Recently updated open-source projects

MathLens is a Cursor Agent Skill that accepts a math problem as text or an image and carries it through analysis, visual explanation, storyboarding, voiceover generation, and Manim animation. Its workflow validates audio timing and writes synchronization details back into the storyboard before rendering. The repository includes initialization, TTS, structure-checking, validation, and rendering scripts for producing step-by-step teaching videos.

Documenso is an electronic document-signing application that can be used as a hosted service or deployed on infrastructure you control. Teams can run and inspect the code instead of relying entirely on a closed signing provider, with documented options for Docker, Kubernetes, and several cloud platforms. The project remains open source, although external pull requests are currently limited to contributors invited by the maintainers; other proposals should be filed as detailed issues.

Voicetypr turns a global hotkey into voice input for the application currently in focus on macOS or Windows. By default, recordings are transcribed locally with a downloaded model, while cloud speech services are used only when the user selects one; telemetry is opt-in and disabled by default. Initial setup requires microphone access, a model download, and accessibility permission on macOS. It is suited to people who frequently write, code, or compose long prompts by voice.

Open Generative AI combines image, video, audio, lip-sync, and workflow tools in a shared creative interface. Users can call hosted models or connect local inference engines, with desktop installers for macOS, Windows, and Linux as well as a hosted browser version. Cloud generation generally requires an account and API key, while local models require suitable hardware and additional runtime setup.

lark-cli is the official Lark and Feishu command-line tool for both people and AI agents. It covers messaging, documents, Drive, Base, Sheets, Slides, calendars, tasks, mail, meetings, wiki content, and other business domains, with companion agent skills. Three command layers provide friendly shortcuts, curated API commands, and raw platform access, alongside structured output, pagination, schema inspection, and dry-run previews. OAuth authorization is required, and agents act within the permissions granted to the selected user or bot identity.

This library gives medical and biomedical researchers more than 550 skills spanning evidence discovery, protocol design, clinical statistics, bioinformatics, academic writing, and lab operations. The skills work with several SKILL.md-compatible agents, including Claude Code and Codex, and the project states that each one is evaluated through its medical research skill audit framework before release.

SnapOtter is a self-hosted file-processing platform for images, video, audio, PDFs, and general documents. Users can run conversions, compression, editing, OCR, transcription, and on-demand local AI through its web interface, REST API, or reusable pipelines, while keeping files inside their own network. A single Docker command starts the basic installation; the documented production setup adds separate Postgres and Redis services for durable jobs and larger deployments.

Recordly combines cross-platform screen capture with a renderer-driven editor for trims, zoom regions, speed changes, annotations, audio overlays, styled backgrounds, cursors, and webcams. Projects can be reopened and exported as MP4 or GIF. Capture behavior varies by operating system: macOS requires version 14 or newer, and Linux system audio typically depends on PipeWire.

Agent Sprite Forge turns natural-language requests into a structured 2D game-asset workflow. An agent generates raw visuals, then local processors remove backgrounds, split frames, validate results, and export PNGs, sprite sheets, GIFs, layered maps, and scene data for Godot, Unity, or other 2D projects. Standard sprite and map skills require built-in image generation. The denser video-to-sprite workflow depends on Grok Build's image-to-video tools, though exported frames can be reprocessed locally with ffmpeg and Pillow.

anarlog is a local-first meeting notetaker that records meetings, transcribes audio on the device, and saves each result as a Markdown file on disk. Notes can be searched or synchronized with tools of your choice, without a hosted account or tracking backend. Summaries can use OpenAI, Anthropic, Gemini, OpenRouter, Ollama, LM Studio, or another OpenAI-compatible provider. This repository continues the former Hyprnote open-source path and is not the team's current char codebase.

SimpleX Chat routes messages through disposable relay queues and uses pairwise connection identifiers instead of a global username or account identifier. Mobile, desktop, and terminal clients are available, and users can run their own messaging servers. The platform is still relatively early, so people relying on it for sensitive workflows should evaluate its current limitations and possible bugs.

This collection offers 75 skills for designers and builders working on web interfaces, animation, asset sourcing, and prompt creation from videos or existing pages. Its plain Markdown, folder-based format lets compatible agents load only the relevant SKILL.md workflow instead of depending on a single vendor-specific runtime.

Organize remote hosts, restore sessions, split terminals, and transfer or edit files through a dual-pane SFTP browser on macOS, Windows, or Linux. The built-in Catty Agent can run diagnostics and coordinate multi-host operations from natural-language requests. Netcatty connects to remote or local shells through SSH and related protocols; it is not a shell replacement.

paper2patent provides prompt templates and an agent skill for transforming a research paper into a Chinese invention-patent application draft. It guides the model through the abstract, up to ten claims, the specification, and black-and-white flowcharts or block diagrams under explicit formatting rules. Supporting scripts can generate figures, export DOCX or PDF files, and check whether a draft introduces content not supported by the source paper. The workflow can be loaded into Codex, Claude Code, Cursor, Windsurf, and similar skill-aware tools.

Apply a focused set of skills to context degradation, compression, memory design, tool interfaces, multi-agent coordination, and long-horizon task execution. The repository also covers operational concerns such as evaluation, LLM-as-a-judge methods, harness safeguards, and self-improvement loops. Complete examples show how several skills can be combined into larger agent systems across different hosts.

BrowserAct gives AI agents commands for navigation, element inspection, clicking, form input, and data extraction while keeping parallel tasks and accounts in separate sessions. For protected pages, it can use stealth modes, CAPTCHA handling, or a remote-assist link for human takeover. Core browser automation is free, while managed proxies and more than five stealth browsers are paid.

drawio-desktop wraps the core draw.io editor in an Electron application for creating and editing diagrams on a computer. Official binaries are distributed through GitHub Releases. The app is licensed under Apache 2.0, can be used for any purpose, and is provided as-is.

Attu provides web and desktop workspaces for switching among Milvus clusters, browsing collections, importing or exporting data, running vector searches, and tracking backup, restore, copy, and other long-running tasks. Its optional AI agent can query data and perform approved administrative actions through natural language. The current v3 release is still distributed as a beta and supports Zilliz Cloud or open-source Milvus 3.x, not Milvus 2.6 or earlier 2.x versions.