AI News

‹All dates

0126

DeepGEMM adds Ascend support and new kernels, trends on GitHub

DeepGEMM is DeepSeek's open-source, high-performance tensor-core kernel library that unifies many core LLM computation primitives — FP8/FP4/BF16 GEMMs, fused MoE with overlapped communication (Mega MoE), MQA scoring for the lightning indexer (including a sparse version) and HyperConnection (HC) — in a single CUDA codebase, with all kernels compiled at runtime via DeepJIT and no CUDA compilation at install time. According to the repository's changelog, DeepGEMM-Ascend became available on 2026.09.30 alongside new optimizations such as locality domain features, after earlier additions of the Sparse Indexer, Mega Gate, Mega mHC and MoE/Indexer optimizations. The repo also says its performance matches or exceeds expert-tuned libraries across various matrix shapes, and that it reached up to 1550 TFLOPS on H800 in April 2025.

GGitHub Trending（每日）·2h

0223

text-to-cad plugin gives coding agents local CAD generation and DFM checks

text-to-cad is an agent plugin that adds local CAD workflows: it generates 3D models as STEP, GLB, STL or 3MF from plain-language or image requests, runs design-for-manufacturing checks, produces dimensioned engineering drawings, and connects to 3D printing, sheet metal and CNC fabrication services. It installs into Claude Code, Claude Desktop, Codex, Cursor, Gemini and Grok Build — any agent supporting plugins or the skills framework — or as skills alone via npx skills add; it runs through uv and downloads the cadgen runtime on first start. Skills cover CAD, off-the-shelf STEP parts, engineering drawings, DXF, URDF/SRDF, SDF, SendCutSend pre-upload checks, DfAM printability checks, OrcaSlicer G-code and Bambu Lab printing.

GGitHub Trending（每日）·2h

039

TII introduces Falcon-ASR for Arabic and Emirati speech

Technology Innovation Institute (TII) has introduced Falcon-ASR, a 1.6B-parameter speech recognition model focused on Arabic, including the Emirati dialect. TII says it supports English, French, Spanish and Portuguese, offers word-level timestamps, and achieved a 20.92% average WER across six Arabic test sets in its evaluation.

Hugging Face·2h

0432

i-have-adhd skill makes coding agents lead with the answer

The GitHub project ayghri/i-have-adhd offers a skill/plugin that claims to stop coding assistants from burying the answer: lead with the next action, number multi-step tasks, restate state every turn, end with one concrete next step, and suppress preamble and closers, with 10 rules listed in SKILL.md. Its README's before/after example contrasts a "Great question! Let me think about this..." explanation with direct instructions such as "Edit src/auth.ts:42" and "Run npm test -- auth.spec.ts." The project is MIT-licensed; the author says it is loosely based on The Adult ADHD Tool Kit by J. Russell Ramsay and Anthony L. Rostain, adapted for how an LLM should respond rather than how a person organizes their day, and users can fork it and edit SKILL.md.

GGitHub Trending（每日）·2h

0513

Cognee ships v1.6.1 with Google Drive, Gmail connectors and graph visualization

Cognee is an open-source AI memory platform that gives agents persistent long-term memory across sessions, turning documents, code and conversations into a self-hosted, searchable knowledge graph that can run locally with small models and no cloud LLM key. Per its release notes, v1.6.1 (September 24) bundles Google Drive and Gmail OAuth connectors into the SDK, streams visualization of large graphs in chunks, and moves the GLiNER installer off the event loop; v1.6.0 supports keyless memory building and retrieval. The project reports BEAM conversational-memory scores of 0.79 at 100K tokens and 0.67 at 10M tokens, while noting the two settings use different conversations, ingestion models and retrieval procedures.

GGitHub Trending · Python·2h

0612

OpenMontage ships as open-source agentic video production system

OpenMontage has appeared on GitHub as a self-described first open-source, agentic video production system, bundling 12 production pipelines, 100+ tools and 700+ agent skill and production-knowledge files. It plugs into AI coding assistants such as Claude Code, Cursor, Copilot, Windsurf and Codex, letting an agent handle research, scripting, asset generation, editing and rendering using tools including Remotion, Blender, FFmpeg and Piper TTS. A local Backlot board turns the asset-generation contact sheet into an approval gate before the final render.

GGitHub Trending · Python·2h

0712

Rapid-MLX: open-source LLM server for Apple Silicon tops GitHub Trending

Rapid-MLX is an Apache-2.0 local LLM inference server built on MLX for Apple Silicon, exposing both OpenAI (/v1/chat/completions, /v1/responses) and Anthropic Messages (/v1/messages) APIs and focusing on reliable tool calling for coding agents. The project claims up to 4× faster decoding than Apple's mlx-lm on identical weights: a 1.50× median across 18 streaming tasks with Qwen3.5-9B 4-bit on an M4 Pro Mac mini, and up to 4.27× on a whole-file code edit (46.4 to 198.4 tok/s), plus 3.0× Ollama's aggregate decode at 8 concurrent streams on Qwen3.6-35B-A3B. All figures are the project's own measurements, and much of its comparison table is drawn from other projects' own documentation.

GGitHub Trending · Python·2h

0810

Stacklok's Mecatl: a Kubernetes-style cloud-native harness for agents

Stacklok, founded by Kubernetes creators Craig McLuckie and Joe Beda, pivoted from software supply chain security to Kubernetes-based agentic infrastructure. Its open source “cloud-native harness” Mecatl, started in June, separates the agent loop from tool execution, session management and memory; the companion open source project ToolHive runs and governs MCP servers on Kubernetes. The company says its AI Gateway is not yet open sourced but is on the roadmap, with an “enterprise spine” control plane as the commercial product.

LLatent Space·1h

0911

Anthropic open-sources knowledge-work plugins for Claude Cowork

Anthropic has open-sourced a set of role-based plugins on GitHub, with an initial 11 covering sales, customer support, product management, marketing, legal, finance, data, enterprise search and bio-research. Built for Claude Cowork and also compatible with Claude Code, each plugin bundles skills, connectors, slash commands and sub-agents as plain markdown and JSON files with no code, infrastructure or build steps. Teams can swap connectors, add company terminology and processes, or write new plugins following the same structure.

GGitHub Trending · Python·2h

1011

thedotmack/claude-mem

Persistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More https://claude-mem.ai 🇨🇳 中文 • 🇹🇼 繁體中文 • 🇯🇵 日本語 • 🇵🇹 Português • 🇧🇷 Português • 🇰🇷 한국어 • 🇪🇸 Español • 🇩🇪 Deutsch • 🇫🇷 Français • 🇮🇱 עברית • 🇸🇦 العربية • 🇷🇺 Русский • 🇵🇱 Polski • 🇨🇿 Čeština • 🇳🇱 Nederlands • 🇹🇷 Türkçe • 🇺🇦 Українська • 🇻🇳 Tiếng Việt • 🇵🇭 Tagalog • 🇮🇩 Indonesia • 🇹🇭 ไทย • 🇮🇳

GGitHub Trending（每日）·2h

1112

Agent-Reach: an open-source CLI that gives AI agents internet access

Agent Reach is an open-source "capability layer" that lets command-line-capable agents such as Claude Code, OpenClaw and Cursor read web pages, YouTube captions, GitHub and RSS through a single CLI, and — after login setup — search and read Twitter/X, Reddit, Bilibili and Xiaohongshu. The project claims all upstream tools are open source and that no API fees apply, with cookies and tokens stored only locally in ~/.agent-reach/config.yaml (mode 600), plus an agent-reach doctor command to check each channel's health. Its README also warns that cookie-based platforms carry detection and ban risk and recommends using a dedicated throwaway account rather than a primary one.

GGitHub Trending · Python·2h

1211

Agency Agents ships a native desktop app for its agent roster

Agency Agents is a growing collection of AI agent personas that started from a Reddit thread and months of iteration, spanning engineering categories such as frontend, backend, DevOps, security and data, with each agent given its own identity, workflows and deliverables. The project now offers a native macOS, Linux and Windows app that browses the full roster and installs agents into Claude Code, Cursor, Codex, Gemini CLI, OpenCode, Qwen and Osaurus in one click, alongside script-based installs and auto-updates. The repo describes the agents as "production-ready," a claim that comes from the project itself rather than independent testing.

GGitHub Trending（每日）·2h

1311

Omnigent open-source meta-harness orchestrates Claude Code, Codex and other agents

Omnigent is an open-source AI agent framework and "meta-harness" that, per its project documentation, mixes Claude Code, Codex, Cursor, OpenCode, Hermes, Pi and user-defined YAML agents in the same session, with sessions that follow you between terminal, browser, phone and a macOS desktop app. The README also claims support for first-party API keys or Claude/ChatGPT subscriptions, cloud sandboxes such as Modal, Daytona, E2B and Kubernetes, and policies that pause for approval, cap spend or limit tool access, plus session sharing for teammates. The repository is currently listed on GitHub Trending for Python.

GGitHub Trending · Python·2h

1410

Impeccable: open-source design skill for AI coding agents

Impeccable is an open-source design-language skill for AI coding agents by pbakaus, described as building on Anthropic's frontend-design skill for Claude to counter the same SaaS-template look models tend to produce. It ships one skill, 24 commands (init, audit, critique, polish, live and more) and 60 deterministic detector rules that the CLI and browser extension run with no LLM or API key. Install is via npx impeccable install, with support for Claude Code, Cursor, Codex CLI, Gemini CLI, GitHub Copilot, Grok Build, Hermes Agent and others; some harnesses require preview channels or manual hook trust.

GGitHub Trending（每日）·2h

1510

Open-source AI engineering curriculum trends on GitHub

The open-source project "AI Engineering from Scratch" has appeared on GitHub's Python trending list. It describes 20 phases, 523 lessons and roughly 342 hours of material in Python, TypeScript, Rust and Julia, released free under an MIT license, with each lesson having learners implement algorithms such as backpropagation, tokenizers, attention and the agent loop from raw math before running the same thing through production libraries like PyTorch. It also offers focused Model Context Protocol (MCP) and Agent Skills routes plus preparation for Claude certification tracks, and can be installed as learning skills for coding agents including Codex and Claude Code.

GGitHub Trending · Python·2h

1610

HexStrike AI v6.0 ships 150+ security tools for autonomous MCP agents

HexStrike AI published v6.0 on GitHub, an MCP server that lets agents such as Claude, GPT and Copilot autonomously run 150-plus cybersecurity tools for penetration testing, vulnerability discovery and bug-bounty automation. The project describes a multi-agent architecture with 12+ autonomous agents, decision engines for tool selection and attack-chain discovery, and real-time visualization dashboards. It is developed by OTT Cybersecurity LLC and documents setup for MCP clients including Claude Desktop, Cursor and VS Code Copilot.

GGitHub Trending · Python·2h

176

llm-openai-decisions 0.1a0 plugin wraps OpenAI's Jev-style Decisions API

Simon Willison released llm-openai-decisions 0.1a0, an LLM plugin for OpenAI's new Decisions API announced at last week's DevDay; he had GPT-6 Astra read the new API docs and modeled the plugin on his existing llm-typesafe plugin for Jev. The new decision model, gpt-6-luna, accepts image input as well as text. According to the post, both models charge for input only and not output — OpenAI at 10 cents per million input tokens versus Jev's 4.2 cents — and both support yes/no, choice and score question types.

SSimon Willison's Weblog·16h

188

Analogue Pocket SNES Developer Criticizes AI-Generated Code Submission

The developer responsible for porting the SNES FPGA core to Analogue Pocket criticized a contributor for submitting an experimental MSU-1 streaming feature as nearly 50,000 lines of large-language-model-generated code. The developer said the unrelated code would require extensive manual review and asked contributors not to submit work they do not fully understand; RPCS3 developers have expressed a similar position.

ccnBeta·9h

196

Ollama v0.40.0 defaults to MLX on Apple Silicon

Ollama v0.40.0 makes MLX the default runtime for supported model architectures on Apple Silicon. The release also adds MLX support for decision models and the EmbeddingGemma 2 embedding model, with additional model families listed in the release notes.

OOllama Releases·21h

2017

Reflection AI debuts Beam, a 501B open-weight model aimed at Chinese rivals

Reflection AI, the Nvidia-backed US startup founded by ex-DeepMind researchers, unveiled Beam — its first open-weight model — a text-only mixture-of-experts system with 501B total and 23B active parameters, pretrained on 23.8T tokens with a 1M-token context window. The company says Beam matches Z.ai's GLM-5.2 on reasoning, coding and agentic benchmarks while using 3–4x less inference compute, and approaches Qwen 3.8-Max; the claims have not been independently verified. Full weights are promised this month under an Apache 2.0 license.

IIT之家 AI·2d