Top Dev Tool Projects : jcode, herdr, Plane, PostHog, OpenClaw & Wake Transcript, AI Summary & Key Points

00:00 Every week, developers release powerful open-source dev tools, and this video rounds up the ones worth your time. Top trending open-source and best dev tool projects this week span coding agents, terminal runtimes, memory tools, project management, and product analytics platforms built by real developers for real workflows. Without wasting time, let's get started.

00:22 >> Before we jump into today's project updates, here's a quick 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.

00:48 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, I have ADHD. Make coding agent replies short and actionable. I have ADHD is an open-source skill for AI coding assistants that stops them from burying the answer under preamble and padding.

01:13 It reshapes how the agent writes back to you. So instead of opening with great question and walking through its reasoning before getting to the point, it leads with the next action, then lays out the rest as short-umbered steps. The skill is a set of 10 plain rules the agent follows. State the action first. Number multi-step tasks. End with one concrete next step.

01:33 Cut tangents. Restate the current state each turn. Give realtime estimates. Keep lists of five items. Report errors matterofactly and drop the recaps and closers. Despite the name, it needs no diagnosis. It just targets output that is easier to scan and act on. It ships for many assistants including Claude Code, Codeex, Cursor, Gemini, Quinn, Kimmy, and Open Code.

01:55 And you install it by pointing your agent at the repository or copying its files in. Then fork and edit the single skill file to tune the tone. It suits anyone who wants tighter agent replies. Add it and cut the fluff from your agent. Project number two, Nenge J-Code. Fast memorydriven terminal coding agent. J-Code is an open- source coding agent harness that runs in your terminal and is built for speed, low memory use, and long multi-session work.

02:24 Written in Rust, it starts almost instantly and stays lightweight even with many sessions open, which matters when you run several agents at once. Its standout feature is automatic memory. It embeds each turn as a semantic vector and quietly recalls related past context when relevant, so the agent remembers earlier work without you calling memory tools.

02:46 and it can search across previous sessions. A swarm mode lets you run multiple agents in one repository that message each other and resolve file conflicts as they edit. And agents can spawn their own helper agents to work in parallel. It works with the model subscriptions and API providers you already have, including Claude, Open AAI, Gemini, Copilot, and many OpenAI compatible and local endpoints through Alama or LM Studio.

03:11 and it can resume sessions started in other harnesses like codecs or claude code. A self-development mode lets the agent edit, build, and reload its own source. Side panels render files and diagrams in line, and a built-in browser tool adds web automation. It runs on Linux, Mac OS, Windows, and Termox. Install it and drive your models from one fast harness.

03:35 Project number three, Herd, always on terminal runtime for coding agents. Herdora is an open-source terminal runtime that hosts the AI coding agents you already run, keeping them alive and organized in one place. It works like a terminal multipplexer built for agents. Instead of launching each agent in a fragile terminal tab, you run them inside herd, which is a background server that owns their terminals.

03:59 Because the server keeps running, your agents keep working even if you close the lid, lose the network, or restart the machine. And you can reattach the same sessions from any terminal or over SSH. Every pane is labeled working, blocked, or idle. So you can see at a glance which agent has stopped and needs an answer rather than hunting through tabs.

04:18 It runs the tools you already use, including clawed code, codeex, cursor, open code, and grock without wrapping or replacing them. And agents themselves can drive herd through its command line and socket API to spawn panes and prompt each other. It supports both T-Max style keyboard control and full mouse use. Extends through a plug-in marketplace and ships as a single Rust binary with no Electron.

04:42 It runs on Mac OS, Linux, and Windows in beta. Install it and give your agents a home to run in. Project number four, Mazia's Icoding Assistant Data Extraction. Export your AI coding chat histories. AI coding assistant data extraction is an open-source Python tool that pulls your own local chat history out of AI coding assistance and saves it in one clean normalized format.

05:05 Many of these tools keep conversations in their own private local databases and if the app clears them, you lose years of history. This extractor reads that data so you can back it up, analyze it, or use it to fine-tune a model. It autodiscocovers installations across Mac OS, Linux, and Windows, so you don't tell it which system you're on. And it supports 10 sources, including Claude Code, Codeex CLI, Cursor, Windsurf, Trey, Continue, Gemini CLI, Open Code, Klein with Rue Code, and Ader.

05:37 Each conversation along with messages, code context, diffs, tool calls, and timestamps is written to JSON L with one conversation per line, and an optional merge combines everything into a single file. It reads every store safely, opening databases read only, so a running editor is never blocked and skips corrupt files instead of failing. It needs only Python and its standard library, runs from an interactive menu or flags, and is easy to extend with new sources.

06:07 It suits developers who want to own and reuse their assistant history. Run it and export your conversations. Project number five, Silva interactive 3JS procedural moss landscape. Silva is an open-source interactive 3JS landscape study that grows a procedural moss covered route through an editorial web page and makes it react to your pointer as if it were alive.

06:29 It is a browser experience and a reference for creative coding rather than a reusable library showing how much can be built from a single page. The whole thing lives in one HTML file that uses a bundled three JS build to generate the root arches, ferns, flowers, moths, and up to 250,000 instanced moss blades at runtime from seated noise. So the same landscape grows on every load with no model file or pre-rendered image downloaded.

06:56 The moss parts around your cursor and releases a short pollen trail as you move. And the interface adds a spring-driven navigation dock with proximity magnification, stage text, and image reveals, and two sandboxed WebGL2 buttons with animated liquid metal surfaces. It runs entirely from the repository with no build step and makes no network request after loading, and it includes a responsive layout and a reduced motion path.

07:23 Note the author grants no license to reuse the code or artwork. So it is for viewing and learning. Serve it locally or open the live page to explore it. Project number six, free claude code. Route Claude Code and Codeex to any provider. Free Claude Code is an open-source local proxy that lets you run Claude Code and Codeex against the model provider of your choice instead of their default paid backend.

07:47 It sits between the coding agent and the internet, accepting the exact API each client speaks. Anthropics messages format from Claude Code and OpenAI's responses format from Codeex and forwarding the traffic to whichever model you configure. So the client keeps working unchanged while you pick free paid or local models. It supports 18 provider backends including Nvidia Nim, Open Router, Google Gemini, Deepseek, Mistral, Grock, Cerebras, Fireworks, Z.AI, Kimmy, and local runtimes like LM Studio, Llama.cpp, CPP and Olama

08:20 and it can route Claude's Opus, Sonnet and Haiku tiers to different providers at once. You run a small Python server then configure everything through a local browser admin UI where you paste keys, pick models and validate them and launcher command start cla or codecs already pointed at the proxy. It also works with the VS Code and Jet Brains integrations.

08:43 handles streaming tool use and reasoning blocks and adds optional Discord and Telegram bots with voice note transcription for remote sessions. It suits developers who want flexible, lowcost model access in their existing tools. Install it and point your agent at your own provider. Project number seven, AI job search. Run your whole job search on Claude Code.

09:04 AI job search is an open-source job application framework that turns Claude Code into a full job search assistant running on your own machine. You fork the repository, fill in your profile, and drive the process through /comands, keeping all your career data local rather than on a third-party site. A setup command builds your profile from a documents folder, a pasted CV or a guided interview.

09:26 Then a scrape command searches job portals and presents matches sorted by fit with an optional rank command to batch score a short list against a structured evaluation framework. The core apply command runs a drafter reviewer workflow. It evaluates fit, drafts a tailored CV and cover letter in LTE, spawns a second agent to research the company and critique the drafts, then revises and it compiles and visually inspects each PDF so the layout is clean and the text extracts correctly for applicant tracking systems.

09:56 Never inventing skills you don't have. Further commands prep you for interviews, record outcomes, sync to Notion or Gmail, and analyze skill gaps. The search skills ship for the Danish market, but are designed to be swapped for your local boards with generators for new portals and custom templates. It suits job seekers comfortable running a command line agent.

10:17 Fork it and run your own search. Project number eight, Plane, open-source project management for teams. Plane is an open-source project management platform for tracking work, running cycles, and managing product road maps. Positioned as an alternative to tools like Jira, Linear, Monday, and ClickUp. It gives teams one place to plan and follow their work without heavy setup.

10:39 Work items are tasks you create and manage in a rich text editor that supports file uploads, subpropies, and links between related issues. Cycles track progress over a set period using burndown charts, while modules break large projects into smaller pieces. And customizable views let you filter to the issues that matter and save and share those filters.

11:00 Pages provide a rich text space with AI help for capturing ideas and turning notes into actionable items. And an analytics view shows real-time insights across all your data. You can use it through plain cloud, the hosted option, or self-host it for full control over your data with deployment methods including Docker, Kubernetes, and managed hosting.

11:20 And instance admins configure global settings through an admin mode. It is built with Django, React, Posgress, and Reddus, and released under the AGPL license. It suits teams that want a flexible project tracker they can run themselves, deploy it, and organize your team's work. Project number nine, Claude Plugins Community. Approved community plugins for Claude Co-work and Code.

11:42 Claude Plugins Community is the community plug-in marketplace for Claude Co-work and Claude Code published as a readonly mirror on GitHub. It is a curated list rather than a codebase you run. A marketplace file in the repository holds the set of community contributed plugins available to install and it is synced nightly from Anthropic's internal review pipeline.

12:04 Every plug-in listed has been submitted through the official submission page past automated security scanning and been approved for distribution. So the repository acts as a vetted directory rather than an open drop zone. You install plugins in claude co-work from the plugins page in the product or in claude code by adding this marketplace and then installing a plugin by name from the command line.

12:25 Because it is a mirror, pull requests open directly against the repository are closed automatically and all changes flow through the review pipeline. New plugins are added only by submitting them through the linked form. It sits alongside related repositories for anthropic maintain plugins and rosp specific knowledge work plugins. It suits Claude users who want to extend their setup with reviewed community tools.

12:48 Browse the marketplace and install a plugin. Project number 10, Host Hog. Open-source product analytics and developer platform. Host Hog is an open-source platform that bundles the tools a team needs to build and understand a product into one place. Rather than stitching together separate services, it gives you product analytics to track events and user behavior.

13:09 Web analytics for a dashboard view of traffic and conversion and session replays to watch real user interactions and diagnose issues. It adds feature flags for safe rollouts, experiments for measuring the statistical impact of changes, error tracking, log ingestion and no code surveys, plus a data warehouse that syncs from external tools like Stripe and HubSpot so you can query it alongside your product data and data pipelines to transform and route events.

13:38 It also captures LLM traces, latency, and cost for AI observability. And a self-driving mode turns signals in your data into researched reports and pull requests you review. You instrument it with a JavaScript snippet, one of many SDKs across front-end, mobile, and backend, or the API, and you can drive it from Slack, the web, a desktop app, or any MCP compatible agent like Claude Code or Cursor.

14:03 Use Post Hog Cloud with a free monthly tier or self-host the open-source deploy with Docker. It suits product teams and developers. Sign up or deploy it and start understanding your product. Project number 11, Open Claw, self-hosted personal AI assistant in your chats. Open Claw is an open-source personal AI assistant that runs on your own devices and meets you in the messaging apps you already use.

14:27 It is built for a single operator and keeps your data with you connecting models, tools, messaging channels, and companion apps through one local component called the gateway. The gateway is the control plane for sessions, tools, events, and channel connections. And you reach it through a web control UI, a command line interface or a terminal UI. channels bring the assistant to services like WhatsApp, Telegram, Slack, Discord, Signal, and iMessage, so you can talk to it from wherever you already chat, while companion

14:59 apps and device nodes add voice, camera, screen, and local actions on supported platforms. It works with both hosted and local model providers, and its behavior is extended through tools, skills, and plugins with new capabilities usually built on a plug-in SDK and shared through a hub. It installs on Mac OS, Linux, and Windows with a script or as an npm package, and an onboarding wizard verifies model access and sets up the gateway.

15:25 It treats inbound messages as untrusted and pairs unknown senders before acting. And it is developed in the open by a nonprofit foundation. It suits people who want a private assistant they control. Install it and bring your assistant into your own chats. Project number 12, MIA Engineering from Scratch. Build AI Endtoend through 511 lessons. AI engineering from Scratch is an open-source curriculum that teaches AI engineering by having you build each algorithm from raw math before touching a framework.

15:57 It is a structured course rather than a tool spanning 511 lessons across 20 phases that stack from linear algebra and classical machine learning up through deep learning, transformers, large language models, agents, and production systems. Every lesson follows the same loop. read the problem, derive the math, write the code, run a test, and keep a reusable artifact.

16:18 And it splits each topic into a build it step where you implement the algorithm by hand, and a use it step where you run the same thing through a production library like PyTorch. So you understand what the framework does underneath. Lessons ship in Python, TypeScript, Rust, and Julia. And each one ends with something you can reuse, a prompt, an agent skill, an agent, or an MCP server.

16:43 You can read it on the website, clone it and run the code or install its learning skills into a coding agent like Claude Code or Codeex. So, the agent acts as a tutor, gives a placement quiz and teaches one lesson per session. It also compiles into a six volume book and includes free preparation tracks for claude certifications. It suits developers who want to understand AI rather than only call APIs.

17:05 Clone it and build your first lesson. Project number 13, Free LMA PI stack free LLM tiers behind one endpoint. Free LMA PI is an open-source proxy that combines the free tiers of many LLM providers behind a single OpenAI compatible endpoint. On its own, each provider's free tier is small, but stacked together, they add up to a large pool of monthly inference capacity across dozens of providers and hundreds of model endpoints.

17:32 from fast small models to more capable ones. You run a local server, add your provider keys, which are stored encrypted, and point any OpenAI client at your unified endpoint. A router then picks the best available model for each request, falls over to the next provider when one is rate limited, and tracks per key usage, so you stay under every free tier cap.

17:55 It exposes the full range of OpenAI surfaces plus Anthropics messages format and native Gemini and Alma endpoints. So it works with coding agents like Claude code, codeex, Klein, ADER and cursor, many of which configure themselves with one command. It supports tool calling, structured outputs, image generation, texttospech, and a fusion mode that synthesizes answers from several models.

18:20 and it ships an admin dashboard, a desktop app, and a self-updating model catalog. The maintainers are clear it is for personal experimentation, not production, and that each provider's terms still apply. It suits developers prototyping against free models. Install it and route your agents through one endpoint. Project number 14, Modular Unified Platform for AI development and deployment.

18:43 Modular is the open-source home of the Modular platform, a unified system for building and deploying AI that brings together two main pieces, the Max framework and the Mojo programming language. It aims to give developers one stack for writing high performance AI code and serving models rather than stitching together separate tools. Mojo is a programming language designed for AI and systems work and the repository holds its compiler and standard library.

19:10 Max is the inference and acceleration side. It includes an accelerator library of GPU and CPU kernels, model pipelines defined as Python graphs, and an inference server that exposes an open AI compatible endpoint so you can serve a model and call it with standard clients. The project is being open- sourced piece by piece, and the repository collects the parts released so far along with code examples for both Max and Mojo and documentation for developers working inside the codebase.

19:39 Contributions are welcome to the Mojo standard library, the Max kernels and model architectures, examples, and docs. Though the Mojo compiler itself is not yet open to contributions. It is licensed under Apache 2.0 with LLVM exceptions with separate terms for max distribution. It suits developers building or serving AI models who want performance. Follow the quick start and serve your first model.

20:05 Project number 15, AI memory, long-term memory and handoff for coding agents. AI memory is an open-source tool that gives coding agent CLIs a persistent long-term memory and lets you hand off a session between different agent vendors without losing context. It solves the problem that each agent normally forgets everything between sessions and can't share what it learned with another tool.

20:29 You run it as a local server, usually in Docker, bound to loop back so nothing outside your machine can reach it, backed by a data volume, then wire your agent in a couple of commands that install its MCP server and life cycle hooks. From then on, every prompt and tool call is captured automatically. And when you open the next session in that project, the agent sees a handoff describing where you left off.

20:50 The store is a searchable markdown wiki owned by a Git repository. So every consolidation and session end produces a durable commit and you can edit memories by hand in an editor while a watcher reconciles outside changes. Search combines full text and graph-based link neighbor ranking with optional embeddings plus consolidation decay and scheduled maintenance to keep the store useful over time.

21:15 It works with agents including Claude Code, Codeex, Open Code, Cursor, and Gemini CLI with Claude Desktop and Open Claw supported as MCPon clients and it can run with or without an LLM. Written in Rust, it is still beta. Install it and keep your agents memory across sessions. Project number 16, AI Infard full stack red teaming platform for AI systems.

21:38 AI Infard is a self-hosted security scanning platform built by Tencent for testing AI infrastructure, agents, and connected tools. It runs as a Docker deployment with a web interface and scans in several ways. It fingerprints live AI services like VLM, Olama, and Comfy UI against known CVE vulnerabilities. Checks MCP servers and agent skills for risk categories using either source code or remote URLs.

22:06 evaluates AI agent workflows on platforms like DFI and co and runs jailbreak evaluations against a configured LLM endpoint. It also exposes a task creation API so scans can be triggered programmatically. This suits security teams and AI developers who need to check their own deployed AI stack for weaknesses before attackers do. Deploy it locally and scan your own AI infrastructure.

22:29 Project number 17, ESA AWS SDK for Python. Async per service AWS clients for Python. ESay AWS SDK for Python is a library from AWS that provides async clients for connecting Python applications to AWS services. Unlike BTO3, these clients are distributed per service. So developers install only the packages for the specific AWS services they use instead of one large SDK.

22:57 The design targets Python's async functionality, letting requests to AWS run without blocking the rest of an application. Code for each supported service lives in its own client package generated from AWS API models. This suits Python developers building applications that call multiple AWS services and want lighter async first clients. Explore the client packages to see which AWS services it currently supports.

23:23 Project number 18, Agent Plugins for AWS. Reusable AWS expertise packs for coding agents. Agent Plugins for AWS is a plug-in marketplace that equips AI coding agents like claude code, codecs, and cursor with AWS specific skills. Each plug-in bundles agent skills, MCP servers, hooks, and reference documentation into one package. So, an agent gets step-by-step playbooks for tasks like serverless deployment, database design, or code-based documentation instead of relying on long-pasted prompts.

23:56 MCP servers connect the agent to live AWS documentation and pricing data, while hooks add automated checks such as validating a SAM template after edits. Developers install plugins like AWS serverless, deploy on Oz, or AWS Amplify directly into their coding agent. This helps developers who build and deploy on AWS get consistent reviewed guidance baked into their agent workflow.

24:20 Add a plug-in to your coding agent and start building on AWS with built-in expertise. Project number 19, Iris, a fast screenshot CLI and MCP tool for coding agents. Iris is a command line tool and MCP server written in Rust that captures screenshots of live websites. It runs a real installed Chrome family browser over the dev tools protocol. So it renders pages exactly like a real browser waiting for fonts, images, and animations to settle before capturing.

24:52 Developers run it locally with simple flags to grab full pages, mobile viewports, or a single CSS selected element, and it can batch capture many URLs concurrently. The built-in MCP server lets coding agents like Codeex call a single capture tool and receive the image in line, giving the agent visual feedback on what it just built without hunting for a saved file.

25:11 This matters for developers using AI coding agents who need those agents to see their UI changes. Install Iris and give your coding agent eyes. Project number 20, Wake, a unified browser for all your coding agent sessions. Wake is a native Mac OS desktop app built with Rust and GPUi that gathers coding agent session history scattered across tools like Claude Code, Codeex CLI, Open Code, Kira, and Gemini CLI into a single window.

25:41 It reads each agent's local data directories readonly, then indexes sessions with a SQLite full text search engine that covers CJK text and code snippets alike, letting developers jump straight to a matched message inside a transcript. From there, a session can be resumed with one click, reopening it in terminal or iTerm at its original project directory.

26:01 It also supports starring, exporting to markdown, and deleting sessions, all stored in its own local database, separate from the original files. Wake makes zero network requests, keeping everything on the machine. This suits developers who juggle multiple AI coding agents and want one searchable home for their history. Build it from source and browse your agent sessions in one place. Thanks for watching.