, chat with their data, and AI agents on their own hardware or servers. Its embeddings database combines sparse and dense vector search with graphs and relational data, so the same system can find related content and supply context to language models. It's open source under Apache 2.0.

DSPy is a Python framework for developers building AI applications whose tasks need clear inputs, predictable output types, and measurable results. You define what a language model should produce, then compose those tasks into a larger program. It's open source under the MIT license.

FastMCP is an open-source Python framework for developers connecting AI agents to their own tools and data through the Model Context Protocol (MCP). It supports locally running servers and connections to remote servers, with server development and client access in the same framework. It uses the Apache 2.0 license.

Jina Reader turns web pages and documents into text that LLMs can use, with Markdown or JSON output. It's for developers building AI agents, search tools and systems that answer questions using retrieved documents. You can self-host the Apache 2.0 service code in Docker or use Jina's hosted API.

Playwright MCP lets AI agents control a browser by reading structured page information rather than interpreting screenshots. It's for developers building browser automation, exploratory tests or agent workflows that need to keep a browser session open across repeated actions. The server runs locally on macOS, Windows and Linux, or as a self-hosted service.

PandasAI is a Python library for people who want to ask questions about their data in plain language. It works with SQL databases, CSV and parquet data, and can answer questions across multiple pandas DataFrames. Developers can use it in their own applications; analysts can use conversational queries to reduce the code they write for individual questions.

Megatron-LM is a Python framework for research teams training large language models on NVIDIA GPU infrastructure. It pairs ready-made training scripts with Megatron Core, a library developers can use to build their own training systems. Its focus is distributed training, with benchmarks on H100 clusters spanning thousands of GPUs.

smolagents is an open-source Python library for developers building AI agents that carry out tasks by writing and executing Python. Its CodeAgent can combine tool calls with loops, conditionals, and calculations in one action. This approach suits tasks that need several operations, rather than a single model response.

CrewAI is an open-source Python framework for developers building workflows with multiple AI agents on their own hardware or servers. It supports local models through Ollama and defaults to the OpenAI API for model requests. The framework uses the MIT license; a separate commercial platform provides managed deployment and governance, with on-premise and cloud options.

LlamaIndex is an MIT-licensed Python framework for developers building AI agents and apps that answer questions using their own data. It connects documents and other sources to language models, then helps an app find the relevant material when a user asks something. Its open source framework can work with models served through Ollama.