Trending Open-Source GitHub Projects : RAD Debugger, OpenRig, cua, Hindsight & mygo #300 Transcript, AI Summary & Key Points

Agent Memory Repo is an open standard and agent skill for storing and carrying AI-agent context between coding-agent sessions in a separate Git repository of Markdown notes. Agents load a short MEMORY.md entry point, follow links to notes, projects, queries, and scripts, then update and commit memories as they learn. Git provides history and merging for memory changes and flags conflicting edits when multiple agents update the same files. It includes a dreaming agent that cleans up stored notes and supports personal memory, shared team knowledge, and agent swarms. The skill can be installed for Devin CLI or through the skills installer for Claude Code, Cursor, and other agents, and requires Node.js and Git for the latter. A separate persistent local or remote repository is needed for memory across machines or cloud sessions. The specification was developed by Cognition, is open to ecosystem contributions, and is licensed under MIT.

Agent Reach is an open-source command-line capability layer for AI agents. It selects, installs, configures, checks, and routes tools that let command-line agents read and search websites and services including web pages, YouTube, RSS, GitHub, Twitter/X, Reddit, Bilibili, Facebook, Instagram, Xiaohongshu, LinkedIn, V2EX, and Xueqiu. It uses an ordered list of backends for each channel and probes them for actual availability, selecting the first usable option and reporting failures and repair guidance through `agent-reach doctor`. The repository lists integrations such as Jina Reader for web pages, yt-dlp for YouTube transcripts, `gh` for GitHub, feedparser for RSS, Exa through mcporter for web search, and alternative CLI or browser-session-based routes for sites requiring authentication. Agents invoke the upstream tools directly rather than through a data-wrapping layer. The CLI supports environment checks, explicit system installation, dry runs, updates, and uninstalling. Credentials and cookies are documented as being stored locally in `~/.agent-reach/config.yaml` with owner-only permissions; authenticated services may require user-provided cookies or browser sessions. The project is distributed under the MIT license.

bbport is an experimental native Linux port of Bloodborne for PlayStation 4 version 1.09 (CUSA03173) on x86-64 PCs. It converts the game's executable offline into a flat memory image, runs its x86-64 code directly without CPU emulation, and supplies a small runtime implementing the PS4 functions the game uses, including memory, threads, files, audio, gamepad input, saves, and application content. Its graphics layer translates the PS4 command stream and GCN shaders to Vulkan using a renderer derived from shadPS4, with a two-stage GPU command pipeline, game-specific motion-vector generation, temporal upscaling through FSR 3.1, FSR 4, or FSR 4.1.1, unlocked frame-rate patches, and an in-game settings menu. A GTK4 launcher and Steam Deck AppImage provide configuration, controller support, mod and patch selection, and resolution and upscaler controls. The project requires Linux, a Vulkan 1.3 GPU, and the user's own decrypted Bloodborne version 1.09 dump; it includes no game files and is not affiliated with Sony, FromSoftware, or AMD. The new memory and translation mode is experimental and currently requires an AMD GPU. bbport is licensed under GPL-2.0-or-later.

Bops is a Mac app developed with Orgo that provides a team of AI bots for business operations such as inboxes, pipelines, invoices, and reports. A main bot delegates work among specialized bots, which chat with users as teammates, remember workspace context, and run tasks on cloud computers with screens that users can watch or take over. Bots can connect to Gmail, Calendar, and about 1,400 other apps, as well as Slack, Telegram, and Discord; reading runs immediately, while sending, creating, and paying normally require approval unless configured otherwise. Bops can also run tasks on the user's Mac through its own Chrome environment, with optional access to local apps and files. Hosted operation uses Orgo cloud computers and Bops Cloud, while self-hosting calls providers directly with keys configured in environment variables and stores state locally. The open-source repository includes the Electron and Next.js Mac app, bot server, cloud-computer templates, hosted Bops Cloud components, and a relay for self-hosting, and is distributed under the Functional Source License. Hosted Bops uses OpenAI for chat, tasks, and voice, can provide bots with email addresses and phone numbers, and stores logins in the Mac Keychain.

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.

Cua is an open-source computer-use platform from Cua AI, Inc. that provides desktop automation, isolated cloud desktops, local virtual machines, and benchmarks for AI agents. Its Cua Fleets provision Linux desktops whose capacity is managed in pools; agents use the Sandbox SDK to run commands, capture screenshots, and interact with applications. Cua Driver exposes tools for inspecting and operating native desktop applications and browsers on macOS, Windows, and Linux through a CLI, MCP, or typed SDKs, with background delivery where supported. Lume creates and manages local macOS and Linux virtual machines on Apple Silicon using Apple's Virtualization framework. Cua Bench builds computer-use tasks, evaluates agents, and exports trajectories for training, including simulated tasks that do not require a VM, Docker, or a model API key. The project is MIT-licensed, while some third-party components have separate licenses.

FreeVideo is an open-source local inference engine and ComfyUI plugin for generating videos with the MiniMax H3 model on consumer hardware. Built on OpenVDN's eight-step VDN-H3 model and Video DeltaNet's hybrid attention, it adapts weight placement, compute precision, attention kernels and acceleration to available hardware. Weight streaming, asynchronous prefetching and chunked computation reduce peak memory use, with the project reporting operation on systems with as little as 8 GB of VRAM and 16 GB of RAM. It accepts text prompts, first and last frames, and image, video and audio references, and supports MiniMax H3 LoRAs, two-pass sampling, batch generation, four quality levels and preview-first generation. Windows and Apple silicon launchers configure ComfyUI, the runtime and models, while Linux supports setup through a shell script and command-line generation. The code is licensed under Apache License 2.0; the model weights use the MiniMax H3 Community License.

gdp-ts is a TypeScript library, linter, and coding-agent skill that applies Ghosts of Departed Proofs to authorization and entitlement checks. It makes sensitive functions require a compile-time proof that a specific user can act on a specific resource, rather than accepting an unconnected boolean permission check. The library's `name()` function gives runtime values distinct compile-time identities, trusted proof modules verify relationships between those named values, and protected functions demand proofs tied to their exact arguments; using the wrong proof, a proof for another value, a raw identifier, or no proof produces a type error. Proofs are represented by small runtime objects while most of the safety comes from TypeScript's type checker, with negligible runtime overhead and no added dependencies for the core package. Its ESLint and Oxlint presets detect practices the type checker cannot, including forged proofs and proof minting outside designated modules. The companion skill installs through `npx skills add rauchg/gdp-ts` and provides a workflow, implementation recipe, linting guidance, examples, and limitations for applying the pattern. The core package requires TypeScript 5.4 or newer and is licensed under MIT.

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.

Hindsight is an agent memory system developed by Vectorize.io for storing, retrieving, and synthesizing long-term memories for AI agents. It organizes memories into world facts, experiences, observations, and mental models within isolated memory banks, with support for multilingual data and optional per-bank scanning for secrets and personally identifiable information. Its retain operation uses an LLM to extract facts, entities, relationships, and temporal data, then normalizes them into searchable representations. Recall combines semantic vector search, BM25 keyword matching, entity/temporal/causal graph links, and time-range filtering, merging results with reciprocal-rank fusion and cross-encoder reranking. Reflect performs deeper analysis of stored memories to form connections and answer questions requiring more than retrieval. Mental models and knowledge pages provide continuously updated answers or documents derived from a bank's memories. Hindsight can run as a Docker service, Python package, embedded database, Kubernetes deployment, or hosted Hindsight Cloud service. It provides Python, Node.js, Go, CLI, and REST clients, an OpenAI-compatible LLM wrapper, integrations for agent frameworks and coding agents, and a built-in Model Context Protocol server. Self-hosted storage uses PostgreSQL with pgvector or Oracle AI Database 23ai; the project is MIT-licensed.

Knowledge Work Plugins is an open-source repository of role-specific plugins for Claude Cowork and Claude Code. Each plugin bundles domain skills, MCP connector configuration, slash commands, and, where applicable, sub-agents for workflows such as productivity, sales, customer support, product management, marketing, legal, finance, data, enterprise search, and biomedical research. Plugins use file-based Markdown and JSON rather than application code, infrastructure, or build steps; users can install them through Cowork or the Claude Code plugin marketplace and customize their connectors, company context, and workflow instructions.

Leviathan is an open-source command-line tool and optional MCP server for giving coding agents searchable memory over large datasets. Its single static binary indexes JSONL, JSON, CSV/TSV, SQLite, or database-CLI exports into one local SQLite file with an FTS5 index. Searches resolve groups and filters, rank records with BM25 and configured boosts, and return capped result cards with citations instead of exposing raw logs; agents can also resolve groups, retrieve records, inspect the index, and maintain it through upsert and delete operations. It has no runtime dependencies, does not hold database credentials, supports shell and MCP integration, and is licensed under Apache-2.0.

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.

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.

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.

OpenRig is an open-source, self-hosted multi-agent harness for coordinating AI coding agents such as Claude Code and Codex as one managed team. It uses a local daemon, CLI, terminal UI, MCP server, SQLite, tmux, and runtime adapters. Users define agent topologies in YAML with pods, seats, communication edges, and continuity policies, then use commands such as `rig up`, `rig send`, `rig broadcast`, `rig chatroom`, `rig down --snapshot`, and `rig discover`/`rig adopt` to launch, coordinate, inspect, snapshot, restore, or take over existing sessions. Its MCP tools let agents manage their own topology, while the TUI exposes topology, seat, project, feed, terminal, and system views. OpenRig requires Node.js and tmux; herdr, cmux, and Docker are optional for terminal workspaces and service-backed rigs. It is distributed through the `@openrig/cli` npm package and licensed under Apache 2.0.

RAD Debugger is a native, user-mode, multi-process graphical debugger for local-machine Windows x64 debugging with PDB debug information. Its control system provides process control, stepping, breakpoints, thread and launch management, while the debugger engine and frontend support debug-information loading, expression evaluation, disassembly, caching, and visualization. The debugger converts native debug information into the project's custom RAD Debug Info (RDI) format on demand. The accompanying radbin utility can perform these conversions and dump RDI contents as text. The repository also contains the RAD Linker, an x64 PE/COFF linker that can generate standard PDB files or RDI data, plus standalone libraries for reading and constructing RDI data. The project is in alpha and currently targets Windows x64 debugging; native Linux debugging and DWARF support are listed as future work. The source can be built on Windows or Linux x64, producing the debugger and, optionally, the RAD Linker and radbin utilities.

RemoveMacAI is a macOS debloating and privacy tool from Om Lahore, distributed as a native app and command-line program for supported Apple Silicon Macs running macOS 27. It can disable selected Apple Intelligence features, remove their models and block future downloads, turn off analytics, personalized ads and other settings, silence pop-ups and background items, and clean up selected macOS and developer files. Changes are shown for review before application and can be undone through a shared history; the CLI can also export configuration profiles for MDM deployment. Locked settings are managed through a configuration profile, while other preferences are recorded in a journal before being changed, and storage cleanup moves files to the Trash except for certain simulator and Time Machine operations. The project states that it keeps System Integrity Protection enabled, modifies nothing under /System, makes no network requests and collects no data. It is released under the MIT license.