A collaborative canvas lets several coding agents work in one repository while file ownership prevents agents from overwriting one another.
Behind this: 7 build steps · 2 tools and how each is used · how to validate demand · 3 things the video never answers.
A desktop application indexes local AI coding chats, embeds them on the device, and lets developers search past conversations by meaning and reopen sessions with context.
Behind this: 7 build steps · 2 tools and how each is used · how to validate demand · 3 things the video never answers.
An iPhone app uses the camera to track full-body movement, count repetitions, provide real-time form corrections, and adapt future workouts to observed performance.
Behind this: 7 build steps · 1 tool and how each is used · how to validate demand · 3 things the video never answers.
An open-source terminal layer lets a teammate join a live Codex or Claude Code conversation from their own terminal while preserving each participant's identity in the transcript.
Behind this: 8 build steps · 3 tools and how each is used · how to validate demand · 3 things the video never answers.
A terminal game turns waiting time during Claude Code sessions into a local game in which prompts advance steps and players battle, catch, and level up creatures.
Behind this: 8 build steps · 2 tools and how each is used · how to validate demand · 3 things the video never answers.
An AI-native CRM builds a living customer model from calls, emails, and meetings, then uses agents to generate pipeline views, meeting preparation, and stale-deal recovery.
Behind this: 9 build steps · 2 tools and how each is used · how to validate demand · 3 things the video never answers.
A tool turns a short product brief into mascot concepts and a complete gesture set that can be downloaded as animated SVGs and added directly to an application.
Behind this: 8 build steps · 2 tools and how each is used · how to validate demand · 3 things the video never answers.
An AI agent researches a market, creates and audits Google Ads campaigns, connects performance to CRM priorities, and applies only human-approved fixes.
Behind this: 10 build steps · 3 tools and how each is used · how to validate demand · 3 things the video never answers.
A managed service launches persistent agents in sandboxed cloud environments with continuous history, snapshots, recovery, connectors, and access across messaging and developer channels.
Behind this: 9 build steps · 3 tools and how each is used · how to validate demand · 3 things the video never answers.
A voice-to-text tool transcribes natural speech, removes filler words and typos, formats the result at the cursor, and synchronizes personal language settings across devices.
Behind this: 9 build steps · 2 tools and how each is used · how to validate demand · 3 things the video never answers.
A personal AI extension represents a user through its own phone number, inbox, and computer, handling calls, email, waiting, scheduling, and approved device tasks while escalating decisions to the user.
Behind this: 10 build steps · 2 tools and how each is used · how to validate demand · 3 things the video never answers.
A payment skill lets AI builders generate a payment link and add a paywall without implementing the usual merchant onboarding, KYC, or payment infrastructure themselves.
Behind this: 6 build steps · 1 tool and how each is used · how to validate demand · 3 things the video never answers.
A Windows application arranges several coding-agent sessions in one window, alerts users when approval is needed, and lets them search and resume saved sessions.
Behind this: 9 build steps · 3 tools and how each is used · how to validate demand · 3 things the video never answers.
An India-first platform lets businesses build voice and chat agents that handle inbound and outbound calls, switch languages mid-call, use uploaded knowledge, and connect to a CRM and chat widget.
Behind this: 11 build steps · 2 tools and how each is used · how to validate demand · 3 things the video never answers.
An email assistant scans sent conversations for genuine open loops, filters out completed and out-of-office threads, and creates editable follow-up drafts without sending automatically.
Behind this: 9 build steps · 2 tools and how each is used · how to validate demand · 3 things the video never answers.
A Mac binary attaches to an existing coding-agent session and lets an iPhone user read its transcript, receive completion or approval notifications, and approve work remotely.
Behind this: 9 build steps · 3 tools and how each is used · how to validate demand · 3 things the video never answers.
An open-weight general-purpose model accepts relationships among text, images, video, and audio in one prompt and generates video with native stereo sound.
Behind this: 8 build steps · 1 tool and how each is used · how to validate demand · 3 things the video never answers.
A natural-language data platform searches for prospects, enriches and verifies contact information, and returns CRM-ready lists using a waterfall of more than 15 providers.
Behind this: 10 build steps · 3 tools and how each is used · how to validate demand · 3 things the video never answers.
A read-only data platform connects vendor administration consoles, normalizes seats, usage, and spend into one schema, and produces measured, derived, or modeled reports for leadership.
Behind this: 10 build steps · 2 tools and how each is used · how to validate demand · 3 things the video never answers.
A Windows system-tray application analyzes faces locally during video calls and alerts users when a face appears synthetic, without recording or transmitting frames.
Behind this: 9 build steps · 2 tools and how each is used · how to validate demand · 3 things the video never answers.
Searchable transcript of Top AI Agent Projects : Murmell, Lightfield, AgentSky, Wispr Flow & Halo — ManuAGI - AutoGPT Tutorials (16:22). 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 keeping up with all of them takes time. This is our weekly project update video where we cover the top trending AI agent projects this week so you can find what is worth using in one place. You will discover trending AI tools and projects across coding, voice, and automation. Without wasting time, let's get started.
00:21 >> 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 cutting-edge agent frameworks, make sure to check it out.
00:47 Subscribe now to get weekly videos, in-depth guides, and real-time 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, Murmel, a shared canvas for coding agents. Murmel is a browser canvas where several AI coding agents run in one repository at once. It solves parallel agents overriding each other's files.
01:10 Each agent claims a file before writing, so one holder per path means nothing forks or merges later. It runs Claude code, Codex, Kimmy, and open code on cloud machines using your own accounts, pushing to a private repo you own while your team watches live by link. It is built for developers running many agents together. Open a canvas and bring your agents.
01:32 Project number two, Inventory, private local search for AI coding chats. Inventory is a desktop app that indexes every AI coding conversation on your machine into one searchable timeline. It solves the lost fix buried in a chat you had weeks ago. It reads the local stores that Cursor, Claude code, Zed, Codex, Kiro, and Antigravity already write to, read only, and embeds each thread on device.
01:57 So, you search by meaning, not just keywords. Claude Code and Code X sessions reopen in the terminal with full context. Everything stays on your Mac with no account or cloud. It is built for developers who reuse past solutions. Index your history for a one-time price. Project number three, Coach AI, camera-based personal trainer on your iPhone. Coach AI is an iPhone fitness app that turns your camera into a personal trainer's eyes.
02:24 It solves the cost of a coach and the guesswork of training with bad form alone. You prop your phone against anything and step back, and it reads your whole body on device, counting every rep so you never tap or guess, and calling out form fixes in real time as you move. Each session feeds your plan, so the next workout fits how you actually moved. It needs no equipment.
02:46 It is built for anyone training at home or the gym. Download it and start your first workout. Project number four, Dansh MPI. Make coding agent sessions multiplayer. MPI is an open-source terminal native layer that lets a teammate join your live AI coding session. It solves the lonely context problem where useful work sits inside one person's Codex or Claude Code conversation.
03:09 The host shares one explicit session by named invite, and the teammate enters that exact conversation from their own terminal, prompting as themselves with their name kept in the transcript. It runs over your private Tailscale network, so your machine stays the host with no public relay or transcript cloud, and sharing is opt-in per session. It is built for small teams pairing on agent work.
03:32 Install it on two Macs and open a shared session. Project number five, ClaudeMon, a monster-catching game inside Claude Code. ClaudeMon is an open-source terminal game that plays alongside your Claude Code sessions. It turns the dead time while an agent works into something to do. Your prompts count as steps, so the longer you prompt and the longer Claude runs, the more chances a wild creature appears.
03:55 The status line flags it, and you switch to a second terminal tab to battle, catch, and level up across the original 151. It runs fully local, with no account or back end. Installed as a plug-in from inside Claude code. It is built for developers who wait on agents. Install the plug-in, and step into the grass. Project number six, Mons Lightfield, AI-native CRM that does the work.
04:18 Lightfield is an AI-native CRM that does the admin, instead of asking you to. It solves the logging, field updates, and follow-up drafting that legacy CRMs pile on sellers. It builds a living version model of every customer from your calls, emails, and meetings. Then agents work off that full context to generate pipeline, prep you for meetings, and revive stale deals.
04:40 You ask questions in plain language, and get answers cited to the original conversations. And you build your own agents from skills, knowledge, and automations. Connected over APIs or MCP. It is built for early-stage, high-growth teams. Try it free, and let the agents work. Project number seven, Mascot AI, generate animated mascots for your app. Mascot AI is an AI tool that generates animated SVG mascots for web and mobile apps.
05:06 It solves the fact that most products settle for a static logo, or a generic AI blob, instead of a real character. You write a short brief describing what you are building, and how it should feel. And three directions come back in seconds. You pick one, and it builds the full gesture set around it. You then download animated SVGs, and paste them straight into your code base, tweaking gestures later.
05:30 Every plan includes six models from OpenAI and Anthropic, with token costs shown before you create. It is built for teams shipping an app that wants a personality. Write a brief, and give your app a face. Project number eight, Airtop for Google Ads automation. Run Google Ads from a chat. Airtop is an AI agent that builds, audits, and optimizes Google Ads campaigns from a conversation.
05:53 It solves the hours of expert work that keeping campaigns profitable normally takes. You give it your company URL, budget, and goal, and it researches your market, writes ads, and sets up structured campaigns ready for your sign-off. It reads your CRM and priorities, so bids follow real pipeline, audits your search terms and targeting to find leaking budget, and implements fixes directly in Google Ads.
06:17 Only human-approved changes go live, and each session runs in its own encrypted environment. It is built for marketers without PPC expertise. Try it free and describe your goal. Project number nine, Agent Sky. Cloud-hosted long-horizon agents on demand. Agent Sky is a managed agent as a service that launches a long-lived AI agent in one click. It solves the setup and ops burden of running an always-on agent yourself, with no Mac mini or infrastructure.
06:43 You pick like Claude Code, Codex, Hermes, or Open Claw. Choose a model, add capabilities, and it runs in a managed sandbox that keeps an unlimited continuous history, with snapshots and recovery handled for you. The same agent answers on WhatsApp, iMessage, Telegram, Slack, web, and the command line, and you can clone a local agent to the cloud or swap models mid-task without losing the thread.
07:09 Built-in tools cover scraping, images, email, and thousands more via connectors. It is built for developers and teams. Launch your first agent in one click. Project number 10, WhisperFlow. Effortless voice dictation in every app. WhisperFlow is a voice-to-text tool that turns speech into clear, polished writing across every app. It solves the slowness of typing and the mess ordinary dictation leaves behind.
07:34 You speak naturally, and Flow transcribes and auto-edits on the fly, dropping formatted text at your cursor with filler words and typos removed. It learns your names and terms in a personal dictionary, turns repeated phrases into voice snippets, and detects over 100 languages. It runs natively on Mac, Windows, iPhone, and Android, syncing your style and settings everywhere, and works in tools like Cursor, VS Code, Slack, and email.
07:59 It is built for anyone who writes at a keyboard all day. Download it free and start talking. Project number 11, Zinly. A personal AI extension of you. Zinly is a personal AI extension that represents you with its own phone number, inbox, and computer. It solves the calls, email, and busy work that pile up when you cannot be everywhere. It remembers you and the people in your life, so you never re-explain the backstory, and it acts within your rules, bringing you in when a decision is yours.
08:29 It takes and makes calls, waits on hold, and books meetings when you CC its inbox, then reports back in plain language with a receipt. It works on its own computer and only the devices you approve, alongside apps like Gmail, Slack, Calendar, and Notion. It is built for busy people who want a second self. Claim your Zinly and start free. Project number 12, Uniweb Pay Skill.
08:52 Accept payments the moment you ship. Uniweb Pay Skill is a payment skill for AI builders that turns accepting money into a prompt. It solves the slowest part of shipping an AI product, where merchant onboarding, KYC, and payment integrations stretch a launch to days. You install the skill, generate a payment link, and start accepting payments in minutes without the usual setup or infrastructure overhead.
09:16 It supports global payment methods, so you can add a paywall to a new feature and quickly test whether people actually pay. It is built for builders turning AI products into businesses. Install the skill and generate your first payment link. Project number 13, Termuxo. One window for every coding agent. Termuxo is a local Windows workbench that runs your coding agents side by side in one window.
09:41 It solves the mess of losing track of which terminal is doing what and setting it all up again the next day. You open Claude Code, CodeX, and more in a grid you arrange and it flags the moment an agent is waiting on approval by flashing the taskbar and holding a banner across every project at once. It lists your saved sessions so you can search and resume through the CLI's own commands with full context and it points the same CLI at Deep Seek, Mini Max, or GLM without reinstalling.
10:10 Everything stays on your machine with keys in Windows credential manager and no account. It is built for developers juggling several agents. Run it with one command and keep your agents in view. Project number 14, Bolsho AI, voice AI agents for Indian languages. Bolsho is an India-first voice AI platform for building agents that handle phone calls and website chat in genuine Indian languages.
10:34 It solves the gap left by English-first tools that markup minutes and reply awkwardly to Hindi callers. You build an agent in the dashboard or by describing it aloud, connect a number, and go live with a language director that switches mid-call the instant a caller does across Hindi, Tamil, Telugu, Bengali, and more. It handles inbound and outbound calls, grounds answers in your uploaded knowledge, and ships a CRM plus a one-line chat widget.
11:00 It runs on switchable providers for speech, brain, and voice with automatic failover. Build near provider cost. It is built for Indian businesses. Sign up free and build your first agent. Project number 15, Nudge for me, finds missed follow-ups in your inbox. Nudge for me is an AI follow-up assistant that finds missed opportunities hiding in your sent emails.
11:23 It solves the deals, proposals, and approvals that quietly go cold after your last message. You connect your mailbox and it scans sent conversations for real open loops, filtering out noise like completed threads and out-of-office replies. For the ones worth chasing, it writes a natural follow-up as a normal draft on the right thread, and you edit, send, or dismiss it.
11:44 It starts in draft mode, so you stay in control and stops watching a thread the moment someone replies. It works with Gmail, Outlook, Yahoo, iCloud, and other IMAP mailboxes. It is built for founders, agencies, and recruiters. Scan your inbox free. Project number 16, Port 22, your coding agents in your pocket. Port 22 is a universal remote for AI coding agents that lets you drive them from your phone.
12:09 It solves being stuck at your desk while an agent works and waits on you. You run a Mac binary that attaches to the Claude code, Codex, or open code session already open in your terminal without forking or restarting it. Then pair your iPhone over the same Apple ID with no separate login. You read every token of the live transcript, and a push lands the moment the agent finishes or needs approval, which you grant from your phone.
12:33 It works over your LAN at your desk and an end-to-end encrypted relay when away, so the relay never sees your code. It is built for developers who run agents on the move. Join the wait list for the beta. Project number 17, Minimax H3, open omnimodal model for video generation. Minimax H3 is an open weight general-purpose omnimodal generation model. It solves the fragmentation of tools that split image, video, and audio work into separate specialized models.
13:01 It understands unified context across text, images, video, and audio and generates video with native stereo sound at up to 2K resolution and 15 seconds. You describe the relationship between your inputs in plain language, referencing a video's camera move, an image's character, and an audio clip's vocals in one prompt, and it handles the full modality understanding on its own.
13:24 It follows instructions closely, renders text and brands accurately, and runs at a fraction of mainstream per second cost. It reaches developers through the API with model weights opening soon. It is built for advertising, e-commerce, gaming, and design teams. Try it through the API. Project number 18, Cleanlist AI, one clean data engine for GTM teams.
13:49 Cleanlist is an AI-powered B2B data platform that replaces several separate go-to-market tools. It solves the stale, scattered contact data that slows outbound sales. You type a request in plain English, like 100 founders in New York at companies of 50 to 200 people, and its co-pilot searches, enriches, and verifies in one pass, returning a CRM-ready list.
14:10 Every contact runs through a waterfall of 15-plus providers that compete for the most accurate email and direct dial. You can also enrich any LinkedIn profile in one click, drop in a CSV, or search companies and pull every relevant decision-maker. It syncs natively with HubSpot, Salesforce, Pipedrive, and more. It is built for sales and revenue teams.
14:31 Enrich your first leads free. Project number 19, Depth Data, one governed view of company AI spend. Depth Data is an AI spend intelligence platform that unifies how much your company uses and gets from every AI tool. It solves the problem that ChatGPT, Claude, Co-pilot, and Gemini each have their own console and their own idea of an active user, so no one can say what the spend is worth.
14:56 You connect each vendor's admin console through read-only OAuth with nothing installed on laptops, and it normalizes seats, usage, and spend into one schema with a single definition. It surfaces a workspace health score, adoption by team, waste price to named seats, and board-ready exports. Every figure labeled as measured, derived, or modeled. Prompts are never read.
15:19 It is built for finance, IT, and leadership. Connect your consoles and see what your AI is worth. Project number 20, Halo by Scam AI, on-device deepfake detection for video calls. Halo is an on-device deepfake detector that catches synthetic faces on your video calls in real time. It solves a new attack surface where real-time face swaps pass as a colleague on a live call and email filters and login checks never see them.
15:46 You install it once on a Windows PC where it sits in the system tray beside your conferencing app with no plugins. As you talk, it analyzes every face in the frame several times a second on your device and surfaces an alert during the call if a face looks synthetic in time to stop and verify. Everything runs locally, so frames are analyzed in memory and discarded with nothing recorded or sent.
16:11 It works across Zoom, Teams, Meet, and more. It is built for finance and HR teams. Download Halo and protect your next call. Thanks for watching. See you in the next video.