Trending Open-Source GitHub Projects : RAD Debugger, OpenRig, cua, Hindsight & mygo #300

Author: ManuAGI - AutoGPT Tutorials · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Trending Open-Source GitHub Projects : RAD Debugger, OpenRig, cua, Hindsight & mygo #300 · Channel: ManuAGI - AutoGPT Tutorials · 12:42 · Watch on YouTube

Topics: AI agents, AI developer tools, Local and open-source models, Open source and software ecosystems

Summary

This weekly update presents 20 open-source projects for coding, AI agents, desktop apps, research, and automation, along with a brief announcement for the AI Agent Studio channel. Featured tools include RAD Debugger, OpenRig, Agent Reach, Hindsight, and cua, as well as projects for local AI, agent memory, lightweight desktop development, and authorization checks. The video outlines what each project does, who it may suit, and how to try it. It closes with GDP, a TypeScript library that uses compile-time proofs to help ensure authorization checks are present.

Key points

RAD Debugger is a native graphical debugger from Epic Games with its own debug information format and a high-speed linker.
OpenRig lets developers define and run teams of coding agents, while Agent Reach gives agents web access across multiple platforms.
Hindsight provides long-term memory for AI agents, and Agent Memory Repo stores agent memory as Markdown notes in Git.
cua bundles tools for giving AI agents access to desktops, native applications, and local virtual machines.
Several projects focus on local or self-hosted workflows, including Free Video, Open Dots, and Clear Aid.
mygo is a Go framework for lightweight desktop apps that use the system web view instead of bundling a browser.
Other featured tools include a searchable memory index, a daily model benchmark, and a library for interactive isometric visuals.
GDP uses compile-time proofs tied to function arguments to catch missing authorization checks.

Chapters

Start	Chapter	Summary
00:00	Introduction and projects 1–5	The video introduces the weekly project roundup and covers RAD Debugger, OpenRig, Agent Reach, God’s View, and Hindsight.
03:23	Projects 5–9: memory, plugins, and desktop tools	The roundup continues with Hindsight, Knowledgework Plugins, cua, Remove Mac AI, and a Linux runtime for Bloodborne.
06:10	Projects 10–13: local AI, memory, and benchmarks	The video introduces Free Video, Leviathan, Agent Memory Repo, and a benchmark for tracking model performance.
08:50	Projects 14–16: desktop apps and research agents	The featured projects are mygo, Open Dots, and Clear Aid, covering lightweight desktop apps, persistent AI coworkers, and evidence-backed research.
10:58	Projects 17–20 and closing	The final projects are OmniRush.ai, Hairline, BOPS, and GDP, ending with a tool for compile-time authorization checks.

Quotes

Every week, developers release powerful open-source GitHub projects that make coding, automation, and daily workflows faster. — 00:00

You define the team in YAML and boot it with one command. — 01:37

Agents ask in plain words and get shorts cited result cards. — 07:12

Apps ship as a single small binary written in pure Go without CGO. — 09:23

So calling them without one fails type checking. — 12:42

Concepts and entities

RAD Debugger, OpenRig, Agent Reach, Hindsight, Knowledgework Plugins, cua, Leviathan, Agent Memory Repo, mygo, Open Dots, BOPS, GDP

Full transcript (original language)

00:00 Every week, developers release powerful open-source GitHub projects that make coding, automation, and daily workflows faster. This is your weekly GitHub project update video covering top trending open-source GitHub projects this week. From agent memory tools to native desktop frameworks and computer use toolkits, you’ll discover useful and trending developer tools worth adding to your stack. Without wasting time, let’s get started. Before we jump into today’s project updates, here’s a quick announcement for everyone, we’ve launched a brand new YouTube

00:32 channel called AI Agent Studio, dedicated entirely to AI Agent projects, tutorials, and tools. So, if you’re interested in staying up to date with the latest AI agent open source projects, learning how to build your own agents, or exploring cuttingedge agent frameworks, make sure to check it out. Subscribe now to get weekly videos, in-depth guides, and realtime project breakdowns. The link is right there in the description. Don’t miss it. All right, let’s get into today’s video. Project number one, Shaad Debugger. Epic Games native graphical

01:07 debugger and tool chain. Shaarad debugger is an open-source native graphical debugger from Epic Games user mode and multiprocess. It aims to simplify debugging by unifying the debug info format. Instead of PDB or dwarf, it uses its own Shray debug info RDI format with converters from both. The project also ships the RAD linker, a high-speed drop-in replacement for MSVC’s linker. The Dubiger adds expression evaluation and built-in visualizers. It’s Windows x64 with early

01:37 Linux support and an alpha build it and debug natively. Project number two, Neon’s open rig. Build a persistent team of coding agents. Open rig is an open-source harness for a team of coding agents from Claude Code, Codeex, and Pi. A rig wraps your harnesses. You define the team in YAML and boot it with one command. You talk to a lead agent. Specialists share context and handoff work each in its terminal, and you can step into any conversation. It’s a demon, CLI, and UI on T-Max with starter teams from a GitHub link. Install it and run an agent team. Project

02:13 number three, Agent Reach. Give your agent free web access across platforms. Agent Reach is an open-source CLI that gives your AI agent eyes to read and search Twitter, Reddit, GitHub, Billybe, Shia Hong, Shu, video sites, and any web page with no API fees. Each platform has a preferred backend, plus fallbacks. So, if one breaks, it switches automatically. It sets up free search and registers a skill so the agent knows which tool to call and your cookies stay local. It works with clawed code cursor and open claw. Let your agent browse. Project number four. God’s

02:48 view. Live open-source spatial intelligence on a 3D globe. God’s view is an open-source browser-based spatial intelligence console. A spy satellite simulator with real data. on a 3D globe. It layers live aircraft, ships, satellites, wildfires, and CCTV, so you move between a global view and a single object. You ride a tracked flight, lock onto anything to see its trail, and hand off to the nearest camera. It runs locally on public data, and by design, it tracks assets and systems, not people. Build it and watch the world. Project number five,

03:23 hindsight. Long-term memory for AI agents that learns. Hindsight is open-source long-term memory for AI agents that learns for coding agents. One package builds a per repreo memory bank from your git history and past sessions, injects it as the agent starts and adds knowledge pages. SDKs and a rest API expose retain recall and reflect with light LM covering 100 plus models. It integrates with claude code, codeex, cursor, langraph and cruax. A local demon runs it all in one command.

03:55 Install it and give your agent memory. Project number six, Knowledgework Plugins. Role-based plugins that turn Claude into a team specialist. Knowledgework Plugins is an open-source collection of file-based plugins for Claude Co-work. Also compatible with Claude Code, it solves the gap between a general assistant and rosp specific work by bundling skills/comands, connectors, and sub aents for jobs like sales, legal, finance, data, and marketing. Connectors reach tools like Slack, Notion, and Jira through MCP servers. Everything is Markdown and JSON with no code or build steps.

04:32 This suits teams customizing Claude to their own tools and processes. Add the marketplace and install a plug-in to try it. Project number seven, doshua, desktops, drivers, and benchmarks for computer use agents. SUA is an open-source toolkit that gives AI agents full computers to use. It bundles Shouer spaces for desktops on your own machines. Shuer driver for operating native apps on Mac OS, Windows and Linux. Loom for local VMs on Apple silicon and shoe bench for evaluating agents. Developers connect through a CLI, MCP or Python, TypeScript and Swift

05:07 SDKs using Claude Code, Codeex or Cursor. This suits developers building and testing computer use agents. Install it and have your agent operate calculator to try it. Project number eight, remove Mac AI. Turn off Apple Intelligence and debloat your Mac. Remove Mac AI is a native Swift app and CLI that debloat Mac OS on Apple Silicon. It turns off Apple Intelligence, deletes its models, and stops analytics, ads, and pop-ups. Every change appears for review first, and a journal

05:37 makes each one reversible. Locked settings ship as a configuration profile, which admins can export for any MDM. It collects no data and makes no network requests. This suits Mac users who want privacy and disk space back. Install it and review the changes. Project number nine, Bloodborne DAP EC, a native Linux runner for one PlayStation game. Bloodborne DAP EC, also called BB port, is an experimental runtime that executes the original PS4 build of Bloodborne natively on x86-64 Linux.

06:10 It replaces the PS4 system libraries with a small runtime made for this one game and translates graphics to Vulcan with a renderer derived from Shad PS4. It adds AMD FSR upscaling and unlocked frame rates. Users must supply their own game dump. This suits Linux and Steam Deck gamers and systems programmers. Build it and run it with your own dump. Project number 10, Free Video, local AI video generation on your own GPU. Free video is a desktop launcher that runs the Miniax H3 video

06:41 model on your own GPU inside Comfy UI. It avoids cloudonly video generation by installing Comfy UI or reusing an existing folder and downloading missing models automatically. Users pick a quality level, add reference videos or audio and use community lores. It runs on Windows with an Nvidia GPU and on Apple silicon max. This suits creators who want local AI video. Download the launcher and generate your clip. Project number 11. Leviathan searchable memory that gives agents only relevant

07:12 records. Leviathan is a single binary Rust CLI that turns records from JSONL, JSON, CSV, TSV, SQLite, or any database CLI’s output into a ranked full text index. It solves the problem of agents gpping through huge raw histories. Agents ask in plain words and get shorts cited result cards. The index is one local SQLite file with no runtime dependencies. It works through shell commands and agent skill or an optional MCP server. This suits developers building agents over large data sets.

07:45 Index your data and run a search. Project number 12. Agent memory repo. An open spec that stores agent memory in git. Agent memory repo is an open specification that treats AI agent memory as a git repository of markdown notes. It solves the limits of a single memory file by giving memory history, merging and permissions. Each session clones the memory, searches it, updates entries, and commits. A separate dreaming agent merges duplicates and removes outdated entries. An installable skill

08:15 works with coding agents, and memory stays local unless you connect a private repo. This suits developers building agents that remember across sessions. Install the skill and try two sessions. Project number 13, Liveerf, a daily benchmark that checks if a model gets worse. Liveer is a Python benchmark that tests whether a Frontier model quietly gets worse after launch. It runs a frozen panel of hard questions once a day through a pinned clawed code CLI and compares scores against a launch week baseline. Built on Inspect, it uses exact graders, paired statistics, a control model,

08:50 and raw logs. It also tracks output tokens. This suits researchers and developers who want data instead of guesses. Clone the repo and review the results. Project number 14. My go small Go desktop apps with web interfaces. My Go is a Go framework for building desktop apps with a web front end. It solves the weight of bundled browsers by using the web view already on the system. WK web view on Mac OS, WebKit GTK on Linux, and WebView 2 on Windows. Apps ship as a single small binary written in pure

09:23 Go without CGO. Typed DPC binds Go services to a generated TypeScript client. It also supports menus, tray icons, dialogues, signed auto updates, and installers. This suits Go developers building lightweight desktop apps. Run the init command to scaffold your first app. Project number 15, Open Dots, a self-hosted template for always AI co-workers. Open Dots is an open- source TypeScript template for building persistent AI agents called dots that work through chat, voice

09:53 calls, and Slack. It solves the need for agents that keep context and tools across sessions. Each dot has a role, tool permissions, and its own isolated computer. A human review card pauses actions before they save or run. It is built on C-pilot Kit and AGUI, and you host it yourself. This suits developers building agent workspaces. clone it and run it locally. Project number 16, Clear Aid, a research plugin that grows a verified knowledge graph. Clear AAI is a plug-in for DeepS

10:24 harness that builds a domain ontology from your research. It solves the problem of agents declaring tasks done without proof by running a seven-stage epistemic loop. Frame, hypothesize, plan, observe, verify, evaluate, and record. Refuted hypotheses stay on record and contradictions surface for a human to decide. It installs through the plug-in manager or npm and runs locally. This suits researchers who want conclusions backed by evidence. Install it and pick clear AI in a new session. Project number 17, MIN’s Omni Rush GI, a desktop AI coding agent

10:58 with a data trade-off. Omnirush.ai is a desktop AI coding agent for Mac OS, Windows, and Linux. It lets developers chat with an agent that clones repos, branches, commits, and opens pull requests with approval prompts for write and destructive actions. It supports skills and MCP servers. Free models require an account, and in return, the app uploads chat sessions and workspace snapshots with secrets redacted. This suits developers who accept that trade. Read the privacy notes before you sign in. Project number 18, Hairline. Interactive isometric line art that reacts to your pointer.

11:35 Hairline is a dependency-free TypeScript library of six isometric line figures that respond to the pointer such as rising pillars, a card tray, and a spinning turntable. It solves the need for lightweight interactive hero visuals by drawing into any DOM element with a React wrapper, and a shy uninstall. Developers set intensity, theme, and colors through CSS variables. A companion skill lets coding agents draw new figures. Install it from npm and drop a figure into your page. Project number 19, BOPS. A Mac app with AI bots for business tasks. BOPS is a Mac

12:10 app that gives you a team of AI bots for business ops like inbox, invoices, and reports. It solves repetitive office work by running bots on cloud computers you can watch live and take over. Bots connect to about 1,400 apps, Slack, Telegram, and Discord, and ask before sending or paying. It can run on Orgo’s cloud or be self-hosted with your own keys. This suits small teams. Download it and say hi to the main bot. Project number 20. GDP tests to compile time proofs that authorization

12:42 checks actually ran. GDP tests to su is a small TypeScript library. ESLint and Oxlint llinter, an AI skill for catching authorization bugs at compile time. It solves the risk that a developer or coding agent forgets a permission check before a sensitive call. Functions demand a proof tied to their exact arguments. So calling them without one fails type checking. It works with any TypeScript codebase in runtime with almost no runtime cost. This suits teams securing APIs. Install the package and add the skill. Thanks for watching. See you in the next update.