New
ovg-project/kvcached v0.1.6 — Elastic KV cache for shared GPUs
ovg-project/kvcached v0.1.6 is a KV cache daemon that allocates GPU memory elastically to support LLM serving and training in shared environments.
CURATED OPEN SOURCE INTELLIGENCE
Latest news
Release
fazmin/cve-intel v0.3.0 — Local CVE Triage Tool
Introduces fazmin/cve-intel's first placeholder release, v0.3.0, and the local CVE triage capability described in the repository description.
New
jpb18/polystrata-hip — An AMD HIP fork of the Strata-family inference engine
jpb18/polystrata-hip is a derived repository that ports the Strata-family inference engine strata-poly to AMD HIP so that mixture-of-experts models can run on AMD GPUs.
New
bytedance/deer-flow v2.1.0 — a long-running super-agent harness
bytedance/deer-flow is a long-running super-agent harness designed to handle research, coding, and content work anywhere from minutes to hours.
New
deeplethe/solid-glass — a glass-texture design system for agent screens
deeplethe/solid-glass is a glass-texture design system for AI applications that provides components in the shadcn/ui registry format.
New
feder-cr/git-ai v0.1.0 — a version control server for coding agents
feder-cr/git-ai is a version control repository built to gather the work of several coding agents on one server, and it has released v0.1.0.
New
onehumanbeing/esp32-iphone-agent — Control a real iPhone without a computer
A hardware project where a vision model controls a real device using an ESP32-S3 board and an iPhone.
New
aoisop/mawoagent — a local agent runtime that goes through evidence and tests
aoisop/mawoagent is a Python-based local agent runtime that turns feedback into a skill only when evidence and tests pass.
New
angelkurten/keel v0.26.0 — Agent Context Layer for Vertical SaaS
angelkurten/keel is a Python project that builds a per-customer bitemporal context graph from domain events and exposes it to agents through five MCP tools.
New
mw00/project-maya v1.1.15 — a GLM-5.3-Flash engine for older GPUs
mw00/project-maya is a C++ engine that splits experts across GPU, RAM and NVMe SSD to run GLM-5.3-Flash on a personal GPU.
New
kajisho5/ffmpeg-skill v2.10.4 — local video editing for coding agents
kajisho5/ffmpeg-skill gives coding agents FFmpeg-based video editing tools and an MCP server, using a design that derives tool input specs from the implementation.
New
ipads-sai/mobiinfer-android — On-device MoE for Snapdragon NPUs
ipads-sai/mobiinfer-android is a C++ Android repository that runs 35B-A3B-class MoE models on device on the Qualcomm Hexagon NPU.
New
touri-ai/touri — Local AI Agent Desktop App
touri-ai/touri is a local-first desktop app that runs existing AI agent tools in an isolated environment and supports browser and computer use.
New
vshulcz/deja-vu — searching coding agent session logs
vshulcz/deja-vu is a Go tool that indexes on-disk session logs from coding agents and provides local search plus MCP.
New
vast-ai-research/edenlive — a real-time generative renderer for interactive worlds
vast-ai-research/edenlive separates world state from appearance rendering to generate frames from reference video and sentence instructions.
Release
academa-labs/manimgx v0.1.1 — Fixes a GPU validation bug for meshes with materials
academa-labs/manimgx v0.1.1 fixes a bug where GPU validation failed for meshes with materials in certain scene configurations.
New
rsi-labs/rsi-chip — An agent-centric Rust-native EDA toolkit
rsi-labs/rsi-chip is a suite of Rust-native EDA tools aimed at agent-led design workflows.
New
zfinix/eleven v0.1.9 — a reverse engineering workbench for agents
zfinix/eleven is a reverse engineering workbench that helps ACP agents analyze binaries and apps and propose patches.
Release
artificialanalysis/aa-agentperf-local v0.3.4 — Expanding local LLM benchmarking recipes
ArtificialAnalysis's aa-agentperf-local v0.3.4 adds local LLM benchmarking recipes for NVIDIA DGX and RTX Spark along with new models.
Release
surya-koritala/glyd v0.32.1 — faster Gemma 4 generation
surya-koritala/glyd v0.32.1 improves token generation speed for Gemma 4 under certain conditions and fixes a model invocation issue.
No matching stories.