Creators and small studios set a creative goal and use specialized agents to plan, execute, revise, and deliver work across film, games, music, and social media.
Behind this: 5 build steps · 1 tool and how each is used · how to validate demand · 3 things the video never answers.
Teams connect agents to existing communication channels, assign roles and permissions, and allow agents to collaborate while keeping repositories, credentials, and model traffic on their own machines.
Behind this: 6 build steps · 8 tools and how each is used · how to validate demand · 3 things the video never answers.
Users describe browser tasks in plain language and an AI assistant opens tabs, scrolls, clicks, fills forms, and works with connected services.
Behind this: 5 build steps · 5 tools and how each is used · how to validate demand · 3 things the video never answers.
Organizations provide a self-service path for teams to build AI systems while security and IT retain visibility and control over what exists, who built it, and what it can access.
Behind this: 5 build steps · 1 tool and how each is used · how to validate demand · 3 things the video never answers.
Founders and SaaS teams consolidate repository documentation, help-center material, API references, and product notes into a maintained source that can serve people and AI agents.
Behind this: 7 build steps · 6 tools and how each is used · how to validate demand · 3 things the video never answers.
An AI customer agent handles complex customer interactions, updates accounts, processes payments and refunds, retrieves live data, and operates across multiple communication channels.
Behind this: 7 build steps · 7 tools and how each is used · how to validate demand · 3 things the video never answers.
Developers use an open-source CLI to validate data, select a post-training method, generate configuration and evaluations, gate model saves, and stream large frozen model layers from RAM or NVMe.
Behind this: 9 build steps · 8 tools and how each is used · how to validate demand · 3 things the video never answers.
Enterprises define agents as versioned files in their own repositories, specifying context, skills, tools, permissions, and identity so runtimes can read only what each agent is allowed to access.
Behind this: 8 build steps · 7 tools and how each is used · how to validate demand · 3 things the video never answers.
Teams subscribe to dashboards or charts and receive freshly rendered reports by email, Slack, or both on a schedule.
Behind this: 7 build steps · 4 tools and how each is used · how to validate demand · 3 things the video never answers.
Developers add MCP servers once behind a local gateway, expose them to multiple agent clients, reduce tool-definition tokens through on-demand search, and centrally control keys and destructive tools.
Behind this: 8 build steps · 6 tools and how each is used · how to validate demand · 3 things the video never answers.
Engineering teams coordinate coding agents, turn plain-English web and mobile journeys into self-healing tests, monitor production signals, and open pull requests for review.
Behind this: 8 build steps · 5 tools and how each is used · how to validate demand · 3 things the video never answers.
A company operating system gives a solo founder AI roles for CEO planning, CTO product work, CMO demand generation, and analysis while the founder retains ownership and judgment.
Behind this: 7 build steps · 1 tool and how each is used · how to validate demand · 3 things the video never answers.
Teams describe portal work in plain language, sign in once, and run scheduled automations that retrieve records, download statements, check statuses, and submit forms.
Behind this: 8 build steps · 1 tool and how each is used · how to validate demand · 3 things the video never answers.
AI agents submit text and target languages, complete a payment challenge when required, and receive publication-ready translations handled by professional native-speaker translators.
Behind this: 6 build steps · 2 tools and how each is used · how to validate demand · 3 things the video never answers.
Users import Chrome logins, connect agents over MCP, and delegate web tasks to isolated tabs while keeping the main browser, credentials, screenshots, and session replays local.
Behind this: 8 build steps · 7 tools and how each is used · how to validate demand · 3 things the video never answers.
An AI CRM builds a versioned customer record from calls, emails, and meetings, then uses agents to generate pipeline work, meeting preparation, and follow-ups.
Behind this: 7 build steps · 3 tools and how each is used · how to validate demand · 3 things the video never answers.
Distributed teams schedule stand-ups, retrospectives, planning poker, and mood checks in Slack, collect responses by time zone, integrate Jira activity, and produce reports.
Behind this: 10 build steps · 3 tools and how each is used · how to validate demand · 3 things the video never answers.
Developers provide a search query or URL and receive clean web content, structured data, screenshots, or metadata, with crawling and authenticated interactions available for larger workflows.
Behind this: 9 build steps · 5 tools and how each is used · how to validate demand · 3 things the video never answers.
Users capture complete conversations, condense them into summaries, and inject the saved context into another AI platform without cloud storage or sign-up.
Behind this: 7 build steps · 9 tools and how each is used · how to validate demand · 3 things the video never answers.
Teams perform a task normally while the application records meaningful actions, screenshots, and accessibility labels, then edit and export the resulting guide.
Behind this: 9 build steps · 3 tools and how each is used · how to validate demand · 3 things the video never answers.
Searchable transcript of Top AI Agent Projects : Omniwork, Argos, Fin, Firecrawl & BrowserOS neo — ManuAGI - AutoGPT Tutorials (18:37). 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, browsers, 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, Omniwork, an agent operating system for creative work. Omniwork is a desktop agent OS that runs creative work through expert agents built on top creators' expertise.
01:08 It solves the way creative production is split across many tools and hands so nobody owns the whole pipeline. You set a goal and Omni coordinates the right agents to plan, execute, revise, and deliver drawing on roles like video editor, music producer, film director, and social media operator. A shared memory keeps your taste, standards, and project history with you across every agent and you can turn your own workflows into reusable experts.
01:34 It runs on Mac and covers film, games, music, and social. It is built for creators and small studios. Download it and set your first goal. Project number two, Agent Connect, tag any AI agent into your channels. Agent Connect is an open source self-hostable layer that brings AI agents into the places your team already works. It solves the glue every team otherwise rebuilds to turn a lone agent into a real teammate.
02:00 You run the Damon wherever your code lives, connect any ACP agent like Claude code, code X, or Gemini CLI, and tag them in Slack, Discord, Telegram, or GitHub. You give each a role, memory, and scoped permissions, so agents call on each other, act across a thousand plus apps, and remember what matters between chats. Your repos, credentials, and model traffic stay on your own machines with the control plane holding only configuration, never transcripts or code.
02:30 It is built for teams. Star it on GitHub and put an agent in your first channel. Project number three, manage Argus, AI assistant that controls your browser. Argus is a Chrome extension and AI assistant that takes real actions inside your browser. It solves the manual clicking and repetitive web work that eats your time. You ask it in plain language, and it opens and closes tabs, scrolls, clicks, fills forms, and interacts with any element on the page.
02:59 It connects services like Google Docs, Gmail, and Google Sheets, so it can research topics, automate tasks, and organize your workflow without leaving Chrome. It requests access to your data only to carry out the actions you ask for. It is built for anyone who wants their browser to do more on its own. Add it to Chrome and hand it your first task. Project number four, Times, a secure place for every team to build.
03:23 Times is a secure environment where every team builds AI agents, apps, and automations while IT and security keep control. It solves the risk that spreads once everyone starts building alone with keys committed to GitHub and apps shipped on personal accounts. Teams self-serve on a governed path, so security sees what exists, who built it, and what it touches.
03:42 Its agents run sensitive workflows like alert triage, IT deployments, finance requests, and hiring, and it suggests fixes for anomalies you decide what goes live. It is vendor agnostic across your stack and lets you choose where and how your models run. It is built [clears throat] for security, IT, and the teams that depend on them. Sign up free or book a demo.
04:04 Project number five, Dia knocks a lot, documentation humans and AI agents can read. Doc A Lot is a documentation platform that publishes docs both people and AI agents can onboard from. It solves the drift where scattered help pages and API notes leave assistance citing stale or missing answers. You connect your GitHub docs, help center, API reference, and product notes, and it normalizes them into one maintained source, then packages that same content as clean markdown, stable anchors, llms.aesk, skill.md, and MCP
04:36 ready chunks. It hosts a polished doc site, exposes a hosted MCP endpoint, so tools like ChatGPT, Claude, and Cursor retrieve the right pages, and runs a visibility audit showing what agents can actually cite and where to fix next. A CLI and skill let your coding agent create docs from the repo. It is built for founders and SaaS teams. Connect your docs and publish an AI-readable source.
05:01 Project number six, Fin, AI customer agent across the whole journey. Fin is an AI customer agent from Intercom that handles support, sales, and e-commerce across the entire customer journey. It solves the tickets and inbound questions that pile up on human teams around the clock. It runs on Apex, models custom trained on billions of customer interactions, so it reasons through complex, multi-step issues, updates accounts, and processes payments and refunds.
05:29 It pulls real-time data and takes actions through APIs, data connectors, or MCP. Works natively across voice, chat, email, and social in dozens of languages, and plugs into help desks like Intercom, Salesforce, Zendesk, and Freshdesk. You configure its tone, knowledge, and behavior yourself, test changes before they go live, and pay only for resolved conversations.
05:51 It is built for support and CX teams. Start a free trial or watch a demo. Project number seven, Soup CLI. Fine-tune large models on small GPUs. Soup is an open-source command-line tool that runs the whole model post-training stack in one place. It solves the tuning and guesswork of fine-tuning, and the fact that big models usually will not fit on a small card.
06:15 It checks your data first, picks the method, and writes the config from rules rather than a search. Then derives evals from your own data and gates every save. When the model is bigger than your GPU, it streams the frozen base from RAM or NVMe one layer at a time, and quantizes it to 4-bit. So, an 8-bit model fine-tunes on a 4-GB laptop GPU with preference methods like DPO and Orpo streaming, too.
06:38 It runs on your own hardware, offline, and works with Hugging Face, Ollama, vLLM, and Apple MLX, plus an MCP server for coding agents. It is built for developers training their own models. Install it and run your first fine-tune. Project number eight, Bevel. Git back control plane for enterprise agents. Bevel is a vendor-agnostic control plane that defines your company's AI agents as files you own in your own Git repo.
07:04 It solves the lock-in of agents living inside a vendor's harness, where context, tools, and lessons scatter, and switching is costly. Four things fully specify an agent. Context as typed knowledge nodes with provenance, skills as plain markdown procedures, tools and permissions as manifests with vaulted secrets, and file-level access rules, and identity as per agent credentials.
07:29 Your repo is the source of truth, and any runtime like Claude Code, Cursor, ChatGPT, or a background agent reads exactly what it is permitted to over MCP, with changes reviewed as diffs. Its core Hexis is open source. It is built for enterprises. Try the open source version. Project number nine, based as subscriptions. Schedule dashboard snapshots to inbox or slack.
07:51 Based as subscriptions is a feature in the Based as BI platform that delivers a fresh dashboard or chart on a schedule you set. It solves the manual reporting errand where someone has to remember to open a dashboard, screenshot it and paste it before a meeting. You open any dashboard or chart, hit subscribe, pick a cadence in a plain sentence like every Monday at 9:00 a.m.
08:14 and choose email, a slack channel or both. When the schedule fires, it renders each chart from live data and sends it. So, a subscription never ships a stale report with a link back to the interactive dashboard. Subscription stack, so one dashboard can feed several audiences and you can pause any without deleting it. It is built for teams with reporting rituals.
08:34 Start free and subscribe to your first dashboard. Project number 10, Toolport. One local MCP gateway for every agent. Toolport is a free open source local gateway that puts all your MCP servers behind one port. It solves the cost and repetition of wiring the same connectors and keys into every agent and the token tax of dumping hundreds of tools into context.
08:59 You add each server once or import the ones your agents already have, then toggle it on in every client with no restarts. Instead of loading every tool definition, agents search a handful of meta tools on demand, cutting tool token sharply at the same task success. It fingerprints every tool to catch rug pulls and hidden instructions, keeps API keys in your OS keychain and lets you switch off destructive tools fleet wide with per server latency and a full audit trail.
09:26 It works with Claude, Cursor, VS Code, Codex and more and runs on Windows, Mac and Linux with no cloud. It is built for developers and teams. Download it and share your servers. Project number 11, coldti.ai. Agents for the rest of software delivery. Coldti is an agentic development environment that covers what happens after code is written. It solves the risk that move downstream as agents write more code into regressions nobody caught and incidents users report first.
09:56 It brings your terminal, end-to-end testing, and production monitoring into one Mac app that runs next to your repo. You steer several coding agents as a team, describe web and mobile test journeys in plain English that become self-healing tests gating every pull request, and let monitoring agents watch error logs, feedback, and sessions. Then open a pull request for your review.
10:16 You can sync your board and hand tasks to background agents in the cloud only when you turn it on, so code stays on your machine. It works with Claude code, Codex, and open code. It is built for engineering teams. Download it for macOS. Project number 12, Solop, an AI founding team for solo founders. Solop is a one-person company OS that gives solo founders an AI founding team to grow the product they built.
10:41 It solves the moment where a founder ships a demo but stalls, unsure how to iterate and turn it into revenue. An AI CEO plans the next move and coordinates the work while a CTO pushes the product forward. A CMO turns attention into demand, and an analyst hunts for the next signal. You keep ownership and judgment while delegating the unfamiliar tasks to the team, which runs continuously so the insights are ready when you open your laptop.
11:07 It is built for solo founders. Start the loop and run your company. Project number 13, Rindler, automates the web work behind logins. Rindler is a web automation agent that signs into the sites your team already uses and does the busy work. It solves the manual grind on portals that have no API, where operators pull records, download statements, check statuses, and submit forms by hand.
11:30 You describe the job in plain words, connect the site by signing in once, and it runs showing every step and screen it saw. It walks each site first, so a run follows a route it already knows, and when the site changes, Rindler repairs that route centrally for everyone with no ticket or extra charge. Every action follows a rule you set, allow, ask first, or never per site, and your login stays encrypted and scoped to that one site.
11:55 You can schedule as many automations as you need. It is built for procurement, government, healthcare, HR, and finance teams. Start a free trial and name your first site, project number 14, Nitro 4.0 human translation API built for agents. Nitro API is a human translation service that AI agents can call on their own with or without an API key. It solves the setup wall of ordinary translation APIs where an agent needs an account, keys, and a dashboard before its first request.
12:27 Your agent posts the text and target languages with no auth, gets back a payment challenge, obtains a token from any MPP compatible payment provider, retries the same request with the credential, then polls each order until it is done. Professional native speaker translators handle over 80 languages preserving HTML, JSON, and other markup with no minimum order and a rough 2 to 24-hour turnaround.
12:51 Endpoints for rates and price estimates need no authentication at all. It is built for developers whose agents need publication-ready translation. Read the API docs and let your agent order. Project number 15, Browser OS Neo, local browser your AI agents drive. Browser OS Neo is a free, open-source browser that your AI agents drive using your logged-in accounts.
13:16 It solves the gap where assistants sound capable in chat but cannot open a tab, sign in, and click through a real site. You import your logins from Chrome in one click, Connect an agent like Claude Code, Co-work, Codex, or Cursor over MCP and hand off web tasks. Each agent works in its own tabs signed in as you, so several run in parallel while your main browser stays yours, and it reads compact page snapshots to use fewer tokens.
13:43 You watch every step live and replay any session as a scrubbable video. Everything runs on your machine with logins, screenshots, and replays kept as local files. It runs on macOS and Windows. It is built for people who run agents on the web. Download it and give your AI a browser. Project number 16, Lightfield, AI native CRM that does the admin. Lightfield is an AI native CRM that handles the busy work instead of asking sellers to log it.
14:11 It solves the field updates, note taking, and follow-up drafting that legacy CRMs pile onto your team. It builds a [clears throat] living versioned picture 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 stalled deals. You ask questions in plain language and get answers cited back to the original conversations, and you assemble your own agents from skills, knowledge, and automations connected over APIs or MCP.
14:39 It is built for early-stage, high-growth teams. Try it free and let the agents work. Project number 17, Troop AI Scrum Master, automate Agile ceremonies inside Slack. Troop is a Slack-based assistant that runs your Agile ceremonies on autopilot. It solves the time lost to status meetings and manual context switching for distributed teams. You pick a template, set a schedule, and it asks your team stand-up, retrospective, planning poker, or mood questions on its own, collecting answers asynchronously in each person's
15:13 time zone without leaving Slack. It pulls activity from Jira, links and expands referenced issues, and lets people update their tickets straight from Slack, then rolls everything into one report with insights, history, and dashboards. It handles time zones, holidays, absences, and external participants, and exports reports as PDF. It is built for Agile and engineering teams.
15:35 Add it to Slack and automate your first stand-up. Project number 18, Firecrawl, context API to search and scrape the web. Firecrawl is a web data API that lets AI agents search, scrape, and interact with the live web at scale. It solves the fact that the web is built for humans, so pages come back as messy HTML full of navs, ads, and scripts an agent cannot use.
16:01 You give it a query or a URL and it returns clean markdown, structured JSON to your own schema, screenshots, or metadata, rendering JavaScript, and waiting for content on its own. Search returns full-page content in one call. Crawl follows links across a whole site and interact clicks, fills forms, and navigates flows behind logins. It parses PDFs and DOCX too, and strips the noise so your model spends far fewer tokens.
16:30 An official MCP server, a CLI, and skills connect Claude, Cursor, Codex, and other agents in one command. It is open source with a free tier. It is built for developers powering agents and research pipelines. Sign up and make your first call. Project number 19, Prompt Bridge, carry AI context across every platform. Prompt Bridge is a free Chrome extension that carries your full conversation context across major AI platforms.
16:56 It solves the restart problem where switching from one assistant to another means re-explaining everything and losing your thread. You capture any full conversation from ChatGPT, Claude, Gemini, Grok, or Deep Seek in one click, and its AI condenses long chats into tight summaries to cut token costs. When you move to another tool, it injects that saved context into the input box with one click, and it auto-detects the active platform on tab switch, so there is nothing to reconfigure.
17:24 Everything stays on your device with no servers, cloud, or sign-up, and it works on Chrome, Edge, and Brave. It is built for anyone who works across several AI tools. Add it to Chrome and keep your context. Project number 20, StepShot, turns Mac workflows into step-by-step guides. StepShot is a Mac app that records what you do and turns it into a polished step-by-step guide.
17:47 It solves the slow part of writing documentation, which is never doing the task, but capturing it, cropping screenshots, drawing arrows, and writing captions. You work as usual, and it watches clicks and keystrokes, making a step for each meaningful action with a before-click screenshot and a title read from the system's accessibility labels, rather than guessed.
18:08 You then reorder, merge, annotate, and redact, and export to PDF, Markdown, or HTML with your logo and theme. Optional AI cleans up structure and wording, off by default, and everything runs locally with no account, telemetry, or upload, and redaction burned into the image. It is built for teams writing SOPs, support answers, and training docs. Download it and record your first guide. Thanks for watching. See you in the next video.