Bookmarks
Top 10 GitHub Repos
- Microsoft Agent Framework
- An open, multi-language framework for building, orchestrating, and deploying production-grade AI agents and multi-agent workflows in Python and .NET. Microsoft built it for teams taking agents beyond prototypes into real deployments, with orchestration and workflow support baked in. It's the clearest first-party signal yet of where the agent framework landscape is consolidating.
- OpenAI Codex
- A lightweight coding agent from OpenAI that runs locally in your terminal, with IDE integrations for VS Code, Cursor, and Windsurf, plus a desktop app path and the cloud-based Codex Web. It remains the reference point for terminal-native coding agents and a backbone many agent tools build on or integrate with.
- Switchyard
- A Rust proxy and library for LLM traffic that routes requests across models and providers while preserving native OpenAI and Anthropic API compatibility. It records operational metrics and enables flexible model selection, benchmarking, and cost/performance tuning without rewriting application code.
- gortex
- A high-performance, graph-based code-intelligence engine for AI agents and IDEs supporting 257 languages and multi-repository setups, exposed via CLI, MCP Server, and API. It aims to expose only the information an agent needs, claiming token usage reductions of up to 50x — all fully local.
- OpenViking
- An open-source, self-evolving context database for AI agents from Volcengine that unifies agent memory, knowledge RAG, and skills in one store. It positions itself as the persistence and context layer for agents that need long-term memory and retrievable knowledge across sessions.
- AI-Infra-Guard
- A full-stack AI red-teaming platform from Tencent's Zhuque Lab that scans agents, skills, MCP servers, and AI infrastructure, and includes LLM jailbreak evaluation. It's one of the first tools to treat MCP servers and agent skills as first-class security surfaces.
- Apache Maka
- Apache Maka (Incubating) is a local-first AI agent workspace that inspects projects and runs tools behind a sandbox boundary, recording model messages, tool calls, tool results, permission decisions, and termination events as an append-only log. The audit-first design makes agent behavior inspectable and reproducible.
- Claude Plugins — Community
- Anthropic's read-only mirror of the community plugin marketplace for Claude Cowork and Claude Code; the repo's marketplace.json is the canonical list of community-contributed plugins. It's the discovery surface for extending Claude's agent tooling, with submissions routed through the official plugin directory.
- Ollama
- The de-facto local model runner — a single CLI and API for getting up and running with Kimi-K2.6, GLM-5.2, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma, and more on macOS, Windows, Linux, and Docker. It keeps trending because local inference remains central to agent and automation stacks.
- Microsandbox
- An easy, fast, local-first microVM runtime and library for running untrusted workloads — AI agents, user code, plugins, CI jobs, scrapers, and automation — with hardware-level isolation. Cross-platform on Linux and macOS, it gives agent developers a safe execution boundary without the overhead of full cloud sandboxes.
Honorable Mentions
- usestrix/strix
- Open-source AI pentesting tool whose autonomous agents find and fix app vulnerabilities; integrates with GitHub Actions to block insecure code on every PR.
- agent-substrate/substrate
- Google's high-density runtime and control plane for large-scale agent deployments, with full lifecycle management (not an officially supported Google product).
- PostHog/posthog
- Product analytics plus AI observability that captures all the context agents need; steerable from Slack, web, desktop, or MCP.
- unslothai/unsloth
- Desktop app to run and train LLMs and diffusion models locally, covering Qwen3.8, Kimi K3, Gemma 4, DeepSeek-V4, and FLUX.
- ml-explore/mlx-lm
- Python package for generating text and fine-tuning LLMs on Apple Silicon with MLX, including quantization and Hugging Face Hub integration.