New AI GitHub Repos — Trending Open-Source Drops | Blog Network

Trending open-source AI projects fresh off GitHub — new repos, libraries and frameworks, with what each one does and why it's picking up stars.

85 releases tracked

• Ray 2.58.0 — KV-cache-aware routing lands for LLM servingRay Project · 2026-08-23 · major

  Ray's AI compute engine finishes cache-aware LLM routing and adds TPU training plus a gVisor sandbox for task code.

• OpenViking v0.4.16 — agents can run Skills hosted on another serverVolcengine · 2026-08-21 · major

  VikingBot can now pull Skills from a different OpenViking server, so one team can host a set of tools and let many agents share it.

• Munder Difflin v0.4.5 — cost reporting and Apple Silicon memory get fixedMunder Difflin · 2026-08-22 · notable

  Munder Difflin puts your agent CLIs on a pixel-art office floor and lets them message each other.

• OpenViking v0.4.14 — agents can now write and edit inside the memory storeVolcengine · 2026-08-17 · major

  Volcengine's open-source context database turns its viking:// namespace into a place agents can write to, not just read from.

• Agent Lightning v1.0 — Microsoft's RL trainer plugs into real agent harnessesMicrosoft · 2026-08-17 · major

  Agent Lightning v1.0 lets you reinforcement-learn a model without rewriting the agent that will run it in production.

• DesktopFly — a Mac desktop fly wired from the real FlyWire connectomeDenis Sergeevitch · 2026-08-18 · notable

  A desktop pet whose behavior comes from a real fruit fly wiring diagram instead of hand-written animation.

• Sentence Transformers v6.0 — ColBERT-style retrieval joins the libraryHugging Face · 2026-08-18 · major

  Late interaction retrieval is now a built-in Sentence Transformers model type, not a separate framework.

• Mojo goes open source — Modular ships the compiler under Apache 2.0Modular · 2026-08-18 · major

  Modular published the Mojo compiler and toolchain under Apache 2.0, four years after the language first appeared.

• turbovec 1.0 — Rust vector index fits 10M embeddings in 4 GBRyanCodrai · 2026-08-18 · major

  A Rust vector index that fits 10 million embeddings in 4 GB and takes new vectors without a training pass.

• h3.c — antirez ships a C engine that runs MiniMax H3 on Apple Siliconantirez · 2026-08-09 · major

  A single C program that turns a prompt into video and audio on a Mac, with no Python stack underneath.

• LoopX 0.4.1 — durable state kernel for long-running coding agentshuangruiteng · 2026-08-04 · notable

  LoopX is a small durable-state kernel that stops long agent runs from forgetting their own plan.

• TencentDB Agent Memory v2.0 — team-level memory hub for AI coding agentsTencent Cloud · 2026-08-03 · major

  A shared memory layer that lets a team's coding agents keep learning from each other's work.

• DeepSeek V4-Flash on one MI300X — 168 tok/s decode, no quantizationryanzhou · 2026-08-04 · notable

  Docker Compose stack that runs the full 304B DeepSeek V4-Flash unquantized on a single AMD MI300X, hitting 168 tok/s decode.

• Swiftlet — 80B Qwen runs in 4.3 GB on a Mac, 35B on an iPhoneleonickson1 · 2026-08-03 · notable

  Swift and Metal runtime that runs Qwen 35B on an iPhone and 80B on a Mac by streaming MoE experts from local storage.

• WASTE — run 2.78T Kimi K3 on a 64GB laptop by streaming from NVMeSQLite AI · 2026-07-30 · major

  SQLite AI's WASTE streams Kimi K3's experts from an NVMe SSD so the full 2.78-trillion-parameter model runs on a 64 GB laptop.

• QM — Y Combinator open-sources the multi-agent harness it runs internallyY Combinator · 2026-07-31 · major

  Y Combinator's in-house multi-agent harness, now open-sourced under MIT — one core, four backends, Slack + web from day one.

• OpenWork — open-source Claude Cowork alternative hits 18.5K starsDifferent AI · 2026-07-30 · major

  OpenWork ships an MIT-licensed desktop workspace that shares AI skills across Claude Code, Cursor, and Codex.

• ECC 2.1.0 — Plan Canvas turns terminal agent plans into a clickable browser reviewECC · 2026-07-27 · major

  A 234k-star agent harness adds a browser plan-review canvas, a Kimi Code target, and self-hosted GPUs.

• esp32-ai — 28.9M-parameter LLM runs entirely on an $8 ESP32-S3slvDev · 2026-07-25 · major

  A hobbyist ports Gemma's Per-Layer Embeddings trick to a microcontroller and gets a 28.9M-param LLM running on $8 of silicon.

• Open Interpreter 0.0.26 — Rust coding agent tuned for open modelsOpen Interpreter · 2026-07-16 · major

  The 66k-star coding agent has been rewritten in Rust to make cheap open models act more like Codex.

• Vibe-Trading 0.1.11 — HKUDS ships Indian equity and a 460-factor libraryHKUDS · 2026-07-10 · notable

  AI trading agent gains India equity, a fundamentals factor layer, and research delivery over 16 messaging apps.

• Colibri — pure-C engine runs GLM-5.2 744B on a 25 GB laptopJustVugg · 2026-07-09 · major

  Colibri streams GLM-5.2's experts from disk so a 744B frontier MoE runs on a 25 GB laptop.

• Hugging Face Transformers 5.13.0 — nine new architectures and unified HfExporterHugging Face · 2026-07-03 · major

  The reference AI library gains nine architectures in one drop plus one export command for every runtime.

• Chrome DevTools MCP 1.5.0 — heap snapshot comparison lands for coding agentsChrome DevTools · 2026-07-03 · notable

  Chrome DevTools MCP 1.5.0 gives coding agents a proper heap-snapshot diff workflow, plus tighter default file permissions.

• Cline 4.0 — SDK rewrite rolled back to 3.89 two days after launchCline · 2026-06-26 · major

  Cline shipped a major SDK rewrite as 4.0.0, then rolled it back two days later as 4.0.1 after launch-day regressions.

• cognee v1.2.2 — truth-subspace reranking for the open-source agent memory platformtopoteretes · 2026-06-26 · notable

  Open-source agent memory gets a reranker that prefers facts the graph has already learned.

• DSpark + DeepSpec — DeepSeek opens its speculative decoding stackDeepSeek · 2026-06-26 · major

  DeepSeek shipped a free codebase for training speculative-decoding drafters, plus DSpark drafters bolted onto V4-Pro and V4-Flash.

• OpenKnowledge v0.18.0 — local-first AI markdown wiki with codebase generatorInkeep · 2026-06-25 · notable

  Local-first AI markdown wiki — now scaffolds a codebase wiki an agent fills in.

• OpenMontage — AGPL agentic video studio for AI coding assistantscalesthio · 2026-06-22 · major

  An agentic video studio that turns any AI coding assistant into a producer, editor, and renderer.

• Anthropic-Cybersecurity-Skills v1.3.0 — 817 security skills across 6 frameworksMahipal Jangra · 2026-06-22 · notable

  55 new agent skills for AI security, supply chain, and hardware/firmware land in the most-starred open cybersecurity skills pack.

• Hermes Agent v0.17.0 — iMessage, WhatsApp, and async subagents from NousNous Research · 2026-06-19 · major

  Nous Research's self-improving agent gains iMessage and WhatsApp adapters, Raft-network integration, and background subagents in 1,475 commits.

• cuTile Rust v0.2.0 — NVIDIA Labs ships NVFP4 GPU kernels in safe RustNVIDIA Labs · 2026-06-16 · major

  cuTile Rust is NVIDIA Labs' safe tile-based GPU kernel DSL for Rust, now with NVFP4 packing and a new performance paper.

• agent-skills 0.6.2 — Addy Osmani adds /build and /webperf for AI coding agentsAddy Osmani · 2026-06-11 · notable

  Addy Osmani's open-source agent-skills bundle hits v0.6.2 with new slash commands and hardened security skills.

• OpenCV 5.0 Ships With Built-In LLM and VLM Inference — Rewritten Graph-Based DNN Engine Pushes ONNX Coverage From ~22% to Over 80%, Native Tokenizer and KV-Cache Run Qwen 2.5, Gemma 3, PaliGemma, and GPT-Family Models Out of the BoxOpenCV · 2026-06-08 · major

  The most-installed computer-vision library lands version 5.0 with a graph-based DNN engine and built-in LLM / VLM inference, timed for CVPR 2026 in Denver.

• Nous Research Ships Hermes Agent v0.16.0 'The Surface Release' — Native Electron Desktop App for macOS, Linux, and Windows With In-App Self-Updates, Drag-and-Drop, and a Full Web Admin PanelNous Research · 2026-06-05 · major

  Nous Research's self-improving agent gets a native desktop client and a no-config web admin panel; the project crossed 183K stars on the same day.

• Headroom v0.26.0 — context compressor adds Copilot BYOK and Bedrock proxychopratejas · 2026-06-16 · notable

  Drop-in token compression layer for AI agents that claims 60–95% fewer tokens with single-digit accuracy hit.

• PewDiePie Open-Sources Odysseus — Self-Hosted AI Workspace Hits 20K Stars in 24 Hours With Chat, Agents, MCP, Deep Research, and Email TriagePewDiePie · 2026-05-31 · major

  Felix Kjellberg drops a 20K-star self-hosted AI workspace that runs entirely on your hardware — chat, agents, deep research, and email.

• LlamaIndex Rewrites LiteParse in Rust for v2.0 — 5–100× Faster Open-Source PDF and Office Parser, 457-Page Doc in 0.777 Seconds, Native + Python + Node + WASMLlamaIndex · 2026-05-27 · notable

  LlamaIndex rewrote its open-source PDF parser in Rust — same CLI, 5–100× faster, runs on Linux, macOS, Windows, Python, Node, and in the browser via WASM.

• Nous Research Ships Hermes Agent v0.15.0 'The Velocity Release' — Multi-Agent Kanban Platform, 76% Smaller Core Loop, 4,500× Faster Session SearchNous Research · 2026-05-28 · major

  Velocity release rewires Hermes Agent into a multi-agent Kanban swarm and shrinks the core loop by 76%.

• DeepSeek-Reasonix — Terminal Coding Agent Built Around DeepSeek's Prefix Cache Reports a 99.82% Cache Hit Rate, Cutting a 435M-Token Day From ~$61 to ~$12esengine · 2026-05-24 · notable

  A terminal coding agent that bets on DeepSeek's prefix cache to keep long sessions cheap.

• Remove-AI-Watermarks Hits the HN Front Page — MIT-Licensed CLI Strips Visible Gemini Watermarks Plus Invisible SynthID, C2PA, and EXIF Provenancewiltodelta · 2026-05-18 · major

  An MIT-licensed library lifts every major AI provenance marker — and lands on the HN front page the same week OpenAI adopts SynthID.

• CodeGraph 0.7.10 — Local SQLite + Tree-Sitter Knowledge Graph Cuts Claude Code, Codex, and Cursor Exploration Tool Calls by 94%Colby McHenry · 2026-05-19 · notable

  Pre-index a repo into a local knowledge graph and let your coding agent query it instead of grepping every file.

• Forge — Open-Source Guardrails Lift a Self-Hosted 8B Local Model From 53% to 99% on Multi-Step Agentic Tool-Calling WorkflowsAntoine Zambelli · 2026-05-19 · notable

  Drop-in reliability layer that pushes 8B local models toward frontier-API accuracy on tool-calling agents.

• Semble — Code Search for AI Agents Uses ~98% Fewer Tokens Than grep+read, Hits the Hacker News Front PageMinish Lab · 2026-05-17 · major

  A CPU-only code-search MCP server that cuts the tokens an agent burns hunting through a repo.

• InsForge v2.1 — Open-Source Backend Platform for Coding Agents Adds Compute Services, Realtime Presence, StripeInsForge · 2026-05-06 · notable

  All-in-one open-source backend (DB, auth, storage, compute, AI gateway) that coding agents drive end-to-end via MCP or CLI.

• ds4 — Antirez Ships C/Metal Inference Engine for DeepSeek V4 Flash on Apple SiliconAntirez · 2026-05-06 · major

  Antirez's first AI repo: a DeepSeek V4 Flash-only inference engine in C, built for Apple Silicon with a compressed disk-backed KV cache.

• vLLM v0.20.0 — DeepSeek V4, FlashAttention 4 Default, TurboQuant 2-bit KV CachevLLM Project · 2026-04-27 · major

  vLLM's biggest release of 2026: DeepSeek V4, FA4 as default, and 4× KV cache via 2-bit compression.

• SGLang v0.5.10 — Native Apple Silicon Backend, Elastic MoE Fault Tolerance, sglang-kernel 0.4.1SGLang Project · 2026-04-09 · major

  SGLang adds Apple Silicon inference, resilient MoE failover, and a 1000× RDMA reduction for large-scale clusters.

• llama.cpp Build b8738 — Vendor-Agnostic Tensor Parallelism, 1-Bit Quantization, AMD CDNA4ggml-org · 2026-04-09 · major

  llama.cpp gains true multi-GPU tensor parallelism across CUDA, ROCm, and Metal — no vendor lock-in.

• RAGFlow v0.25.0 — Seven Pipeline Templates, Agent Publishing, User Memory, Mobile-ReadyInfiniFlow · 2026-04-21 · major

  RAGFlow v0.25 ships production-ready agent publishing, persistent user memory, and ingestion pipelines for seven document types.

• PageIndex File System — Vectorless Reasoning RAG Scales to Millions of DocumentsVectifyAI · 2026-05-03 · major

  PageIndex File System adds a query-time virtual tree that makes vectorless reasoning RAG work across millions of documents.

• Unsloth Studio v0.1.37 — New UI Redesign, Qwen3.6 Support, Preserve Thinking ModeUnsloth AI · 2026-04-23 · major

  Unsloth Studio's April redesign adds Qwen3.6 support, Preserve Thinking toggle, and developer role for agentic coding tools.

• Google ADK Python v1.32.0 — Native OpenTelemetry Metrics, Anthropic Thinking Blocks, Visual Graph CanvasGoogle · 2026-05-01 · major

  Google's ADK now ships native OpenTelemetry metrics and visual agent graphs in its 2026 multi-agent framework.

• Llama Stack v0.8.0 — Native Anthropic Messages API, Gemini Interactions, 91% OpenAI CompatibilityMeta · 2026-05-01 · notable

  Llama Stack now speaks Anthropic's Messages API and Google's Interactions API from a single unified GenAI stack.

• Goose v1.2 — Automatic MCP Server Discovery, ACP Protocol, 44k GitHub StarsAAIF / Block · 2026-04-10 · major

  Goose adds MCP auto-discovery and ACP agent-to-agent protocol — the reference implementation for the evolving MCP ecosystem.

• ml-sharp-web — Apple's SHARP 3D Gaussian Splat Model Runs Entirely In Your Browserbring-shrubbery · 2026-05-03 · notable

  Apple's one-shot photo-to-3D Gaussian splat model, ported to ONNX and shipped as a 100% in-browser web app.

• DeepClaude — Run Claude Code's Agent Loop on DeepSeek V4 Pro for 17x Cheaper Tokensaattaran · 2026-05-03 · major

  Point Claude Code at DeepSeek V4 Pro and pay $0.87 per million output tokens instead of $15 — same agent loop, same tools, different brain.

• DeepSeek-TUI v0.8.7 — Rust Terminal Coding Agent for DeepSeek V4 With 1M-Token ContextHmbown · 2026-05-03 · notable

  A native, single-binary terminal agent built specifically around DeepSeek V4's 1M-token context and prefix cache.

• Ruflo v3.6.10 — 32 Plugins, Agent Federation, IoT Cognitum for Multi-Agent Claude Orchestrationruvnet · 2026-04-30 · major

  Open-source orchestrator for swarms of Claude Code agents picks up federation and an IoT bridge in a 32-plugin v3.6 release.

• jcode — Rust Coding Agent Harness Hits #4 on GitHub Trending With Swarm Mode and Self-Modification1jehuang · 2026-04-30 · major

  A solo-developer Rust coding-agent harness tops GitHub trending with claims of 14 ms boot and built-in swarm mode.

• OpenWarp — Community Fork of Warp Opens Up Bring-Your-Own AI Providerzerx-lab · 2026-04-29 · major

  Plug any OpenAI-compatible model into Warp's terminal — credentials stay local, no cloud relay.

• Mike — Open-Source, Self-Hostable Alternative to Harvey and Legora for Law FirmsWill Chen · 2026-04-29 · major

  Self-hostable open-source legal AI: chat with documents, cite verbatim, run multi-step workflows, draft contracts.

• Dirac — Open-Source Coding Agent Tops Terminal-Bench-2 on Gemini 3 Flash PreviewDirac Delta Labs · 2026-04-28 · major

  A solo-built Cline fork takes #1 on Terminal-Bench-2 with a Flash-tier model — the harness, not the model, was the bottleneck.

• Beads — Git-Backed Dependency Graph Issue Tracker Built for AI Coding AgentsGastown · 2026-04-24 · major

  A dependency-aware issue tracker built for AI coding agents — versioned graph storage with hash IDs that survive multi-agent merges.

• GoModel — Lightweight AI Gateway in Go, LiteLLM AlternativeENTERPILOT · 2026-04-25 · notable

  A Go-based AI gateway that exposes a single OpenAI-compatible endpoint across 10+ LLM providers, with caching, cost tracking, and an admin dashboard.

• mattpocock/skills 1.1.0 — engineering skill pack for AI coding agentsMatt Pocock · 2026-07-08 · major

  Matt Pocock's composable skill pack teaches AI coding agents how to plan, TDD, and review like a senior engineer.

• GitNexus — Code Intelligence Graph with MCP for AI EditorsAkon Labs · 2026-04-24 · major

  A code knowledge graph that gives AI editors structural awareness of your entire codebase via 16 MCP tools.

• OpenChronicle — Open-Source Local-First Screen Memory for LLM AgentsEinsia · 2026-04-25 · major

  Free, local-first screen memory for AI agents — built in 48 hours as a direct response to OpenAI paywalling Chronicle at $100/month.

• Browser Harness: Self-Healing LLM Browser Control Directly on CDPbrowser-use · 2026-04-24 · notable

  A minimal Python shim (~592 lines) that connects an LLM directly to Chrome via CDP — and lets the agent edit its own harness when it gets stuck.

• free-claude-code — Route Claude Code CLI to Free LLM ProvidersAlishahryar1 · 2026-03-08 · major

  A Python proxy that lets you use the Claude Code CLI and VSCode extension with free LLM backends instead of Anthropic's API.

• vercel-labs/skills — Universal CLI for Discovering and Installing Agent Skills Across 45+ Coding ToolsVercel Labs · 2026-04-17 · major

  One CLI to install, share, and discover reusable instruction sets for any AI coding agent.

• RAG-Anything — All-in-One Multi-Modal RAG Framework for Text, Images, Tables, and EquationsHKUDS · 2026-03-24 · major

  RAG for real-world documents — handles images, tables, equations, and charts alongside text in a single pipeline.

• Claude Context — Semantic Codebase Search MCP with ~40% Token Savings for Coding AgentsZilliz · 2025-06-06 · notable

  One MCP install gives your coding agent semantic search over the entire codebase — no per-conversation file loading needed.

• CrabTrap — LLM-as-Judge HTTP Proxy to Secure AI Agents in ProductionBrex · 2026-04-17 · notable

  CrabTrap sits between your AI agent and the internet, vetting every outbound request against natural-language security policies before they leave.

• LuceBox Hub — Hand-Tuned LLM Inference Reaching 207 tok/s on an RTX 3090Luce-Org · 2026-04-20 · notable

  LuceBox rewrites LLM inference from scratch for one GPU at a time, achieving 207 tok/s on Qwen3.5-27B on a consumer RTX 3090.

• RuView — WiFi DensePose: 17-Point Body Pose Through Walls, No Camera Requiredruvnet · 2026-04-16 · notable

  Real-time body pose and vital sign monitoring through walls using commodity WiFi — no camera, no wearable, $9 hardware.

• OpenCode — open-source terminal AI coding agent crosses 146k GitHub starsAnomaly · 2026-04-18 · major

  The largest open-source AI coding agent on GitHub — terminal-first, provider-agnostic, and a viable swap for Claude Code if you want fully open infra.

• Mozilla Thunderbolt — Open-Source Self-Hostable Enterprise AI ClientMZLA Technologies · 2026-04-16 · notable

  Mozilla's open-source enterprise AI client that runs on your own infrastructure and connects to any model provider.

• Hermes Agent v0.10.0 — Tool Gateway for Self-Growing Multi-Platform AI AgentsNous Research · 2026-04-16 · major

  Self-growing agent framework that learns from interactions and runs on every platform from CLI to WhatsApp.

• Lyra 2.0 — NVIDIA's Explorable Generative 3D World FrameworkNVIDIA · 2026-04-15 · major

  NVIDIA's open-source framework for generating interactive 3D worlds — walk through, explore, export to Isaac Sim.

• ClawGUI — Unified Framework for Training, Evaluating, and Deploying GUI AgentsZJU-REAL · 2026-04-13 · notable

  Train, benchmark, and deploy mobile GUI agents — all from one auditable open-source repo.

• Open Agents — Open-Source Cloud Coding Agent Platform by Vercel LabsVercel Labs · 2026-04-14 · major

  Vercel Labs' forkable cloud coding agent: chat drives the task, an isolated sandbox runs it, and Workflow SDK keeps it durable.

• Shep — parallel AI coding agent orchestrator across git worktreesshep-ai · 2026-02-02 · notable

  Run ten feature branches in parallel — each with its own AI agent, git worktree, and PR — from a single CLI command.

• AutoResearch — autonomous ML experiment runnerAndrej Karpathy · 2026-03-06 · major

  Give an AI agent a small LLM training setup, go to sleep, and wake up to 100 completed experiments — Karpathy's viral 630-line repo.

• OpenWorldLib — unified world-model codebasePeking University, Kling/Kuaishou, HKUST · 2026-04-06 · notable

  One codebase, one definition, six modules — an attempt to pull scattered world-model research into a single inference framework.

← All releases