AIKON is an open-source AI chat client for Nokia Series 40 and Symbian S60 phones, plus a Go server. The Java ME phone app runs on CLDC 1.1/MIDP 2.0 devices and lets users choose among Claude, OpenAI, Gemini, and Grok models, send keypad-typed messages, browse paginated replies, search the web for news, weather, and exchange rates, and use photos, voice transcription, calendar entries, and to-do actions. It also provides chat history, saved replies, message actions, quick prompts, and interfaces in eight languages. The phone communicates with the server over TLS 1.0 using a certificate from a private certificate authority installed on the device; the server then uses modern HTTPS or provider SDKs to access the model services. The server provides pairing, per-device access tokens and usage limits, SQLite chat storage, optional speech-to-text, photo handling, and a localhost-only admin API. It is tested on the Nokia 6300 and Nokia E63, requires users to supply their own model-provider API keys, and is distributed under the MIT license.
audiocn is a React audio UI component library and shadcn registry for level meters, visualizers, waveforms, faders, knobs, channel strips, mixers, players, sound pads, and audio-device controls. Components are copied into an application with the shadcn CLI rather than installed as a package, allowing their source to be modified. The project includes Web Audio hooks for microphones, system or tab audio, audio analysis, mixer state, playback, sound effects, waveform data, and device selection; components can also receive data from another audio engine. It is built with Base UI and Tailwind CSS, uses shadcn-compatible theme variables, and includes keyboard controls, ARIA roles, and reduced-motion support. The repository contains the documentation site, registry source, component source, audio core, examples, and tests, and is licensed under the MIT License.
Backburner is a pre-release local-LLM inference tool by StayLameBro that uses an iPhone alongside an Apple Silicon Mac to run Qwen3.8-27B over a USB-C connection. Its llama.cpp-based engine splits prefill computation between the Mac and phone: the Mac runs the initial model layers while the iPhone runs later layers, and for contexts beyond the Mac's available 8-bit capacity the phone stores older key-value pages and computes attention over them. The phone can also use its GPU and Neural Engine during decoding, while the Mac uses custom Metal, SME2, and DFlash2 speculative-decoding kernels. It exposes an OpenAI-compatible server at localhost:8080 and supports iPhones 15 Pro or newer and M-series iPads, although the documented testing uses iPhone 16 Pro Max and 17 Pro Max devices. The project requires a 10 Gb/s USB-C cable, distributes its iPhone app through AltStore or an Xcode build, and is released under the MIT license.
ClearAI is a local-first ontology discovery and exploration platform delivered as a native DeepSeek Harness (DSH) plugin. It uses an epistemic loop—question, judgement about what would disprove a claim, a potentially failing test, evidence, and a bounded conclusion—to build a domain ontology whose concepts, relations, entities, evidence chains, boundaries, and support levels can be searched and explored. Refuted hypotheses remain in the record, while contradictory conclusions are surfaced as conflict readings for a human to retain or retract. ClearAI adds this epistemic layer through a DSH agent preset and client module without changing the DSH engine. It is distributed as a prebuilt npm plugin, supports English and Chinese output, requires DSH, and is licensed under Apache-2.0.
Comma is an open-source, sessionless personal AI agent built by AFK Inc. It turns requests into persistent Tasks and runs agent loops that plan, execute, review, and verify work until completion, pausing in a Needs Review state when a human decision is required. Loops can monitor inboxes, repositories, or feeds and wake the agent only when action is needed; work is organized on a board, and routines can deliver scheduled briefings. Comma can use connected computers for files, commands, and other operations, and can invoke agents such as Codex and Claude Code as workers or sub-agents. It is available as a hosted service or can be self-hosted with Docker Compose, using a stack that includes PostgreSQL, Redis, MinIO, ClickHouse, and Mailpit. The runtime is primarily built with Elixir/OTP, with Lean and TLA+ used to machine-check selected reliability and distributed-protocol properties. The repository's original code is licensed under AGPL-3.0-only.
decider is an open-source AI model family and Python package for one-pass typed decisions with calibrated probabilities. Instead of generating text, it reads a text or JSON state and a set of typed questions—Choice, Score, or Noul yes/no questions—and returns a probability distribution over each question's allowed answers from one forward pass. It renders each question as an answer slot, reads the model's hidden state at that slot, projects it onto the valid option-label tokens, and applies a fitted temperature; there is no decoding or output parsing. The package provides `decide()` and `system_one()` interfaces, an HTTP server with `/decide` and `/v1/systemone` endpoints, and GGUF loading through llama.cpp. Models are available in sizes intended for CPUs and laptops as well as CUDA and Apple Silicon GPUs, and the repository includes training, calibration, serving, evaluation, and schema-caching components. The project is an independent reproduction of the System One model class, is not affiliated with TypeSafe AI, and is released under the Apache 2.0 license.
An open-source tool that lets a coding agent create narrated explainer videos and product demos as MP4 files on the local computer. The agent writes a Markdown script and JavaScript scene definitions; Kokoro generates local speech, Whisper provides word-level timing, Chrome renders the scenes on a canvas, and FFmpeg assembles the narration, animation, music, sound effects, and captions. Scenes can include hand-drawn graphics, charts, code, icons, screenshots, or reconstructed product interfaces with simulated pointer and typing actions. explainroo saves stills and frame sheets for review, and provides layout and speech checks for overlapping or clipped text, pronunciation errors, and small text in feed-sized videos. It supports multiple visual styles, output aspect ratios, adjustable pacing, and optional AI-generated illustrations through OpenRouter. The software is MIT licensed and requires Node.js, FFmpeg, and Chrome or Chromium; its voice and timing models run locally after setup.
Felucca is free, GPL-3.0-only custom firmware for the M-VAVE FM-1 synthesizer, developed by Hügelton Instruments. It replaces the device's official firmware with a multi-engine synthesizer supporting thirteen engines, four tracks with eight shared voices, a 64-step sequencer per track, chord and scale keyboard modes, an arpeggiator, modulation, per-track and master effects, presets, projects, and USB MIDI and audio. It can be installed over USB through a web installer in Chrome or Edge, or with a Python terminal tool, without additional hardware. Its web editor exposes synthesizer parameters, a six-operator FM patch editor, step grids, mixing, preset management, sample upload and recording with trimming, and full backup and restore; the installer can also return the device to official firmware. The repository includes firmware, hardware-layer, web, build, installer, sample-upload, and test sources, with bundled fonts, DSP components, SDK files, and samples retaining their stated separate licenses.
Fieldwatch is a receive-only Android observer for nearby Wi-Fi access points and Bluetooth Low Energy advertisements. It passively listens through the phone’s built-in radios, working offline without a dongle, account, backend server, or ads. The app provides configurable views, filtering, an extensible device-signature library, local storage, and reports of detected radio sources. It is distributed as a sideloadable APK for Android 10 and later, with source code and installation materials in the repository, and is licensed under the MIT License. It does not support Wi-Fi monitor mode, Bluetooth Classic inquiry, cellular detection, or direction finding.
Hairline is a dependency-free, ESM JavaScript library of interactive isometric line figures for React and other DOM environments. Its figures respond to pointer movement, including pillars that rise around the pointer, layered objects that separate, cards that stand up, and other scenes such as terminals, keyboards, routers, and commit graphs. The package provides React components and DOM functions, supports Svelte-style actions, and lets each figure accept intensity, theme, accessible label, and caption callbacks. Figures draw at a 5:4 aspect ratio; the library includes server-rendering behavior, reduced-motion handling, shared animation scheduling, CSS custom properties for theming, and no runtime dependencies. It is distributed through npm under the MIT license. Hairline also provides hairline-create, a skill for coding agents that generates new figures as standalone HTML files according to the project's rules.
Inlark is a keyboard-first desktop email client for Linux, Windows, and macOS that connects directly to users' mail servers through JMAP or IMAP/SMTP. It combines multiple accounts in a unified, account-marked inbox, supports keyboard commands such as J/K navigation, archiving, replying, search, and Ctrl+K actions, and provides undo for mailbox actions. Searches run server-side across folders and years, while IMAP accounts build a local index in the background; the interface virtualizes large inboxes by drawing only the messages currently on screen. Inlark stores credentials in the operating-system keyring when available, requires encrypted connections, has no hosted service, subscription, Inlark account, or telemetry, and includes native desktop notifications and mailto handling. It is distributed as desktop builds and source under the GNU Affero General Public License v3.0, but the repository describes it as an early preview; Gmail and Microsoft OAuth are not yet supported.
Jevbox is a self-hostable, permission-aware document library from Extend for uploading, visually browsing, and inspecting parsed documents. It uses Extend for document parsing and stores parsed structures, heading outlines, source passages, and original files; supported viewers include PDF, DOCX, XLSX, PPTX, Markdown, code, JSON, CSV/TSV, images, archives, audio, and video, with a safe download fallback for unknown formats. Its JEV retrieval system performs permission-filtered hierarchical beam search across categories, documents, and sections, followed by usefulness scoring of source passages. Users can ask questions through the AI SDK or web chat, view inline citations and source previews, attach accessible documents to scope an answer, and connect through REST and MCP APIs. Uploads can be automatically filed by walking existing folders and validating model-proposed branches, while organization permissions are evaluated through SpiceDB. The repository supports local operation with Node, pnpm, Docker Compose, PostgreSQL, and SpiceDB, and provides Docker, Render, Helm, and AWS/EKS deployment options. It includes organization accounts, invitations, role-based sharing, durable background jobs, queued chat processing, and configurable model providers. Jevbox's original code is licensed under the MIT License; third-party code and assets retain their own licenses.
Jumper is an open-source workflow for designing and training a 22-degree-of-freedom crab robot. It uses natural-language instructions and associated tools to create robot appearances and scenes, exporting them as `.skin` and `.map` files, and to train motions such as gaits, gestures, dances, jumps, and grasps. Trained motions can be replayed and evaluated, then packaged with their controllers as `.app` bundles. Its training workflow builds on mjlab, rsl_rl, MuJoCo, and MuJoCo Warp. KingKong Robotics maintains the project; its project materials are licensed under Apache-2.0, while third-party materials and generated outputs have separate licensing terms.
Kharcha is a private spending ledger for Android and other Expo-supported platforms. It parses bank transaction SMS messages or pasted messages on the device, extracts debit or credit amounts and merchant details, preserves each raw message beside the parsed entry for review and correction, and stores the ledger locally in SQLite. Users can add categories, budgets, and insights; the Android reader scans messages only from enabled sender IDs, filters common OTP and promotional patterns, and uses heuristic parsing with uncertain results sent to review. The app works without an account or backend. An optional backup flow encrypts the ledger on the device with AES-GCM and sends only ciphertext to a small Render API backed by Postgres; the recovery code contains the access token and encryption key. iOS and web versions support manual or pasted entries, while general inbox reading is Android-only. The project is MIT-licensed. Its README notes that parsing may miss or misclassify messages, the Android build has not yet been smoke-tested on a physical device, and Google Play distribution using READ_SMS requires policy approval.
LinkedIn Agent Skill is a free, MIT-licensed set of eleven Claude skills for drafting and managing LinkedIn content. Its commands generate posts from 21 hook formulas, classify and write comments and replies, score profiles against a 12-part rubric, plan weekly publishing and engagement, create carousel copy, repurpose source material, draft direct messages, triage inboxes, and analyze previously published posts. The skills produce copy-ready drafts for manual review and publication rather than posting to LinkedIn automatically. The package also includes two dependency-free Python scripts: humanize.py removes invisible Unicode characters, normalizes typography, and replaces entries from an editable 113-term vocabulary; detect.py scores drafts on burstiness, specificity, slop density, fingerprint, and voice using local heuristics. These checks run on the user's machine and are not integrations with commercial AI-detection services. The repository states that the skills do not upload text, fabricate unsupported metrics, or fill in unspecified numbers, instead using placeholders such as {{your number}}. It can be installed in Claude or Claude Code, as a Claude plugin, or by copying the skill folders into a project; MIT licensed and made by Jake Schincariol.
Lipflow is an open-source, cross-platform silent lip-reading dictation tool that uses a webcam to turn mouthed words into text pasted at the cursor. It runs locally on macOS, Windows, and Linux: face landmarks are tracked live, 96×96 grayscale mouth crops are sampled at 25 fps, and an Auto-AVSR visual-speech-recognition pipeline combines a 3D-convolutional ResNet front end, Conformer encoder, Transformer decoder, CTC scoring, and a subword language model with beam search. Users can train the language model and lip reader on their phrasing and face, import local Wispr Flow dictation history, and optionally apply cleanup through offline rules, an on-device Qwen model, Claude, Codex CLI, or Ollama. Whisper mode combines lip and audio input with an audio-visual model while the hotkey is held. The repository is MIT-licensed, although its LRS3-trained model weights are identified as for non-commercial research use.
livenerf is an open-source benchmark for detecting post-launch capability changes in frontier models. It calibrates a panel of intermittently solved questions from GPQA Diamond, MMLU-Pro, competition mathematics and AIME, then runs the same panel repeatedly through a hermetic, pinned Claude Code harness and compares paired per-item scores with a launch-period baseline. The benchmark records raw prompts and responses, usage, latency, CLI and harness versions, and token counts; it also runs a control model and applies preregistered statistical thresholds rather than relying on an LLM judge. It is built on the Inspect evaluation framework, supports scheduled daily runs on Linux, macOS and Windows, and is distributed under the MIT license. The repository describes it as an independent project, not affiliated with Anthropic.
Mesh Avatar Studio is a local editor and coding-agent-assisted tool for turning a single illustration into an animated 2D mesh avatar. A coding agent prepares the avatar by reading the image, placing the rig, cutting it into layers, and reviewing fixed poses; the editor then provides a live preview for adjusting mesh points, testing head turns, tilts, blinking, lip-sync vowel shapes, breathing, and hair movement. Users can rebuild layers after changing eye, hand, head, hair, or bun boundaries, and can generate imported drawn eye and mouth variants through Codex. It runs locally with Node.js 22.17 or later and is released under the MIT License, except for the bundled Miko sample character and its assets.
Mobile Dev is a Codex plugin for developing and debugging Android and iOS applications alongside a Codex desktop chat. It supports Expo, React Native, SwiftUI, and other native mobile projects, and can stream iOS simulators, Android emulators, and connected devices into a device panel. The plugin provides device selection and booting, taps, drags, text input, accessibility inspection, screen or element annotations, and MCP-based app control. It combines iOS unified logs, Android logcat, and Metro messages; logs can be searched, filtered, scoped to the foreground app, and sent to chat with an error stack trace. Its performance tools record CPU and memory activity, thread charts, display FPS, Android frame-pacing and jank statistics, and selected interaction ranges for comparison and analysis in Codex. The prebuilt distribution targets Codex desktop on Apple Silicon Macs and includes bundled runtimes. It is installed through Codex's plugin marketplace and requires the relevant Xcode, Android SDK, simulator, emulator, or device tooling. The repository is maintained by Callstack's incubator organization.
An open-source SDK and firmware project for building custom Muse gadgets from off-the-shelf ESP32 boards, Raspberry Pi devices, or other Linux boxes. The ESP32 and Linux device SDKs let developers connect displays, buttons, audio, sensors, actuators, and custom commands, then pair the gadgets with the Muse app on iOS or Android. Gadgets require an SDK token and developer mode in the Muse app before pairing. The project is licensed under Apache License 2.0, with stated exceptions for specified third-party files and dependencies that retain their upstream licenses.
MyGo is a Go framework for building desktop applications with either a web frontend or a native Go UI. Web windows use the operating system's webview and communicate with Go services through a TypeScript client generated from Go code; native windows use MyGo's GPU-rendered UI toolkit without HTML, JavaScript, or a webview. It provides desktop APIs such as windows, menus, tray icons, dialogs, notifications, global shortcuts, deep links, and file associations, and supports macOS, Linux, and Windows through pure Go without cgo. Its distribution tooling creates app bundles, disk images, Windows installers, Debian packages, and Linux install scripts, with code signing, notarization, and signed delta auto-updates. MyGo is distributed under the MIT license.
NeonPlan 3D is an open-source Home Assistant integration and dashboard card for drawing a home's floor plan and viewing it as a live 3D model. Its built-in editor supports floors, rooms, walls, doors, windows, furniture, roofs, stairs, outdoor areas and other structural elements; the 3D view updates alongside the plan. In Home Assistant, the view can control lights, covers, windows, doors, climate, media and other devices, while displaying cameras, sensors, heatmaps, alerts, scenes and room details. It includes a sidebar panel and the `custom:neonplan3d-card` dashboard card, with wall-tablet features such as kiosk mode, idle return and night dimming. The integration and core features are free under the MIT license and installable through HACS or by copying the custom component into Home Assistant; optional furniture packs and Pro add-ons are sold separately. It stores plan data, pictures and packs in Home Assistant storage and contacts the vendor's server for pack updates only when a license key is entered.
OmacVM is an open-source tool that installs and runs Omarchy, an Arch Linux ARM desktop, in a virtual machine on Apple Silicon Macs. A setup script builds the VM and configures it for OmacVM.app, UTM, VMware Fusion, or Parallels Desktop, while Mac-side helpers pass hardware and system features to the VM over a private network secured by a secret token. These integrations include trackpad gestures, macOS-like scrolling, keyboard and media keys, Wi-Fi, Bluetooth, audio, battery status, displays, camera, microphone, clipboard sharing, theme and wallpaper synchronization, and the Omanotch bar beside the MacBook notch. The command-line interface can build VMs, switch features, update installations, check compatibility, manage VM resources, and list VMs. It requires an Apple Silicon Mac, macOS 15 for OmacVM.app or macOS 14 for the other virtualization routes, and about 30 GB of free disk space. OmacVM is released under the MIT license.
omnirush.ai is a desktop coding agent for macOS and Linux that works on local projects. It lets users select models and effort levels, add provider keys, and configure skills and MCP servers. Its Git workflow can clone repositories, create branches and worktrees, make commits, push changes, and open or review pull requests through gh, with approval prompts for tool calls and write commands. The app is distributed through platform-specific installers and releases under the repository's MIT terms, while code in the ee/ directory has a separate license.
OpenDots is an open-source, self-hostable template for building persistent AI coworkers, called Dots, each with its own role, instructions, permitted tools, browser profile, files, and workspace. It provides a web and mobile workspace with Spaces, searchable pages, visual editing, page-specific conversations, and specialist Dots. Its workflow connects agents to interfaces through AG-UI and uses CopilotKit components, TanStack AI for model streaming and server-side tool execution, Intelligence for durable conversation Threads, and Channels SDK for Slack. Dots can perform browser and workspace actions through OpenBot computers, with permissions, activity records, human takeover, and persistent browser profiles and files. Drafts can be paused for human approval before being saved as pages; the template also includes text conversations, configurable voice calls, background work, research, memory, and automatic-learning flows. OpenDots is a starting point rather than a hosted product: operators clone and run the application, configure its model, conversation, channel, speech, browser, and computer services, and customize the workflows. The repository describes the project as early-stage, with some connected-service behavior still requiring verification. It is distributed under the MIT license.
Papermorph is an AI skill that turns PDF books into animated, narrated, interactive web books. Its pipeline converts a PDF into a book plan, storyboards, narration, animation, quizzes, and a browsable web-book experience for a specified target audience. It is currently designed to run with Opus 5.5 and supports English content; the project says image-model support, multilingual output, and background music are not yet available. It can be installed with `npx skills add DozenTwelve/Papermorph --skill papermorph --agent claude-code` and invoked in Claude Code with `/papermorph`. The repository includes a live bookshelf of examples and is released under the MIT License.
pi pod is a self-hosted platform for running pi coding-agent sessions in isolated remote sandboxes called pods. Its command-line client and iOS and Android apps connect to a server that manages pod lifecycles through a REST API, session gateway, lifecycle workers, and Postgres; a native sandbox service runs multiple isolated pods in one container, with pi inside each pod behind a small shim. The CLI can launch and attach to a pod for the current project, while the server uses Zitadel for OIDC identity and stores no passwords. The repository includes a Docker Compose deployment, CLI package, server, sandbox service, mobile apps, and self-hosting tools. It requires a Linux host with Docker Compose, at least 8 GB of RAM, and Node 22.19 or later for the CLI, and is licensed under AGPL-3.0-only.
RE:Dox is an open-source, Apache-2.0-licensed structured-data engine for .NET, developed by CAPCOM as a core component of its REX technology. It parses JSON and other supported formats into a compact token DOM/intermediate representation in which fixed-size tokens can retain source offsets, lengths, links, or inline values; this representation supports reading, editing, serialization, deserialization, and format conversion. The mutable token model provides node-like insert, remove, and replace operations without rebuilding a managed object tree, while lazy decoding can defer materializing strings, numbers, timestamps, and binary values. Its asymmetric I/O design writes known values directly through a DataWriter, while deserialization first builds an addressable token DOM for look-ahead, out-of-order constructor binding, reference resolution, and optional automatic parallel processing of large arrays. The core packages support JSON, JSON5, CBOR, MessagePack, INI, and the binary DOX format, with TOML, XML, HTML, and CSV integrations listed as preview components. A common DataReader/DataWriter and DataConverter model enables conversion between formats such as JSON, CBOR, MessagePack, TOML, XML, and DOX, subject to structural projection where formats have different data models. JSON5 editing can preserve comments and other trivia, and JsonSequence supports asynchronous processing of NDJSON or large top-level JSON arrays without loading the entire input. Compatibility packages are provided for System.Text.Json, Newtonsoft.Json, and DataContractJsonSerializer. The project requires .NET 10 or later, is distributed through CAPCOM.REDox.* NuGet packages, and is under active development; its README notes that APIs, package boundaries, and preview-format support may change before the first stable public release.
Remocn Studio is a macOS agent-native video editor that creates editable Remotion projects from natural-language direction. A coding agent builds the video while the user watches a live preview and provides notes by typing, selecting elements in the preview, or framing parts of a video snapshot. The resulting project contains readable source code and production documents for analysis, branding, scripting, motion, building, choreography, and review. The app supports direct manipulation of preview elements, reusable assets and components, brand settings, motion-design skills, and exports to MP4, WebM, GIF, or ProRes at up to 4K. It drives supported coding-agent command-line tools, including Claude Code, Codex CLI, GitHub Copilot CLI, and Grok Build, using the user's existing sign-in. Projects remain in folders on the user's Mac as standard Remotion projects rather than a proprietary format. Remocn Studio is built with a Tauri v2 Rust core, a Next.js static webview, and a Bun sidecar for agents, history, previews, and rendering. It requires macOS, an installed and authenticated supported agent, and Remotion project dependencies. The studio is released under the MIT License; Remotion has separate license terms.
SCM is a local-first macOS application for searching photos and videos by visual content, visible text, and spoken dialogue. It indexes whole files with local vision models, segments videos into scenes so searches can jump to a timecode, runs Tesseract OCR for literal text matching, and uses Whisper transcripts for exact spoken-word searches. An optional local LLM can answer questions over extracted dialogue, OCR, and filename evidence with citations. The app imports files through manual selection, drag-and-drop, or watched folders. It content-hashes files for rename-proof deduplication, stores embeddings and media metadata locally, and can re-embed a library when the selected vision model changes. Model weights and language packs download once; subsequent inference runs offline, with no accounts, uploads, or telemetry. SCM is distributed as an Electron application for macOS, including a Homebrew cask and packaged DMG/ZIP builds.
Strands Decider is a small decision model for agentic AI workflows, developed for use with the Strands Agents SDK. It selects among options, answers yes/no questions, or scores inputs against an ordered rubric, returning a calibrated confidence value for each decision. These capabilities support tasks such as model routing, tool selection, argument checking, triage, guardrail checks, and evaluation. The model uses a pretrained decoder LLM torso with its language-modeling head replaced by a pointer head. The head compares hidden states at the answer position and at each option's final token, scoring all options in one forward pass without text generation or a decoding loop. The same masked-softmax design supports its choice, noul, and score question types. The repository includes a command-line interface, a local HTTP server, an example that gates a Strands agent's tool call, and optional image inputs through a vision mode. It can run on CUDA, Apple silicon through MLX or MPS, and CPU. The project is distributed under the Apache License 2.0 and can be installed with pip.
Television is a graphical interface for personal AI agents developed by Telepath. It lets an agent create, display, and modify persistent interactive artifacts—including documents, data, visualizations, live web pages, and applications—in a visual workspace used alongside the agent's existing chat interface. The system consists of a lightweight server running with the agent, a skills bundle that teaches the agent how to create and manage artifacts, and a client app. It provides an installable Electron app for macOS and an experimental browser interface; the agent itself must run on Linux or macOS, and the project states that Windows agents are not currently supported. The software is open source under the MIT license.
terrahour is a terminal-based world clock that renders a Braille day-and-night map alongside 24-hour timelines for tracked cities. Each city has working hours, and an overlap row shows when all tracked locations are at work; users can scrub time in 15-minute steps or jump to a specified time. It also provides daylight-saving-time transition alerts, stock-exchange sessions, configurable reminders, an ambient clock view, optional weather from Open-Meteo, compact and tmux-friendly layouts, status-line and JSON output, and a built-in database of about 12,000 cities. The application is a single Python package with no third-party runtime dependencies. Its map uses a land mask and Braille rasterization, while sun position, time-zone offsets, working-hour status, DST transitions, and overlap are calculated by dedicated modules. Network access is optional and limited to Open-Meteo geocoding and weather; setting TERRAHOUR_OFFLINE=1 disables both. It requires Python 3.9 or newer and a UTF-8, truecolor terminal, is developed on Linux, and is distributed under the MIT license.
VibeWise is a Claude Code plugin that supports learning-oriented software development while Claude writes the code. It asks the developer to propose an approach, works through architecture, tradeoffs, failure cases, and unfamiliar concepts, and uses Build, Design, and Implementation checkpoints before writing code. The developer confirms or discusses each decision; Claude then implements the agreed step, runs tests, and reports what changed, how it works, and which checks passed. Teaching support can be adjusted for beginner, intermediate, or advanced users, with configurable checkpoint frequency and project-specific learning notes and maps stored in `.vibe-wise/`. The plugin requires Claude Code and Python 3, uses no additional Python packages, has no extra account, backend, or telemetry, and is released under the MIT license.
Whirl is a full-stack, self-hostable AI chat application developed by Anterra and published under the MIT license. Built with Next.js and Convex, it routes conversations through multiple models using OpenRouter, allowing users to switch models and adjust thinking levels within one conversation. It provides long-term memory for preferences and projects, live web search and page reading, MCP integrations with OAuth, installable skills, voice input, image generation, file attachments, and editable living artifacts such as documents, charts, and interactive pages that can be shared by link. It also includes encrypted locked chats, incognito mode, message queueing, folders, conversation sharing, and an installable mobile web app.
Searchable transcript of GitHub Trending Today #52: decider, whirl, livenerf, mygo, Comma, Fieldwatch, terrahour, OpenDots — Github Awesome (14:42). Search for a phrase, then click its timestamp to jump straight to that moment in the video.
Captions sourced from the original video on YouTube, published by Github Awesome. The video, its captions and all related intellectual property remain the property of their respective owners; AINotes claims no ownership. Provided for research, accessibility and search — see the Transcript Notice and Copyright Policy.
00:00 Welcome back to GitHub awesome. This is GitHub trending today number 52. 35 trending open source projects on GitHub right now. Let's get into it. Decider gives AI agents a fast way to make structured choices. Feed it a situation and questions like which team handles this? Is it urgent or how severe is it? It returns probabilities for the allowed answers in one model pass with no free form text to parse.
00:26 The project offers small models for laptops, larger GPU options, and an API for putting those decisions into an app. Whirl puts several AI models in one chat app, so you can switch models within a conversation. It can remember preferences across chats, connect your tools through MCP, and keep generated documents, charts, and interactive pages editable beside the conversation.
00:51 There's also web search, voice input, and image generation. The full code behind world.hat is open source with a guide for running your own instance. Livvenerf tests a familiar suspicion. Does an AI model get worse after launch? It runs the same carefully selected questions through clawed opus 5.5 in clawed code each day with the prompts and CLI version fixed.
01:15 Every response is saved. Then scores are compared with an early baseline using rules published before the results. The 30-day run is underway. It hasn't made a verdict yet. My Go lets you build desktop apps in Go and choose the interface for each window. Use a web front end with a generated TypeScript bridge to your Go code or draw the whole UI natively in Go.
01:39 Both get desktop features like menus, notifications, and shortcuts. It targets Mac, Linux, and Windows with tools for packaging apps and shipping signed updates. Comma is a personal AI agent that runs around the clock and manages a request through to done. You ask once, it turns that into tasks on a board, backlog through done, and stops for your decision at the checkpoints that matter.
02:05 Small autonomous loops watch your inbox, repos, and feeds, and wake the agent only when there's something to do. You can reach it from a Mac app, a browser, or Signal and Telegram. Fieldwatch turns an Android phone into a pocket observer for nearby radio signals. It passively lists the Wi-Fi access points and Bluetoothle broadcasts your phone can hear, then lets you filter observations and build a report.
02:32 You can add signatures to recognize familiar devices. It works offline, needs no dongle or account, and keeps the data on your phone until you choose to share it. Terraore is a world clock that lives in your terminal. It draws the planet in Braille dots shaded by where the sun is right now and gives each city a 24-hour timeline so you can see which working hours actually overlap.
02:55 Arrow keys scrub time in 15minute steps and a DST radar warns you which cities are about to change their clocks. There's a stock exchange view with real trading sessions, optional weather. Explain lets you ask a coding agent for an explainer video and get back a narrated MP4. The agent writes the script and draws scenes in JavaScript. Explain handles the voice, word timing, animation, music, and export.
03:23 It even saves preview frames and checks for clipped text or mispronounced words, so the agent has a way to catch mistakes before handing you the video. Open Dots is a self-hostable starting point for building AI co-workers with different jobs. Give one dot a research role, another a writing role, and each can have its own persistent browser and files.
03:47 Their work shows up beside your chat, while a review card lets you approve a draft before it becomes a saved page. The template also includes paths for voice calls and Slack, though those need separate service setup. Redox is Capcom's open-source.NET engine for structured data built as part of its next generation game technology. Instead of treating JSON as a blob to decode and forget, it builds a compact token map you can read, edit, and convert between formats like Seabore and message pack.
04:19 It can even preserve comments while editing JSON 5. Clear AAI is a Deep Seek Harness plugin for research that asks a harder question than did the agent finish. It asks what would prove a claim wrong, runs a test, records the evidence, and gives the result a bounded conclusion. Supported findings grow into a searchable knowledge graph. Refuted claims stay visible, too.
04:42 When conclusions conflict, the system flags them for a human decision. AON puts modern AI chat on a 2007 Nokia. Type a question on the keypad, choose Claude, Chat GPT, Gemini, or Grock, and read the answer on that tiny screen. It can search the web, handle photos, and turn a voice message into text. A small Go server bridges the old phone to today's model APIs, and the project has been tested on a Nokia 6300 and E63.
05:13 LinkedIn agent skill gives Claude 11 jobs for managing your LinkedIn writing. Draft posts, plan a week, suggest comments and replies, review your profile, and turn longer material into shorter posts. A shared voice file helps the draft sound like you while local scripts flag stock AI phrasing and awkward patterns. It never posts for you. Every draft ends as copy you review and publish yourself.
05:39 Remach Studio lets you direct a launch video by talking to a coding agent. Describe the idea, watch a live preview, then click a line of text or point to a frame and say what needs fixing. The agent builds the motion design as a motion project, so you can edit the code later and export the finished video. It's a Mac OS app built around a project folder you own.
06:02 Inlark is a keyboard first desktop email client that talks straight to your mail server over Every account lands in one color-coded inbox and replies always go out from the right address. J and K move, E archives, R repl replies. Control K opens a command pallet and everything can be undone. Vibe Wise turns clawed code into a coding partner that makes you think through the build.
06:30 Before it writes a feature, it asks how you'd handle the tricky parts, helps you weigh tradeoffs, and shows you the design to approve. Then it writes and tests the code, explaining what changed. You can tune how often it stops so the lesson fits your experience and the time you have. Television gives your personal AI agent somewhere visual to put its work.
06:52 Ask for a chart, document, live web page, or small app, and the agent can place it on a separate screen where you can inspect it and keep working on it later. It runs alongside your usual chat interface and supports agents with skills and file access. Right now, the agent needs to run on Linux or Mac OS. Muse Gadget SDK lets you turn an ESP32 board, Raspberry Pi, or Linux box into a custom gadget for Muse.
07:20 Add a screen, buttons, audio, or sensors. Then pair the device with the Muse Mobile app. The repo includes open- source firmware and SDKs for both ESP32 and Linux, so you can start with off-the-shelf hardware and decide what your gadget actually does. Audio CN gives React developers the controls you usually have to build from scratch for audio apps.
07:42 Level meters, waveforms, faders, knobs, channel strips, and a full mixer. Add a component with the Shad CN CLI, and its source lands in your project, ready to restyle or change. It also includes web audio hooks to connect those controls to microphones, playback, and live audio signals. Lipflow lets you type without speaking. Hold a key, silently mouth a sentence toward your webcam, and it pastes its best guess at your cursor.
08:09 It runs on Mac, Windows, and Linux, and you can train it on your face and vocabulary. Lip reading still makes mistakes, so an optional cleanup model can refine the text. There's also a whisper mode that combines lips and audio for better accuracy. Strands decider handles the small decisions an AI agent makes all day. Which tool to call, where to route a request, or whether an answer needs another check.
08:36 Give it options or a scoring scale, and it returns a choice with a confidence score instead of writing a long response. It's a compact model you can run locally with examples for plugging those decisions into a strand's agent. Backburner puts your iPhone to work when your Mac runs a local 27 billion parameter model. Connect them with a fast USBC cable and the phone helps process long prompts or holds older context that won't fit comfortably on the Mac.
09:05 In the project's tests on an M4 Pro MacBook and iPhone 17 Pro Max, that cut prompt reading time for large inputs. It's still a pre-release project with specific hardware requirements. SCM or screen memories makes the photos and videos in your Mac folders searchable by what's actually inside them. Describe a scene in plain English and it can jump to the matching shot even partway through a video.
09:31 You can also search words visible on screen or exact lines of dialogue. Watched folders keep the library current and the models run on your Mac after their first download. Mesh Avatar Studio turns one illustration into a 2D character that blinks, talks, tilts its head, and moves its hair. A coding agent cuts the image into layers, and builds the rig.
09:55 Then you open the local editor, drag points that look off, and watch the pose update live. You can test eye and mouth shapes before using the avatar, so the final movement feels less like a flat picture. Omnirush is a desktop coding agent for Mac OS and Linux. Open a project, choose a model and reasoning level, then let the agent work through code changes and git tasks in one place.
10:19 It supports skills and MCP tools, plus workflows for branches, work trees, commits, and pull requests. You can connect an Omnirush account or add your own provider keys. Neon Plan 3D turns your home assistant dashboard into a live model of your house. Draw rooms, walls, doors, and furniture in its built-in editor, then link them to your devices. Lights glow in their actual colors, blinds move, doors open, and cameras show where they're watching.
10:49 You can control the house from the 3D view or put it on a wall tablet. Feluca gives the MVE FM1 synth a whole new set of sounds. This custom firmware adds nine sound engines from analog and FM to chip tune, samples, and granular textures. You get four tracks, a 64step sequencer, effects, and a browser editor for shaping sounds and patterns. Installation runs over USB through Chrome or Edge with no extra hardware required.
11:19 Jevbox is a document library that gives your team more than folders and a search bar. Upload files, browse them in a visual finder, inspect their parsed sections, and ask questions with citations that point back to the source. It can file new documents into the right part of your library and respects each member's access permissions during search and chat.
11:41 It's a full stack starting point you can run yourself. Hairline gives websites 19 tiny isometric scenes that react to your pointer. Hover over a field of pillars and they rise around you. Move across an exploded laptop and its layers spread apart. Drop a figure into React or any page with a DOM. Then tune its colors and movement. There's also a skill that helps a coding agent draw a new figure in the same style.
12:10 Pipod lets you run PI coding agent sessions in isolated sandboxes on a server you control. Launch a pod for your current project from the terminal, then attach to the same session from the CLI or a phone app. Its server handles pod creation and connection, while a single compose setup gets the self-hosted stack running. It's for people who want their coding agent available remotely without leaving it on their laptop.
12:37 Mobile Dev for Codeex puts your iOS simulator or Android device right beside your codeex chat. You can watch the app run, tap through screens, point codecs at a specific element, and pull native and JavaScript logs into the conversation. It also records CPU, memory, and frame performance. So when a screen feels slow, you can show codeex the interaction instead of trying to describe the hitch.
13:00 Papermorph turns a PDF into an interactive web book you can explore instead of just scroll through. Give Claude code a chapter and a target audience and the skill builds a plan, storyboards, narration, animations, and quizzes before producing the web experience. The project includes a live bookshelf of examples. Right now, it's designed for English content with clawed opus 5.5.
13:26 OMASVM brings the Omari Linux desktop to an Apple silicon Mac through a virtual machine. One setup command builds it for OMAC VM's own app, UTM, VMware Fusion or Parallels. It then connects Mac features like trackpad gestures, media keys, Wi-Fi, audio, and clipboard sharing. So, the Linux desktop feels at home on the hardware. You can switch features later and run a check when something isn't working.
13:50 Cartcha turns bank transaction text into a spending ledger. You can actually check on Android. It can read messages from bank senders you approve, pull out the amount and merchant, and keep the original SMS beside each entry so you can correct mistakes. You can set budgets and view spending patterns without an account. The ledger stays on your device with optional encrypted backup if you set it up.
14:17 Jumper is an open-source crab robot you design and train by describing what you want in a sentence. One line to an AI assistant designs the robot's look as a skin file, trains a motion or builds a scene, and it implements it for you. The project includes guides for replaying and evaluating motions, then packaging a trained controller for the robot.