10 Hot Open‑Source AI Projects on GitHub This Week (Last One Praised by Jensen Huang)

1. Hermes Agent: Self‑evolving General AI Agent Framework

Hermes Agent, released by Nous Research under the MIT license, gained nearly 20,000 stars in a single week, becoming one of the fastest‑growing AI projects on GitHub.

Core capabilities: It features a self‑improvement loop that evaluates task outcomes, encodes successful experiences as reusable Skills, and supports persistent cross‑session memory, enabling continuous user profiling and context continuity. The framework integrates with 15+ platform gateways (CLI, Telegram, Discord, Slack, WhatsApp) and over 200 models via OpenRouter, offering 40+ built‑in tools.

Architecture & value: Implemented in Python, it provides tool invocation, a secure sandbox, and scheduling mechanisms. Its cross‑session memory and self‑improvement directly address key pain points in Agent development, especially for long‑term collaborative scenarios. Tencent Cloud launched a Lighthouse template for one‑click cloud deployment, lowering the entry barrier.

Project URL: https://github.com/NousResearch/hermes-agent

2. Evolver: Genome‑Evolution‑Protocol‑Based AI Agent Self‑evolution Engine

Evolver, open‑sourced by the Chinese team EvoMap on 2026‑02‑01, recorded over 36,000 downloads in three days, connected to more than 130,000 AI Agent nodes and accumulated 46 million calls.

Core capabilities: Leveraging genetic programming, Evolver lets Agents autonomously optimize prompts and action strategies, producing traceable and reusable evolutionary paths. Unlike Hermes Agent’s self‑improvement loop, Evolver manages versioned genomes, offering auditability and scientific rigor.

Architecture & value: The GEP protocol maintains a “genome” per Agent, supporting version rollback and cross‑Agent knowledge transfer, turning prompt tuning into a systematic engineering effort. Over 114 releases demonstrate high community activity. Recent allegations that Hermes Agent copied Evolver’s design further underscore Evolver’s forward‑looking approach.

Project URL: https://github.com/EvoMap/evolver

3. OpenAI Agents SDK: Official Multi‑Agent Orchestration Framework

The open‑source SDK (openai‑agents‑python) has amassed more than 22,000 stars, positioning it as the most trusted official Agent framework.

Core capabilities: It provides a complete toolchain for production‑grade Agents, including multi‑Agent orchestration, task delegation, tool calling, streaming, safety guards, context management, and deep integration with the latest OpenAI models. It focuses on primitive Agent‑to‑Agent collaboration rather than a monolithic approach.

Major updates this week (April 15): Introduction of a Harness mechanism for isolated testing and a SandboxAgent that mounts a containerized filesystem, enabling long‑running tasks and cross‑request environment reuse. A forthcoming “sub‑agent” concept will allow delegation to specialized agents.

Project URL: https://github.com/openai/openai-agents-python

4. Qwen3.6‑35B‑A3B: MoE‑Based Efficient Agent‑Programming Model

Alibaba’s Qwen3.6‑35B‑A3B, released on April 16, uses a sparse Mixture‑of‑Experts architecture with 350 billion parameters but activates only 30 billion per inference, achieving high efficiency.

Core capabilities: Despite its modest active parameter count, it outperforms the dense 270 billion‑parameter Qwen3.5‑27B on several programming benchmarks and rivals Google’s Gemma4‑31B. It matches Claude Sonnet 4.5 on most multimodal language tasks and exceeds it on specific metrics (e.g., RefCOCO 92.0, ODinW13 50.8).

Architecture & value: The MoE design reduces inference cost while preserving capability. It supports “thinking” and “non‑thinking” modes and integrates with OpenClaw, Claude Code, and Qwen Code. Weights are available on Hugging Face and ModelScope for local deployment or via Alibaba Cloud Bailei API.

Project URL: https://huggingface.co/Qwen/Qwen3.6-35B-A3B

5. HY‑World 2.0: Tencent’s Multimodal 3D World Model

Released on April 16, HY‑World 2.0 is a multimodal model that understands text, images, and video to generate, reconstruct, and simulate 3D worlds.

Core capabilities: Unlike Google’s Genie 3, HY‑World 2.0 outputs editable 3D assets (Mesh, 3DGS, point clouds) that can be directly imported into Unity or Unreal Engine, enabling rapid creation of game‑ready maps and level prototypes from natural language or visual inputs.

Architecture & value: It unifies generation and reconstruction within a single offline 3D world model paradigm, adds an interactive “character mode” with navigation and physics, and targets applications beyond gaming, such as digital twins, architectural planning, and cultural heritage preservation. The paper is on arXiv; code and weights are open.

Project URL: https://github.com/Tencent/HY-World

6. NVIDIA Ising: First Open‑Source Quantum AI Model Family

Announced on April 15, NVIDIA Ising tackles two core quantum‑computing challenges: error correction and processor calibration.

Core capabilities: The Calibration model is a 350 billion‑parameter vision‑language model that reduces calibration time from days to hours. The Decoding model, built on a 3D‑CNN framework, outperforms the open‑source pyMatching benchmark by up to 2.5× speed and 3× accuracy.

Technical value: By turning AI into the “operating system” for quantum computers, Ising addresses the “5‑year curse” of quantum scaling, aiming to lower error rates to below one in a trillion operations.

Project URL: https://github.com/NVIDIA/ising

7. Omi: Real‑Time Screen‑Aware Multimodal AI Assistant

Omi, from BasedHardware, captures screen content and spoken dialogue, transcribes in real time, generates summaries and to‑do items, and retains a continuous memory of user interactions.

Core capabilities: Modular design separates hardware abstraction, AI inference, and application layers, supporting desktop, mobile, and wearables (e.g., smart necklaces, AR glasses). Functions include AI chat, contextual memory, action suggestions, and live transcription.

Architecture & value: By continuously perceiving the user’s environment, Omi shifts AI from a query‑only tool to an ambient digital companion, offering a solid open‑source base for custom wearable AI experiences (171 contributors).

Project URL: https://github.com/BasedHardware/omi

8. Google AI Edge Gallery: Offline Mobile AI Model Experience Platform

Google’s AI Edge Gallery lets users run major open‑source LLMs, including Gemma 4, entirely offline on Android 12+ and iOS 17+ devices.

Core capabilities: Offline AI dialogue, visual question answering, audio transcription, and phone‑control functions powered by a 270 million‑parameter FunctionGemma model. Developed in Kotlin, the code is fully open for extension.

Architecture & value: Demonstrates efficient on‑device inference, preserving privacy by keeping all data local. It serves as a reference implementation for privacy‑sensitive industries such as finance, healthcare, and government.

Project URL: https://github.com/google-ai-edge/gallery

9. ElatoAI: ESP32‑Based Real‑Time Voice AI with Edge Computing

ElatoAI combines an ESP32 microcontroller with Cloudflare edge computing to deliver continuous global AI voice interaction lasting over ten minutes.

Core capabilities: ESP32 handles audio capture/playback; Cloudflare Durable Objects manage session state; OpenAI realtime API and Secure WebSockets transmit audio; Deno edge functions run the AI logic, ensuring low latency worldwide.

Architecture & value: By offloading heavy LLM inference to edge nodes, the solution overcomes the limited compute of low‑cost microcontrollers, offering a reference design for smart toys, wearables, and voice assistants.

Project URL: https://github.com/akdeb/ElatoAI

10. OpenClaw: Local‑First, Zero‑Code AI Agent Automation Platform

OpenClaw (aka “Little Lobster”) is an MIT‑licensed, fully open‑source AI agent framework emphasizing local privacy, zero‑code deployment, and comprehensive office automation.

Core capabilities: Three differentiators: (1) Local private execution—data never leaves the device; (2) Zero‑code deployment via a graphical UI, enabling non‑programmers to set up in ten minutes; (3) Full‑scene automation covering file management, data processing, browser automation, and messaging.

Technical ecosystem & value: The ecosystem includes lightweight (Pico/NanoClaw), high‑performance (MaxClaw), and industry‑specific variants (MedClaw, ClawWork). The latest v2.6.2 adds native video/music generation and a “dream” memory system, marking a shift from conversational tools to digital execution.

Project URL: https://github.com/openclaw/openclaw

Summary and Trend Observations

Trend 1 – AI Agents become production tools: OpenAI Agents SDK’s SandboxAgent and Harness, Evolver’s traceable prompt optimization, and Hermes Agent’s persistent memory illustrate the transition from experimental toys to deployable engineering systems.

Trend 2 – Edge AI and lightweight deployment dominate: Google AI Edge Gallery runs large models offline on phones; Omi brings real‑time perception to desktops and wearables; OpenClaw enables local, zero‑code deployment, reflecting a developer focus on secure, efficient on‑device AI.

Trend 3 – Accelerated open‑source contributions from tech giants: Tencent, Alibaba, NVIDIA, and OpenAI all released heavyweight projects within the same week, signaling that open‑source is now a strategic lever for advancing AI infrastructure and cross‑domain innovation such as quantum AI (NVIDIA Ising) and 3D generation (HY‑World 2.0).

Developers are advised to prioritize projects that align with their use cases: production‑grade Agent frameworks (OpenAI Agents SDK, Evolver), privacy‑focused local solutions (OpenClaw, Omi), or frontier research (NVIDIA Ising, HY‑World 2.0).