Agent Clan is a desktop workspace for Claude Code and Codex agents that presents them as interactive 3D characters on top of other apps. It connects to the user's existing Claude Code and Codex accounts rather than providing its own AI service. Users can select agents with hand gestures, issue tasks by voice, or drag files, images, and text onto an agent as prompts. Agents can be organized into leader-and-member groups, and each character indicates whether it is working, finished, or waiting for approval. The workspace also supports mouse and keyboard input, multiple terminals, customizable character appearances, and agent-specific results and permission requests without switching applications. Gesture recognition and speech-to-text run locally on the computer, while agent tasks require an internet connection and are handled by the selected provider. The page describes versions for Mac and Windows; the free tier supports one project and up to two agents, while the Pro version adds unlimited projects and agents and a mod system for changing the workspace through Claude agents.
Arcjet is an AI agent runtime security platform that discovers agents running in an organization and enforces policy across agent actions, prompts, and tool calls. Its protections detect prompt injection, authorize or block tool calls, redact personally identifiable information, control budgets and abuse, and retain evidence of security decisions for auditing. Arcjet provides guards, a plugin, skills, an MCP server, a CLI, SDKs, and integrations with web frameworks and AI agent frameworks.
AXL is an AI-native sales and marketing platform that combines website and funnel building, CRM and sales pipelines, marketing automation, booking, checkout, and customer-history records. It is designed for founder-led businesses, startups, course creators, consultants, communities, and local service businesses. Users can work in the platform's interface or give tasks to its built-in AI assistant. AXL also supports connecting ChatGPT, Codex, or Claude through MCP. Its AI can use recorded visits, purchases, viewing activity, form answers, messages, and other customer events to build pages, create segments and campaigns, configure follow-up workflows, and support sales activity. Additional features include AI chat agents with human handover, configurable inbound and outbound voice calls, browser-based business telephony, appointments, subscriptions, webinars, courses, protected video hosting, communities, forms, quizzes, and analytics dashboards.
OpenAI's image-generation and editing system, built into ChatGPT, Codex, and the API. It creates images from prompts or photos, supports more controlled editing across rounds, can turn prompts or photos into animated GIFs, allows developers to mark regions for editing, and exports SVG files.
ClipsPal is a desktop AI video editor for macOS and Windows that processes talking-head and faceless videos. It imports or records footage, cuts failed takes and silences, adds subtitles, zoom-ins, visuals, sounds and motion slides, and can generate shorts, thumbnails, chapters and articles. Its Director can apply edits from plain-language requests, while the timeline editor allows each cut to be restored or undone and exports MP4 files with subtitles or timelines for Premiere Pro and DaVinci Resolve. ClipsPal includes an MCP server with tools for reading videos, cutting footage, generating shorts, chapters and articles, and exporting results through compatible AI clients. Video, audio and transcription remain on the computer; only the transcript is sent to a selected remote AI, while a local model can keep processing on-device. It supports 40 spoken languages and is distributed as a one-time-purchase desktop app for macOS 14 or later and Windows 11, with a Windows beta.
Edits for Desktop is a free macOS video-editing app from Instagram for creators. It provides a multi-track timeline, keyframes, audio waveforms for beat-accurate cuts, color wheels and curves, layered audio, multiple aspect ratios, and automatic captions. It supports Instagram's licensed and royalty-free audio libraries, imported LUTs and font files, and direct publishing to Instagram. Videos can also be exported in 4K without a watermark. The app requires a Mac with Apple silicon (M1 or later) and an Instagram account; it is not available for Windows.
Egoist AI Passport is a personal context layer for carrying preferences, history, identity, files, notes, and voice memos across connected AI assistants and apps. It lets users review personal details and context before sharing them, choose which data each app can access and for how long, and revoke access with one click. The service also provides audit receipts showing what connected apps used.
Higgsfield Miracle is an AI video-editing tool from Higgsfield that applies described changes to a clip while it plays. It can recolor, move, remove, or restyle an individual object; alter lighting, time of day, shadows, or texture; and edit live text or layers while keeping the rest of the shot, including the face, camera movement, and original action, unchanged. The page describes integrations as a ChatGPT plugin and an MCP for Claude. Miracle is coming soon and offers a waitlist and early MCP access.
Inkling 2 Small is an AI model from Thinking Machines Lab for long-horizon autonomous work, including agentic coding, knowledge work, reasoning, and image-based tasks. It is a mixture-of-experts model with 594B total parameters and 17B active parameters, supports image inputs and variable effort, and can call tools through MCP.
KIKUmap is a free, open-source native macOS menu-bar app and desktop widget for monitoring Claude, ChatGPT, and Claude Code sessions. It reads pinned chats and sessions, shows whether they are running, finished, waiting for approval or an answer, or stopped with an error, and lays them out as a draggable mind map organized by topics, colors, shapes, and tags. It also provides keyboard navigation, search, timelines for past and scheduled work, chat export and import for Claude conversations, PDF map export, backups, usage and time statistics, and optional notifications. The app supports macOS 13 or later on Apple silicon and Intel, uses Swift AppKit without Electron or a web view, and stores its data locally without requiring an account or tracking users.
MCP1 by Appwrite is an MCP gateway for teams that puts Model Context Protocol servers behind a single URL. Administrators install servers from a catalog or remote server URL, keep server tokens sealed in MCP1 rather than on users’ laptops, and share gateways with groups. Group and per-tool policies can grant full or read-only access, restrict destructive tools to administrators, apply rate limits, and control which servers and tools members can use. MCP1 supports clients including Claude, Cursor, ChatGPT, and Codex, and records tool calls, sign-ins, policy decisions, failures, latency, and other workspace activity in an audit log with CSV export. Its context controls expose tools and server instructions as needed rather than loading every available tool up front.
Muse Realtime Avatar is Meta’s real-time avatar system, which turns Muse Realtime Voice into expressive, interactive characters conditioned on reference media. It can animate portraits, illustrations, animals, and everyday objects with facial, hand, and full-body movement while preserving their appearance and mannerisms across conversational turns. The system uses the same speech-token stream for audio and visual generation, keeping spoken content, lip motion, and expression synchronized. Its audio-driven Diffusion Transformer generates video in short causal chunks, feeding newly generated video latents forward as motion context for subsequent chunks. Meta describes a distillation process that converts a bidirectional teacher using guided diffusion into a two-step causal student, and an inference stack using persistent KV caches, dynamic batching, quantization-aware training, fused kernels, and CUDA Graph capture for subsecond interactive latency. Generated video includes an invisible Meta Video Seal watermark for traceability.
NoMac is an on-demand cloud macOS service for AI agents. It provides a real, disposable Mac with administrator access, Xcode, Homebrew, Node, Python and CocoaPods, allowing agents to build, test, sign and publish Apple apps without local Mac hardware. Agents control sessions through MCP, CLI, REST API or SSH. A session can run Xcode builds, use the iOS Simulator, access the Mac desktop through a browser, upload builds to TestFlight and submit apps for App Review. Sessions are isolated and deleted when stopped, while an encrypted persistent drive retains projects, packages and build caches between sessions; secrets can be stored for reuse without passing through chat. NoMac also includes a cloud iOS Simulator for faster agent-driven screen reads, taps and recordings. The service uses pay-as-you-go pricing of $0.90 per hour, billed by the minute, with no subscription.
outages.live is a free live outage-monitoring feed for coding agents. It periodically synchronizes official status pages from cloud, AI, and developer-service providers into one feed, with filtering for time range, state, impact, and maintenance notices. The feed is available on the website, through RSS and a JSON API, or to agents through an MCP server; the videos describe these interfaces as requiring no login or API key. It is intended to help agents distinguish failures in their own code from provider outages.
ProtoMap is a self-hosted context engine and model harness from HoundDog.ai for gRPC and Protocol Buffers. It scans source repositories, identifies services, methods, messages, fields and their relationships, and builds a graph that coding agents can query through an MCP server. It supports monorepos and multi-repository codebases and can be used with agents including Claude Code, Codex, Antigravity, VS Code and OpenCode, as well as other MCP clients. The scanner supports Go, Java, Kotlin, C#, Python, TypeScript, JavaScript, Rust and Swift. ProtoMap runs on-premises with Docker and PostgreSQL, supports bring-your-own AI models, and is designed to keep source code in the user's environment.
Sage2 is a runtime API from Levanto for software that acts on human language. Its extraction API converts defined fields in natural-language messages into typed values such as text, numbers, money, dates, times, and durations, while linking each value to its source text and returning a probability. Its streaming API maintains accumulated conversation context, accepts appended text, and updates answers to developer-defined questions as the conversation changes; sessions support up to 32 questions and 16,384 tokens of context and questions. Extraction and streaming decisions can be combined in applications such as support workflows, document processing, offer selection, and agent tool selection. The service also supports reasoning, web grounding, and vision in beta, and is available through the Sage API, playground, and documentation.
TRACE is an open weekly benchmark of web-scraping APIs used to retrieve pages for AI agents. It tests context.dev, Exa, Firecrawl, and Keenable on a fresh sample of roughly 2,500 web pages each week, measuring page-access rates, latency percentiles, returned-content quality, cost per page, and failure reasons. The sample pipeline discovers URLs from publisher feeds, job boards, regulatory listings, filings, legislation, research sources, sitemaps, Wikipedia edits, and other public sources; freezes the sample with caching disabled; and uses a blind judge to evaluate provider responses. It publishes results by source and category, confidence intervals, page-level data, sample history, methodology, and source code, retaining prior grades when scoring rules change.
Watchflows is a visual workflow-automation app for macOS that runs flows locally to automate repetitive tasks involving files, folders, apps, and other Mac functions. Its Flow Builder turns a plain-language description into a wired canvas flow, explains the nodes, and lets users edit the result manually. Flows can use triggers such as folder changes, schedules, hotkeys, clipboard events, app launches, webhooks, and Apple events, followed by actions including renaming, moving, reading PDFs, running AppleScript, making HTTP requests, and invoking AI nodes with structured output, agents, or tool calls. The app supports Apple Intelligence, Ollama, LM Studio, Claude, ChatGPT, OpenAI, Anthropic, Gemini, and OpenRouter. Files, flow runs, and history remain on the Mac; credentials are stored in the macOS keychain, and the app can work offline without an account. Watchflows is distributed as a $49 one-time purchase with a 14-day trial, and is also available through Setapp for macOS 14 or later.
Searchable transcript of Top AI Agent Projects : Arcjet, ProtoMap, MCP1, Voyager & Watchflows — ManuAGI - AutoGPT Tutorials (12:08). 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 ManuAGI - AutoGPT Tutorials. 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 Everyday developers build new AI tools and it's hard to keep up. This is our weekly project update video. Top trending AI agent projects. This week you'll discover trending AI tools for security automation, local workflows, and coding agents. Without wasting time, let's get started. >> Before we jump into today's project updates, here's a quick announcement for everyone.
00:22 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 real time project breakdowns.
00:48 The link is right there in the description. Don't miss it. All right, let's get into today's video. >> Project number one, Arkjet. Runtime security for AI agents and LLMs. Arkjet is a runtime security platform for AI agents. It sits inside your code and checks each agent action before it runs, then returns allow, block, redact, or hold for review. It detects prompt injection, redacts personal data, limits tool calls, and caps token spend.
01:16 It finds coding agents like clawed code, codecs, and cursor on developer machines. Decisions are logged as audit evidence. Security and engineering teams control what agents can do. Start with one agent. Project number two, Nissan Proto Map, gRPC and Protobuff context for AI agents. Proto MAPAP is a context engine for AI coding agents that work on gRPC and Protobuff code.
01:40 It scans your repos, builds a graph of every service, method, message, and field, and updates it with each merge. Agents like claude code and codeex query that graph through MCP instead of searching files again. Self-hosted, it keeps your source code inside your environment. Engineers get safer API changes with fewer tokens. Start free on GitHub. Project number three, Higsfield Miracle.
02:05 Edit any video clip while it plays. Higsfield Miracle is an AI video editing tool that changes a clip while it plays. It is not a new generation model. You type a change like recoloring an object or shifting the light and only that part changes. Face and camera move stay as shot. Miracle is an early testing. Higsfield offers an MCP for Claude and a plug-in for chat.
02:29 GPT creators and marketers keep the take they picked. Join the wait list. Project number four, the MCP1. one URL for every MCP server. MCP-1 is an MCP gateway for teams that puts every MCP server behind one URL. Admins install servers from a catalog or by URL and tokens stay sealed in MCP-1, never on laptops. Groups decide who can call which tools with readonly access, destructive tool limits, and rate limits.
02:58 Claude, Cursor, Chat, GPT, and Codeex connect. Once every call is recorded in an audit log. Teams give agents safe access to company tools. Start free. Project number five. Voyager. Agentic file browser for your Mac. Voyager is an agentic file browser for Mac OS that puts AI agents inside your folders. You keep tabs and save searches, select files, and ask an agent to compare them.
03:25 You bring the agents you already use and pick one per task. Your files stay on your Mac and you review the context an agent receives first. Knowledge workers handle files faster. Download Voyager Free Project. Number six, Egoist AI Passport. One private memory layer for every AI. Egoist AI Passport is a personal context layer that carries your preferences, history, and identity across the AI apps you use.
03:51 It solves the problem of scattered, uncontrolled personal data. You connect your AI assistants and apps, add files, notes, or voice memos, and review new details before they are shared. You choose which apps see what and for how long. Access can end in one click. It is for anyone who wants private portable AI memory. Claim your passport today. Project number seven, Clips Pal.
04:14 AI video editor that cuts raw clips for you. Clips Pal is an AI video editing agent for Mac and Windows that turns raw talking head clips into finished videos. It cuts failed takes and silences, adds subtitles, zoomins, and visuals, then makes shorts and articles. You can ask its director in plain words, or let agents like Claude Code and Codeex connect through MCP.
04:38 Video and transcription stay on your computer. It suits creators who want faster edits. Download it and try one video. Project number eight, Gennel Clarify, the autonomous CRM that runs your sales busy work. Clarify is an AI native CRM with agents that handle the work around every sale. It removes manual CRM upkeep by recording calls, updating records, watching buying signals, enriching prospects, and drafting follow-ups.
05:03 It syncs with tools like HubSpot, Salesforce, Slack, and Google. You can also use it from Claude, ChatgPT, or Cursor through its MCP server. It is built for fastmoving sales teams that want more selling time. Start for free. Project number nine, Watch Flows. Describe a chore, your Mac automates it. Watch Flows is a visual workflow automation app for Mac OS with built-in AI nodes.
05:28 It handles repetitive Mac chores that need several steps or a decision. You describe a task in plain words and its flow builder drafts the flow on a canvas. Flows run locally so files stay on your Mac. It works with Apple Intelligence, Alama, Claude, ChatgPT, and more. It suits Mac users who want no code automation. Download the free trial. Project number 10.
05:50 Kiku Map, a mind map for your AI chats. Kiku Map is a free open-source Mac OS menu bar app that tracks your AI chat sessions. It shows which claude chatgpted code and CEX chats are running, done, or waiting for you. Pin chats become a mind map you can organize with topics and tags. It stays on your Mac with no account or tracking. It suits people juggling many AI chats.
06:16 Download it and map your chats. Project number 11, DXL. AI sales platform you run through chat. DXL is an AI native sales and marketing platform that brings websites, CRM, booking, and payments into one system. It cuts the manual work between separate tools. You give tasks to its built-in assistant or connect chat GPT, Codeex, or Claude through MCP within permissions you set.
06:42 The AI uses recorded customer history to build pages, run campaigns, and follow up with leads. It suits founder-led businesses. Start on the free plan. Project number 12, Sage 2 by Levanto Labs. Turn customer words into live decisions. Sage 2 is a runtime model API from Levanto Labs that helps software act on human language. It turns messages into typed fields like money, dates, and durations, each linked to its source text.
07:09 Its streaming decisions update answers as a conversation changes. Your language model replies. Sage supplies fields and decisions and your code runs the next step. It also supports reasoning and web grounding. It suits developers building agents. Try the playground project number 13. Inkling too small. Agentic model for coding, vision, and knowledge work.
07:33 Inkling too small is an AI model from thinking machines lab built for longrunning agentic work. It handles coding, knowledge work, and image inputs, and it calls tools over MCP. The mixture of experts design uses 17B active parameters out of 594B total with adjustable effort levels. You can try it in the Tinker playground, then sample or fine-tune it on Tinker.
07:55 It suits developers customizing agents. Try it on Tinker. Project number 14, Nom Mac, a disposable cloud Mac for your agent. Nomac is an ondemand cloud Mac OS service for AI agents. It solves the need for a Mac when building Apple apps. Your agent starts a real Mac in about a minute. Builds and tests with Xcode, then deletes it. Claude, Chat, GPT, Codeex, and Cursor connect over MCP, CLI, API, or SSH.
08:23 Each session runs in isolation with an encrypted drive kept between sessions. It suits developers without a Mac. Get your agent a Mac. Project number 15, match outages.live, a free live outage feed for agents. Outages.live is a free outage monitoring feed built for coding agents. It solves the guesswork of whether a failure comes from your code or a provider.
08:47 It merges official status pages from cloud AI and developer services every few minutes. You can read it live via RSS, a JSON API or an MCP server with no login or key. Claude code, codeex, cursor, and copilot are supported. It suits developers using thirdparty services. Add it to your agent. Project number 16, man agent clan. See and control your coding agents by hand.
09:14 Agent Clan is a desktop workspace that turns clawed code and codeex agents into interactive 3D characters on your screen. It solves the problem of tracking many agents across apps. Each agent sits in the corner with a ring showing working, done, or waiting. You pick agents by hand gesture, give tasks by voice, or drag files onto them. Gesture recognition and speech to text run on your computer.
09:36 It suits developers running many agents. Download it free. Project number 17, Chat GPT images 2.6. Faster images with more control over edits. Chat GPT images is OpenAI's image generation and editing system built into Chat GPT codecs and the API. It makes image creation faster and gives more control over each editing round. You describe an image, then refine it.
10:03 It can turn prompts or photos into animated GIFs like a mascot or product loop. Developers can mark regions to edit and export SVG. It suits creators and product teams. Try it in chat. GPT project number 18. Muse Realtime Avatar. Live expressive avatars driven by voice. Muse Realtime Avatar is a real-time avatar system that turns MW's real-time voice into interactive characters.
10:29 It solves the problem of talking AI that has no face or body. It uses the same speech token stream as the voice model, so lip motion and expression stay synchronized. It animates portraits, illustrations, animals, and objects from reference media with video responses in under a second. This matters for people building embodied AI. Try Muse and meet an avatar live project number 19.
10:52 Instagram edits for desktop. Pro video editing for creators on Mac. Instagram edits for desktop is a Mac video editing app built for creators. It gives more room and power for long complex projects than a phone. It offers a multi-track timeline, key frames, color wheels, curves, and unlimited automatic captions. Creators can use Instagram's audio libraries and import their own LUTs and fonts.
11:16 Videos published directly to Instagram or export in 4K without a watermark. It needs a Mac with Apple silicon. Download edits and start editing. Project number 20. Trace by context.dev. Weekly open benchmark for AI agent web scrapers. Tracy is an open benchmark that tests how reliably web scraping APIs read pages for AI agents. It replaces vendor claims with measured results.
11:41 Each week it sends about 2500 fresh pages to context deva firecrawl and conneal with caching off. It reports access rate, speed and cost per loaded page. Data and code are public. context.dev publishes it and is also tested. Developers get evidence for choosing a scraper. Open the leaderboard and check the results. Thanks for watching. See you in the next