GitHub Repos

The dominant deep learning framework: Python tensors and neural networks with strong GPU acceleration.

The standard Python library for defining, loading, training and running open models for text, vision and audio.

C/C++ engine for running LLMs locally on CPUs, Apple silicon and most GPUs, home of the GGUF model format.

Node-based interface and backend for running image, video and audio diffusion models locally.

Code for Sebastian Raschka's book on building a GPT-style LLM in PyTorch, step by step.

Rust vector database with fast filtered search, hybrid sparse-dense retrieval and quantization, self-hosted or managed.

Skills plugin that makes coding agents brainstorm, plan, test first and review before they ship, across 16 agent harnesses.

High-throughput LLM serving engine for GPUs, with an OpenAI-compatible server and wide model support.

Open-source tracing, evaluation and prompt management for LLM apps, free to self-host and owned by ClickHouse since January 2026.

Low-level framework for building stateful agents as graphs, with durable execution, memory, streaming and human-in-the-loop.

One-command tool for downloading and running open LLMs locally, with a REST API and optional paid cloud models.

Self-hosted ChatGPT-style web interface for Ollama, OpenAI-compatible APIs and other model providers.

MIT-licensed terminal coding agent that works with almost any model provider, from the team behind SST.

Postgres extension that adds vector columns and HNSW or IVFFlat indexes, so embeddings live next to the rows they describe.

Faster, lower-memory fine-tuning for open LLMs, now with a desktop app and web UI for running and training models.

Python library that lets LLM agents control a real web browser to click, type and extract data from sites.

Visual platform for building LLM apps, agent workflows and RAG pipelines, self-hosted or on Dify Cloud.

IBM-born toolkit that converts PDFs, Office files, HTML and images into clean Markdown or JSON for RAG, with strong table handling.

Small, composable skills such as /grill-me, /grill-with-docs and /tdd that keep you in charge of the process instead of a framework.

OpenAI's open-source terminal coding agent, written in Rust and included with ChatGPT plans.

CLI and library for testing prompts, RAG apps and agents, plus red teaming for security issues. Now owned by OpenAI.

Type-safe Python agent framework from the Pydantic team, with composable capabilities, durable execution and Logfire tracing.

Hugging Face's library for post-training LLMs: supervised fine-tuning, DPO, GRPO and other reinforcement learning methods.

Anthropic's open-source sandbox behind Claude Code's /sandbox: it fences any process off from your files and network.

Anthropic's reference Agent Skills, including the docx, pdf, pptx and xlsx skills behind Claude's file features, plus the spec.

CLI that reads coding agents' local usage logs and reports tokens and estimated cost by day, session or 5-hour block.

Open-source coding agent for VS Code, JetBrains and the terminal that asks before each change; bring your own model.

Hugging Face's Python library for running and training diffusion models for image, video and audio.

The UK AI Security Institute's framework for LLM and agent evaluations, used for safety and capability testing.

MIT-licensed, self-hosted ChatGPT-style app for many model providers, with agents, MCP and enterprise sign-in.

EleutherAI's framework for running standard academic benchmarks against language models.

TypeScript framework for agents and workflows, with memory, RAG, evals, tracing and a local Studio for testing.

Official Python SDK for building Model Context Protocol servers and clients, rewritten for the 2026-07-28 spec.

Distributed vector database for billion-scale search, now able to query vectors in Parquet, Lance and Iceberg files where they sit.

Apple's array framework for machine learning on Apple silicon, now with a CUDA build; the engine behind mlx-lm and many Mac LLM apps.

Fair-code workflow automation with 1,500+ integrations and AI agent nodes; self-host free for internal use or pay per execution.

OpenAI's lightweight Python framework for agents with tools, handoffs, guardrails, sessions and sandboxed workspaces.

Self-hosted control centre for coding agents: runs OpenHands, Claude Code, Codex and other ACP agents locally, on VMs or in the cloud.

Fast serving framework for LLMs and multimodal models, built around RadixAttention prefix caching.

A security firm's audit methods, packaged as 46 plugins for Claude Code and Codex.

25 engineering-workflow skills and 9 lifecycle commands, from spec to ship, written by Addy Osmani for any coding agent.

All-in-one desktop and Docker app for chatting with your documents using local or cloud models, with agents and MCP.

Config-driven fine-tuning tool for open LLMs, from single GPUs to multi-node clusters.

Anthropic's plugin catalogue that Claude Code adds by default: 315 entries, mostly vendor plugins, with context-cost estimates.

NVIDIA's open-source vulnerability scanner for language models: point it at an endpoint and it fires hundreds of known attacks.

Zep's open-source framework for temporal knowledge graphs that track when facts became true and when they stopped being true.

Frontend design skill with 24 commands and a 62-rule detector that flags generic AI-looking UI, by Paul Bakaus.

Self-hosted gateway and Python SDK that puts 100+ model providers behind one OpenAI-style API, with keys, budgets and fallbacks.

Fine-tuning framework for 100+ LLMs and vision-language models, with a web UI as well as a command line.

Python framework for connecting LLMs to your documents, with ingestion, indexing, retrieval and event-driven agent Workflows.

Datalab's PDF-to-Markdown converter that reads the text layer first and calls its OCR model only where a page needs it.

Andrej Karpathy's minimal codebase for training a small ChatGPT-style model end to end on one GPU node.

Packs a whole repository into one AI-ready file, with token counts, secret filtering and optional code compression.

GitHub's toolkit for spec-driven development: constitution, spec, plan, tasks, implement and converge commands for 40+ coding agents.

Browserbase's browser SDK for TypeScript, Python and Go that mixes Playwright-style code with natural-language act and extract steps.

Ostris's training toolkit for fine-tuning image and video diffusion models, mainly LoRAs, with a web UI.

OpenTelemetry-based tracing and evaluation workbench from Arize, source-available under the Elastic License and now part of Dynatrace.

Pure-C engine that runs huge mixture-of-experts models on ordinary PCs by streaming experts from disk, plus a Jev-compatible decision API.

Open-source Python crawler on Playwright that turns pages into LLM-ready Markdown, now with a paid hosted cloud.

Python framework for unit-testing LLM apps with ready-made metrics, in a pytest-like style.

Web scraping and crawling API that turns websites into clean Markdown or structured data for LLMs, self-hosted or cloud.

Microsoft's free 21-lesson course on building generative AI apps, with Python and TypeScript samples that lean on Azure.

Google's open-source framework for building, testing and deploying agents, with a graph-based workflow runtime since 2.0.

deepset's open-source Python framework for production RAG pipelines and agents, built from explicit, typed components.

Open-source desktop app for running LLMs offline, with an OpenAI-compatible local API and optional cloud models.

Python and TypeScript framework for building LLM agents, with a standard agent loop, middleware and hundreds of model integrations.

Maxime Labonne's free roadmap for learning LLMs, from fundamentals to fine-tuning and deployment, with Colab notebooks.

Microsoft's agent SDK for Python and .NET, merging AutoGen and Semantic Kernel into one framework with stable 1.0 APIs.

An open-source desktop workspace for running several coding agents side by side, each in its own git worktree.

Microsoft's Python framework for automated and human-led red teaming of generative AI, strongest on multi-turn attacks.

Anthropic's Claude Code plugin that reviews what Claude writes three times: on every edit, at the end of every turn and on every commit.

Go vector database with first-class hybrid keyword-and-vector search, built-in embedding modules and multi-tenancy, self-hosted or cloud.

Python framework and AgentOS runtime for running agents with memory, knowledge and tools on your own infrastructure.

Cross-platform desktop client for many cloud model providers and local servers, with knowledge bases, MCP and an agent.

Open-source vector database that runs embedded in Python or as a server, popular for prototypes and small RAG apps.

Python framework for teams of role-playing agents (Crews) plus event-driven Flows for predictable production pipelines.

Extensible local AI agent from Block, now under the Linux Foundation's Agentic AI Foundation, with desktop and CLI apps.

A /graphify skill that parses your codebase and docs into a local knowledge graph the agent queries instead of grepping files.

Polished multi-model chat and agent workspace, formerly LobeChat, available self-hosted, as a desktop app or on LobeHub Cloud.

Long-established graph database with Cypher queries, built-in vector search and a GPL Community Edition.

NVIDIA's runtime that puts each AI agent in its own sandbox and checks every file, system call and connection against a policy.

Open-source OpenTelemetry instrumentation for LLM calls and vector databases, sent to whichever backend you already run.

Comet's Apache 2.0 platform for tracing, evaluating and optimising LLM apps and agents, with the full server free to self-host.

Stacklok's platform that runs each MCP server in its own locked-down container, with access policy and audit logs for teams.

An agent skill that explains technical ideas as detailed, interactive isometric machines in SVG, with optional WebGL effects.

A guard that reads what a coding agent is about to run and refuses destructive commands and secret-reading before they happen.

Local gateway that routes Claude Code, Codex and other agents to any model provider, with fallbacks, key pools and cost logs.

OpenAI's official plugin to run Codex reviews and hand off tasks from inside Claude Code, using your local Codex CLI.

MIT-licensed coding agent for VS Code, JetBrains and the CLI, with agent modes and 500+ models at provider rates.

Embedded vector database on the Lance columnar format, built for multimodal data with versioning and no server to run.

Memory layer for AI agents that extracts, stores and retrieves facts about users across sessions, open source or hosted.

About 100 lines of Python that solve GitHub issues, from the Princeton and Stanford team behind SWE-bench and SWE-agent.

NVIDIA's toolkit for programmable checks on LLM inputs, outputs, retrieval and tool calls, from topic control to jailbreaks.

DAIR.AI's guide to prompt engineering, context engineering, RAG and agents, also published as a website.

Self-hosted RAG engine with layout-aware document parsing, a web UI and an agent builder, midway through a rewrite in Go.

NVIDIA's library for optimized LLM inference on its GPUs, with a Python API and production runtimes.

Community plugin marketplace of 94 plugins with 202 agents and 185 skills you install one domain at a time.

Data-movement platform with 600-plus connectors that syncs SaaS apps and databases into warehouses and vector stores, self-hosted or cloud.

A skill that turns a coding agent into a six-phase security auditor, with a separate agent trying to disprove every finding.

Confident AI's red-teaming framework for LLM apps and agents, with an OWASP Top 10 preset and dozens of agent-specific vulnerabilities.

A browser-based 3D globe of live public data, from aircraft and ships to satellites and cameras, steered by a voice AI agent.

Local image generation app with a layered canvas, aimed at artists, now maintained by the community.

One CLAUDE.md with four rules (think first, keep it simple, change surgically, define success) drawn from Andrej Karpathy's post.

Graph-plus-vector RAG framework from the University of Hong Kong, a lighter alternative to Microsoft GraphRAG with incremental updates.

Packages an LLM and its runtime into a single executable file that runs on most operating systems without installing anything.

Agile framework that runs product, architecture, UX, dev and test roles as agents, from brief to stories to code.

Claude Code plugin that files sources into a linked, cited Obsidian wiki and answers from it, based on Karpathy's LLM Wiki idea.

Generates command-line wrappers around real desktop software so AI agents can drive Blender, GIMP, LibreOffice and more.

Rust CLI that turns a codebase into one LLM prompt, with Handlebars templates, glob filters, git diffs and token counting.

Open-source memory layer that turns documents, code and conversations into a knowledge graph agents can search.

Large agent-harness kit with 71 agents, 311 skills, hooks, rules and a security scanner for Claude Code, Codex and more.

MIT-licensed visual builder for AI agents, RAG pipelines and MCP servers, owned by IBM via DataStax.

Microsoft's lightweight Python converter that turns Office files, PDFs, HTML and more into Markdown for LLMs.

Unofficial Python API, CLI and agent skill that lets Claude Code drive Google NotebookLM (now Gemini Notebook).

A scanner for the MCP servers, tools and agent skills installed on a machine, looking for tool poisoning and hidden prompt injections.

Python library that partitions many document types into typed elements (titles, text, tables) for RAG and ETL pipelines.

The vLLM project's router that reads each request and picks, or combines, the model to answer it, with built-in safety and PII classifiers.

Small, fast Python library that splits text into chunks for RAG, with eleven strategies from fixed-size to LLM-guided.

Cisco AI Defense's scanner that checks agent skills for prompt injection, data theft and malicious code before your agent reads them.

Plugin that records what your agent does, compresses it with AI and injects relevant memory into later sessions.

Local semantic code search and call-graph tracing for AI agents, as a single binary with an MCP server.

Small, fast TypeScript gateway for 1,600+ models with routing, retries and guardrail hooks, now owned by Palo Alto Networks.

Pioneering terminal pair-programmer that edits your git repo with any model; development has slowed sharply in 2026.

Keeps bulky tool output out of a coding agent's context and rebuilds the working state when a long session compacts.

Python runtime for the Laya decision models: a language router, a Jev-compatible server, framework hooks and free fine-tuning notebooks.

Python library that scrapes by prompt: describe the fields you want and an LLM pulls them out as JSON.

Weights & Biases' toolkit for tracing and evaluating LLM apps and agents, now part of CoreWeave Forge and partly succeeded by Agent Lens.

Microsoft Research pipeline that builds a knowledge graph and community summaries from documents to answer whole-dataset questions.

Hugging Face's minimal Python library for agents that act by writing and running code instead of JSON tool calls.

A local proxy that points Claude Code, Codex and other coding agents at free and cheap models from other providers.

Python library of LLM-judged metrics for RAG pipelines, such as faithfulness and context precision; unmaintained since February 2026.

Rust CLI proxy whose hook rewrites agent shell commands to compress their output; a JetBrains benchmark found no bill savings.

Google's open-source Gemini terminal agent, now closed to free users and being superseded by Antigravity CLI.

Python library of validators for LLM inputs and outputs; its hub closed in 2026 and the team joined Harvey.