Top Dev Tool Projects : MarkItDown, coder, YuE, Spotifast, colibri & Worktrunk
Read the Portuguese translation →
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: Top Dev Tool Projects : MarkItDown, coder, YuE, Spotifast, colibri & Worktrunk · Channel: ManuAGI - AutoGPT Tutorials · 10:57 · Watch on YouTube
Topics: AI agents, Software engineering with AI, AI developer tools, AI content generation, Local and open-source models
Summary
Section titled “Summary”This video presents 20 open-source developer tools covering browser automation, document conversion, AI agents, local models, cloud development, code review, music generation, and more. It describes what each project does, who it may suit, and a suggested first step. Featured tools include MarkItDown for converting documents to Markdown, Coder for self-hosted cloud workspaces, Colibri for streaming large models from disk, and Worktrunk for managing Git worktrees. The video also includes a brief announcement for a new channel focused on AI agents.
Key points
Section titled “Key points”- Von’s Chrome Dev Tools lets coding agents interact with live Chrome pages, inspect network requests, and run Lighthouse audits.
- MarkItDown converts documents and other files into clean, token-efficient Markdown.
- The video introduces tools for interactive AI lessons, extending Cursor with plugins, and booting virtual iPhones on Apple Silicon.
- LLM Fit scores local models against a machine’s hardware, while Coder provisions cloud workspaces and AI coding agents on a team’s own infrastructure.
- PR Lens visualizes pull requests with animated diagrams, and Open Code Review combines deterministic engineering with agent-based judgment.
- Colibri streams routed model experts from disk, while Deep Select provides tuned GPU kernels for specific top-K workloads.
- YuE generates full songs from lyrics and style tags; the video also features an archive of chatbot system prompts.
- Worktrunk simplifies Git worktree management for parallel coding agents, and Spotifast is presented as a lightweight native Spotify client.
Chapters
Section titled “Chapters”| Start | Chapter | Summary |
|---|---|---|
| 00:00 | Introduction and channel announcement | The video introduces its weekly roundup of developer tools and announces a new channel dedicated to AI agent projects. |
| 01:02 | Browser tools, document conversion, and AI learning | The roundup begins with Chrome automation, MarkItDown, personal AI agents, interactive AI classrooms, Cursor plugins, and a virtual iPhone. |
| 04:10 | Local models, development environments, and code review | Projects include hardware-aware model selection, Coder cloud workspaces, pull-request diagrams, whiteboard animation, GPU kernels, Colibri, and Open Code Review. |
| 07:17 | Music generation, agent workflows, and sensing | The video covers YuE music generation, a system prompt archive, Oh My Hermes, and Wi-Fi-based contactless sensing. |
| 09:24 | Worktrunk, Spotifast, and bookkeeping | The final projects simplify Git worktrees, offer a lightweight Spotify client, and provide a self-hosted bookkeeping replacement. |
Quotes
Section titled “Quotes”Every week, developers release powerful open-source DevTools, and this weekly DevTool project update video brings them together in one place. — 00:00
Convert any document to clean markdown. — 01:32
Find which local models run on your hardware. — 04:10
U is an open foundation model that turns lyrics and style tags into multi-minute tracks across many genres and languages — 07:49
Make git work trees as easy as branches. — 09:24
Concepts and entities
Section titled “Concepts and entities”MarkItDown, LLM Fit, Coder, PR Lens, Colibri, Open Code Review, YuE, System Prompts Leaks, Oh My Hermes, Worktrunk, Spotifast, Slowbooks Pro 2026
Full transcript (original language)
Section titled “Full transcript (original language)”00:00 Every week, developers release powerful open-source DevTools, and this weekly DevTool project update video brings them together in one place. Top trending open-source and best DevTool projects this week cover document conversion, local model tools, code review, self-hosted platforms, and browser automation. You’ll discover useful and trending developer tools you can start using right away, explained fast and to the point. Without wasting time, let’s get started. Before we jump into today’s project updates, here’s a quick
00:30 announcement for everyone. We’ve launched a brand new YouTube 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
01:02 right, let’s get into today’s video. >> Project number one, Von’s Chrome Dev Tools for agents. Let your coding agent drive Chrome. Agents debugging web pages usually have to guess at runtime behavior. This MCP server drives a real Chrome browser via Puppeteer, letting agents click, navigate, capture performance traces, inspect network requests, and run Lighthouse audits. With a CLI available outside MCP2, it suits developers who want their agent to test and debug real web pages, add it to
01:32 your client, and inspect a live page. Project number two, mark it down. Convert any document to clean markdown. Feeding messy PDFs, slides, and spreadsheets to a language model waste tokens and structure. Mark it down from Microsoft converts PDF, Office files, images, audio, HTML, and more into clean, token efficient markdown through one CLI or Python interface with optional Azure integration for higher fidelity extraction. It suits developers preparing documents for AI systems.
02:04 Install it and convert your first file. Project number three, Airmes agent. Self-improving personal AI agent that learns you. Most AI assistants start fresh every session, forgetting what they’ve learned about you. Agent from Noose Research, creates and refineses its own skills from experience, recalls past conversations, and builds a model of you over time. Reachable via terminal or messaging apps like Telegram and Discord. It suits people who want a private agent that grows with them. Install it and start a conversation.
02:36 Project number four, Naoi AMA. Turn any topic into an AI classroom. Passive reading is a weak way to learn a new topic. Open NA turns any subject or document into an interactive lesson delivered by AI teachers and classmates, generating slides, quizzes, simulations, and whiteboard discussions from a two-stage outline then scene pipeline. It suits learners and educators who want interactive generated courses. Clone it and teach yourself something new. Project number five, cursor plugins.
03:06 Extend the cursor agent with official plugins. Extending cursor with new tools and integrations lacked a standard structure. This official repo defines the plug-in format and gathers firstparty and third-party plugins, skills, rules, MCP servers into one marketplace with a scaffolding helper for building your own. It suits cursor users who want to extend the agent or publish their own plug-in. Browse the marketplace and add a plugin. Project number six, Vphone CLI. Boot a virtual iPhone on Apple Silicon. Testing iOS
03:39 behavior usually means relying on a simulator rather than the real OS. Vphone CLI boots actual iOS in a VM on Apple Silicon using Apple’s virtualization framework, automating firmware setup, DFU restore, and boot in one command with patch levels from stock to full jailbreak. It suits iOS security researchers and developers. Create a VM and launch your virtual iPhone. Project number seven, Nichl LM Fit. Find which local models run on your hardware. With
04:10 hundreds of open models and quantizations available, it’s hard to know which will actually run on your machine. LLM Fit detects your hardware and scores every model in its catalog on memory fit, speed, quality, and context length, ranking them for your system with transparent, verifiable estimates. It suits developers running local models who want to pick the right one. Install it and see what fits your machine. Project number eight, Coder. Self-hosted cloud development environments and agents. Setting up consistent secure dev
04:41 environments by hand doesn’t scale across a team. Coder provisions cloud workspaces from Terraform templates and runs AI coding agents from a central control plane on your own infrastructure with centralized model governance, cost tracking, and audit logging. It suits platform teams, standardizing development environments and agent workflows. Install the server and provision your first workspace. Project number nine, PR lens. Draw every poll request as animated diagrams. Reviewing AI generated poll requests from a table
05:13 of findings is slow and hard to visualize. PR Lens renders animated architecture and data flow diagrams directly inside the PR showing blast radius and pipeline order as one validated graph available as a GitHub app agent skill action or CLI. It suits developers and reviewers who want changes drawn not just listed. Install the app or skill and diagram your next change. Project number 101 whiteboard animator. Turn a still image into a handdrawn reveal. Producing a whiteboard
05:44 style explainer video normally means manual animation work. Whiteboard animator takes a finished marker style image and writes text, traces outlines, and fills shapes stroke by stroke into an MP4 using only the CPU and no model at render time. It suits educators and creators making explainer videos. Install it and animate your first sketch. Project number 11, deep select fast top kernels for sparse attention. Standard TopK implementations are a bottleneck in sparse attention indexing
06:15 and token sampling. Deep select replaces them with tuned GPU kernels for these two specific workloads, reporting several times to 20 times the throughput of the built-in operation exposed through a single tensor in tensor out function. It suits engineers optimizing inference for these model architectures. Install it and speed up your top K. Project number 12. Theist Collibri stream giant Maui models from disk and C. Running trillion parameter mixture of experts models normally requires more fast memory than most hardware has.
06:47 Collibri written in pure C keeps the dense part of a model in RAM and streams routed experts from disk on demand learning and caching your hottest experts over time without changing model precision. It suits people who want to run Frontier models locally. Download the engine and run a giant model on your machine. Project number 13, open code review. Hybrid AI code review with line level precision. Pure LLM code review tends to be noisy and imprecise about where comments belong. Open code review
07:17 from Alibaba pairs deterministic engineering for file selection and comment positioning with an agent for context and judgment. Producing accurate line-level feedback while using far fewer tokens than a generalpurpose review agent. It suits teams wanting consistent, accurate, automated review. Install it and review your changes. Project number 14, U. Open Foundation model, turning lyrics into full songs. Generating full songs with vocals and accompaniment has mostly meant relying on closed services like Sunno. U is an
07:49 open foundation model that turns lyrics and style tags into multi-minute tracks across many genres and languages with incontext learning from a reference song for style transfer. It suits musicians and developers experimenting with AI music. Set your genre and lyrics and generate a song. Project number 15. System prompts leaks. Archive of AI chatbots hidden instructions. The instructions chat bots receive before your first message are normally invisible. This community-maintained repository collects system prompts from
08:20 Anthropic, OpenAI, Google, and dozens of other providers verbatim, organized by company, and updated as new models appear alongside injected reminders and agent skills. It suits researchers and prompt engineers curious how assistants are configured. Browse the collection and read a model’s hidden instructions. Project number 16. Oh My Herms operating layer adding routing and memory toms agent alone doesn’t route requests by difficulty or verify what actually ran. Oh My Herms adds a governed layer on top
08:52 scoring and routing tasks to model lanes, distinguishing planned from executed work and building a budgeted long-term memory pack across sessions. It suits her users who want disciplined model aare workflows. Install it and run your first governed workflow. Project number 17. Ru View. Turn Wi-Fi signals into contactless sensing. Monitoring a room for presence or vitals usually requires cameras or wearables. Ru reads Wi-Fi channel state information from lowcost ESP32 sensors to detect
09:24 occupancy, breathing, heart rate, falls, and sleep through walls. Running entirely on edge hardware with no cloud and native smart home integration. It suits makers and researchers exploring privacy preserving sensing. Flash a sensor and start sensing your space. Project number 18. Work trunk. Make git work trees as easy as branches. Running several coding agents in parallel means juggling gits clunky built-in work tree commands. Work trunk collapses workree management into three commands. Switch,
09:54 list, remove, address by branch name. With hooks, AI commit messages, and a one-step merge workflow for spinning up multiple agents fast. It suits developers running multiple coding agents at once. Install it and create your first work tree. Project number 19, Spotify. Lightweight native Spotify client in Rust. The official Spotify desktop app is heavy on memory and slow to start. Spotify written in Rust with no bundled browser starts in under a second and uses a fraction of the RAM
10:25 offering gapless playback, full library access, Spotify connect support, and a Win Amp style mini player with visualizers. It suits premium subscribers who want a leaner Spotify app. Install it and sign in with Spotify. Project number 20, Slowbooks Pro 2026. Self-hosted bookkeeping, a QuickBooks 2003 replacement. Old QuickBooks 2003 installs can become unusable once activation servers shut down for good. Slowbooks Pro 2026 is a from scratch self-hosted replacement
10:57 covering full double entry accounting, payroll, inventory, and tax forms with tamper evident PDFs and one command Docker deployment on Python and Postgress. It suits people wanting local control of their books. Run it with Docker and start your books. Thanks for watching. See you in the next update.