← GitHub Repos
GitHub Repos 2026-06-01

Top 10 GitHub Repos — 2026-06-01

Top 10 GitHub Repos — 2026-06-01

Top 10 GitHub Repos — 2026-06-01

As of: 2026-06-01 Focus: AI tools, automations, agents, integrations, MCP, CLIs, and developer tooling. Snapshot: Current top 10 at the moment of generation — not a weekly recap.


Overview

Two threads dominate this snapshot: the agent-harness arms race (optimization layers, plugin marketplaces, and full agent operators all stacking stars) and a sharp turn toward self-hosted, local-first AI surfaces — workspaces, inference engines, and context tooling you run on your own hardware. The standout newcomer is odysseus, a self-hosted ChatGPT/Claude-style workspace that cleared ~18k stars in roughly a day, while antirez and Vercel Labs both shipped notable new low-level pieces (ds4 local inference, zerolang agent language). Picks are all fresh to this digest series (no repeats from May). Note: the serper search/news API returned HTTP 400 throughout this run, so discovery and every star count were sourced directly from the GitHub MCP at generation time.


1. ECC

An agent-harness performance-optimization system layering skills, "instincts," memory, security, and research-first development on top of Claude Code, Codex, OpenCode, Cursor, and other harnesses. Created 2026-01-18, it has rocketed past 200k stars — the single highest-starred AI repo in this snapshot — riding the wave of developers trying to squeeze more reliability and structure out of their coding agents.

Why it's on the list: The most-starred agent-tooling repo on GitHub right now — a harness-agnostic optimization layer the whole coding-agent crowd is piling into.


2. LobeHub

Formerly LobeChat, now repositioned as a "Chief Agent Operator" — a surface for hiring, scheduling, and reporting on a whole team of agents running 7×24. It keeps a large knowledge-base, MCP, and multi-model (Claude, GPT, Gemini, DeepSeek) feature set, and the rename signals the broader shift from single-chat UIs to agent-fleet management. Active development continues on the canary branch.

Why it's on the list: A mature, heavily-starred chat UI pivoting hard into agent-fleet orchestration — a bellwether for where AI front-ends are heading.


3. oh-my-openagent (omo)

Pitched as "the pickaxe for complex software engineering" — an agent harness purpose-built for large, gnarly codebases, with a TUI and broad model support (Claude, GPT, Gemini, OpenCode, Codex). Created 2025-12-03, it has climbed past 60k stars by targeting the exact pain point that generic agents struggle with: navigating and editing genuinely complex repos.

Why it's on the list: A fast-rising harness aimed squarely at hard, real-world codebases rather than toy demos.


4. agents

A multi-harness agentic plugin marketplace spanning Claude Code, Codex CLI, Cursor, OpenCode, and Gemini CLI — subagents, skills, commands, and orchestration workflows in one installable collection. As teams standardize on reusable agent components, a cross-harness marketplace is exactly the kind of glue that keeps accumulating stars and contributions.

Why it's on the list: The closest thing to a package manager for agent skills and subagents across every major coding harness.


5. Odysseus

A self-hosted, local-first AI workspace — the DIY answer to the ChatGPT/Claude UI — bundling chat, an OpenCode-based agent, deep research, model "Cookbook" (hardware-aware model recommendation + serving via vLLM/llama.cpp/Ollama), persistent memory/skills (ChromaDB), plus email, calendar, notes, and a mobile PWA. Created 2026-05-31, it cleared ~18k stars almost overnight — the breakout new repo of this snapshot.

Why it's on the list: The fastest-rising new project right now — a privacy-first, run-it-yourself workspace that replaces a stack of hosted AI tools.


6. ds4

A DeepSeek 4 Flash local inference engine for Metal and CUDA, from antirez (creator of Redis). Created 2026-05-06, it's a lean, readable C implementation in the spirit of his other from-scratch projects — and it's surged on both the model's release and the author's reputation for clear, hackable systems code. A high-signal pick for anyone running DeepSeek models locally.

Why it's on the list: A from-scratch local inference engine for a hot model, written by one of open source's most respected systems engineers.


7. zerolang

"The programming language for agents," from Vercel Labs. Created 2026-05-15, it's an early, experimental take on a purpose-built language/runtime for expressing agent behavior — a notable bet that agents need their own primitives rather than being bolted onto general-purpose languages. Worth watching precisely because of who's behind it.

Why it's on the list: A first-party Vercel Labs experiment in giving agents a native language — an early signal of where agent runtimes may go.


8. semble

Fast, accurate semantic code search for agents, shipped as an MCP server — and pitched as using ~98% fewer tokens than the usual grep-then-read loop. Created 2026-04-06, it's climbing fast because token-efficient retrieval is one of the highest-leverage optimizations for any coding agent, and it drops straight into Claude Code / Cursor via MCP.

Why it's on the list: Token-frugal code search over MCP — a direct fix for the most expensive habit coding agents have.


9. SkillOpt

A Microsoft research project that treats agent skills as something to optimize: a text-space optimizer that trains reusable natural-language skills for frozen LLM agents via trajectory-driven edits, validation-gated updates, and deployable best_skill.md artifacts. Created 2026-05-08, it's an early but credible take on self-improving skills without touching model weights.

Why it's on the list: A first-party Microsoft approach to self-evolving agent skills — turning prompt/skill engineering into a measurable optimization loop.


10. agent-device

A CLI that lets AI agents drive real iOS and Android devices/emulators — ADB, simulators, XCUITest, Appium-style automation exposed for agentic control and E2E testing. Created 2026-01-30, it fills an obvious gap: most agent tooling targets the web or the shell, leaving mobile largely untouched. A clean primitive for agent-driven mobile QA and automation.

Why it's on the list: Brings mobile devices into reach for agents — a new modality (mobile control/testing) most harnesses can't touch today.


Honorable Mentions

  • open-multi-agent/open-multi-agent — TypeScript-native multi-agent orchestration that compiles a goal into a task DAG (new, 6,300★).
  • diegosouzapw/OmniRoute — Free AI gateway: one endpoint, 160+ providers, token-saving compression and auto-fallback (new, 5,622★).
  • holaboss-ai/holaOS — Local-first "super agent for work" that learns your working context in minutes (new, 5,482★).
  • TanStack/ai — Type-safe, provider-agnostic TypeScript AI SDK for chat, tool calling, and agents (2,708★).
  • yvgude/lean-ctx — "Context OS" binary with 63 MCP tools and up to 99% token savings across Cursor/Claude Code/Copilot (2,330★).

Sources

More from GitHub Repos