Recently updated open-source projects

Hollama is a browser-based LLM chat client that can connect to multiple Ollama and OpenAI servers. It supports text, vision, and reasoning models, customizable system prompts, and advanced Ollama settings. Conversations are stored locally in the browser and can be imported or exported; desktop releases and Docker self-hosting are also available.

AingDesk is a desktop and server AI assistant that brings local models, mainstream model APIs, and local knowledge bases into one interface. Users can build agents, search the web, connect MCP services, and share configured assistants for others to use. Desktop releases are available for macOS and Windows, while the server edition can run with Docker; users still need to install local models or configure the relevant provider credentials.

Founder Skills bundles reusable Claude Code workflows for SOP creation, conversion review, product requirements, go-to-market planning, content hooks, lead magnets, and social writing. The pack can be installed in full or one skill at a time through npx or manual copying. A project-level FOUNDER_CONTEXT.md file lets users supply company details, audience, positioning, brand voice, and goals so the same workflows can be adapted to different businesses or clients.

This repository packages four agent skills for analog-circuit work: introductory ngspice simulations, gm/ID characterization and sizing, a PTM transistor-model library, and a Sky130A PDK workflow. The skills include runnable scripts, examples, plots, and model assets for teaching, research, and early design exploration. Users need ngspice, Python, and the listed scientific packages. The Sky130 skill explains how to install or locate the PDK and run PVT and Monte Carlo checks, but it does not bundle the PDK payload.

kftray provides both a system-tray desktop app and a terminal UI for managing many Kubernetes port forwards from shared configuration. It watches pod changes and reconnects after restarts or network interruptions, supports TCP and relay-based UDP forwarding, and can expose local services through a cluster. Optional HTTP inspection is useful for debugging but can capture sensitive request data.

M-Courtyard provides a graphical, local workflow for preparing data and fine-tuning language models on Apple Silicon. Users can import TXT, PDF, or DOCX files, clean and segment them, generate instruction datasets, run MLX-based LoRA training, and test the adapter in a built-in chat. Models can be exported for Ollama, mlx-lm.server, or LM Studio. The application requires macOS 14 or later and an M1, M2, M3, or M4 Mac, with local Python and mlx-lm as the core runtime; 16 GB or more memory is recommended for 7B or 8B models.

skills-npm lets packages ship matching Agent Skills alongside their code, then links those skills into detected coding-agent directories. A one-time setup adds a prepare hook, an ignore rule, and the first synchronization so future dependency installs refresh the links automatically. Configuration can limit packages, skills, agents, or monorepo traversal.

Skill-kits is a local-first, single-binary workspace for inspecting and managing the Skills that Codex, Claude Code, Gemini CLI, and custom agents actually read. Its shared GUI and CLI distinguish global and project Skill directories, plugin caches, managed copies, and read-only sources, then support enablement, deployment, removal, and drift handling. It can also expose the runtime capabilities bundled by Codex plugins. The current setup is macOS-first and is installed from source with Cargo.

The editorial-card-screenshot skill reorganizes source material into editable, screenshot-ready HTML cards and can optionally render PNG output. It supports portrait, landscape, square, widescreen, and banner-like aspect ratios by adapting the information layout rather than merely resizing one template. Installation requires the complete skill directory, including its assets, references, scripts, and optional agent files.

QTranslate is an extensible desktop translation application built around a quick select-and-press workflow: highlight text in any app, press Ctrl+Q, and view the translation without opening a browser. Its main window also handles longer text, engine switching, screenshot OCR, text-to-speech, spell checking, dictionaries, and history. Translation and language services are delivered as runtime-installable JAR plugins, so individual providers can be replaced independently. Java 11 or later is required, and some bundled services need user-supplied API credentials.

cctop is a keyboard-first macOS menu bar app that consolidates session status from tools such as Claude Code, Codex, OpenCode, and pi. It shows which sessions are active, idle, finished, or waiting for input, then jumps to the most specific available target: a terminal pane, editor window, desktop thread, or project. Each coding tool must first be connected through its plugin or event hooks. Session state is stored locally as small JSON files, with no analytics, telemetry, or session uploads.

AutoPTZ is a cross-platform desktop app that detects people in camera feeds and drives pan, tilt, and zoom controls to keep a selected subject framed. Users click a person to lock the target, while optional face recognition and appearance re-identification can recover identity after occlusion. It supports multiple cameras and PTZ backends including VISCA over USB or IP, ONVIF, and NDI. Tracking features require the appropriate models; when they are unavailable, affected controls are disabled while live preview remains available where possible.

JiwuChat combines direct and group messaging, media and file transfer, multi-device synchronization, WebRTC audio and video calls, and screen sharing in a lightweight cross-platform client. AI assistants from providers such as DeepSeek, Kimi, and Gemini can participate in private or group conversations. A public web experience is available, while users who need their own service can deploy the separate jiwu-chat-core backend with Docker. On macOS, the downloaded package may require manually clearing quarantine because of signing-related warnings.

Resend Skills equips coding agents with workflows for the Resend email API, secure agent inboxes, the Resend CLI, React Email components, and email delivery best practices. The collection can be installed as a plugin across several coding environments and also registers a hosted MCP server that authenticates through OAuth. Direct Resend use generally requires an account, a verified domain, and the appropriate credentials.

Nori Skillsets Client downloads and activates packaged skillsets or individual skills from the Nori registry, translating each configuration into the on-disk format expected by Claude Code, Cursor, Codex, Gemini CLI, and other supported agents. It suits users who maintain several agent setups from one profile library. Node.js 22 and macOS or Linux are required, and switching a profile removes manual changes made inside an agent's generated configuration directory.

onedriver gives Linux computers direct filesystem access to Microsoft OneDrive without downloading the whole account in advance. Files are fetched only when an application opens them, remote changes appear in the mounted view, and local edits are written back to OneDrive. Users can add multiple accounts through a desktop launcher or mount them from the command line. It is designed mainly for convenient desktop use rather than behaving like a traditional full-sync client.

asta-skill is an optional instruction layer for Ai2’s Asta MCP server. It maps requests for paper search, identifier lookup, authors, citation traversal, batch queries, or full-text snippets to appropriate tool calls with safer defaults and workflow guidance. Access still requires the Asta MCP server and an API key; the tools work without installing the skill.

Legends-Of-Heroes is a C# client-server game framework based on ET 10, Unity 2022.3.62, and .NET 10. It includes state synchronization, hot updates, skills, buffs, behavior trees, timeline events, server-side collision logic, and a playable ball-survival demo. Setup requires the specified toolchain, and commercial use of the underlying ET framework requires a license from its author.