New
openadaptai/openadapt openadapt-desktop-v0.22.0 — Desktop Release
Introduces OpenAdapt's latest desktop release and the execution structure that records computer work and replays it.
CURATED OPEN SOURCE INTELLIGENCE
Latest news
New
realsenseai/librealsense v2.59.1 — cross-platform depth camera SDK
An introduction to librealsense, which handles depth and color streaming and calibration data for RealSense cameras, and to the latest release on record, v2.59.1.
New
paddlepaddle/paddle-lite — an inference engine for mobile and edge
paddlepaddle/paddle-lite is a lightweight inference framework that runs and optimizes PaddlePaddle models on mobile, embedded and edge devices.
New
standardagents/dmux — Separating Agent Work with tmux and Git Worktrees
standardagents/dmux is a tool that isolates and manages the work of multiple AI coding agents in parallel using tmux and Git worktrees.
New
stakpak/agent v0.3.88 — an always-on DevOps agent
Introduces the always-on DevOps capabilities of stakpak/agent and its latest public release v0.3.88, along with the scope of what can be verified.
New
opennmt/ctranslate2 v4.8.3 — Transformer inference for CPU and GPU
An introduction to the latest release of opennmt/ctranslate2, v4.8.3, and the features of the library that runs Transformer models on CPU and GPU.
New
solvespace/solvespace v3.2 — Parametric 2D/3D CAD
solvespace/solvespace is the source code repository for SolveSpace, a parametric 2D/3D CAD tool that supports constraint-based design and modeling.
New
mcpware/intentprobe architecture-v0.1.2 — a study of a probe-sharing architecture
mcpware/intentprobe is a Python repository that studies an architecture where multiple task-specific probes share the hidden states of a fixed local language model.
Release
ggml-org/llama.cpp b11752 — MoE Router Selection Stabilization Test
ggml-org/llama.cpp b11752 addresses the problem that expert selection can vary due to per-backend numerical noise, and includes MoE router selection tests.
New
opendatalab/pdf-extract-kit 1.0.0 — A Collection of PDF Extraction Models
Introduces the 1.0.0 release of opendatalab/pdf-extract-kit and the collection of models that extract layout, formulas, text, and tables from PDFs.
New
openai/plugins — A Collection of Codex Plugin Examples
openai/plugins is a repository gathering Codex plugin examples and file layouts across several kinds of work.
New
temporalio/sdk-python 1.34.0 — Temporal SDK for Python
A look at the latest release of temporalio/sdk-python and the role of the SDK for writing Temporal workflows and activities in Python.
New
spacering-net/codeg v0.38.0 — A workspace for multiple coding agents
spacering-net/codeg is a workspace that handles the conversations, file changes, and sessions of multiple AI coding agents in one place.
New
qualcomm/geniex — On-Device AI Inference Runtime for Snapdragon
GenieX in qualcomm/geniex is a generative AI inference runtime that runs LLMs and VLMs locally on Snapdragon devices.
New
justrach/nanoterm v0.1.21 — a block-based terminal with agents attached
justrach/nanoterm is a Zig project whose block-based terminal input field handles both shell commands and ACP-based coding agent requests.
New
b0yko/tern — Local media search for Apple Silicon
b0yko/tern is a local tool for Apple Silicon that searches podcast and video libraries by speech, on-screen text, and scenes, then lets you edit and export the results.
New
smtg-ai/claude-squad v1.0.20 — Terminal multi-agent task management
Introduces the public release status of smtg-ai/claude-squad v1.0.20 and how to manage multiple local coding agents from the terminal.
New
openai/openai-agents-js — JavaScript·TypeScript agent framework
openai/openai-agents-js is a framework for building multi-agent workflows in JavaScript and TypeScript.
New
pytorch/fbgemm — Low-precision compute kernels for deep learning
A look at the deep learning operations covered by pytorch/fbgemm's three packages and at what cannot be confirmed from the available materials.
New
cv-superding/hypercontext-mode v0.1.4 — Coding Agent Context Unification
An introduction to the structure in which cv-superding/hypercontext-mode handles tool-output sandboxing, conversation compression, and persistent memory within a single execution environment, plus the search measurements in its README.
No matching stories.