Top AI Agent Projects : Nimble, Slashy, Aside, Juggler & Loqua Transcript, AI Summary & Key Points

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 search, email, and coding. 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 cuttingedge agent frameworks, make sure to check it out.

00:47 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, web search agents by Nimble. Self-learning web search for your domain. Web search agents by Nimble are self-learning agents that become experts at your specific research task.

01:09 They solve the way generic search tools return shallow, redundant results and burn tokens, forcing a model to parse raw pages. You define a domain and schema and the agent crawls the web with precision, combining search with deep domain crawling to reach subpages other tools miss and returns exactly the fields you asked for consistently on every run.

01:29 It accumulates web context over time through a proprietary index in memory. So accuracy compounds with each query and auditable search plans show exactly what was searched, where, and why within guardrails you set. It also builds and enriches data sets and monitors any data point on any page in real time. It handles company research, real estate, finance, product intelligence, and more with native integrations for Anthropic, Open AI, Langchain, and Versel, plus an MCP server, and keeps data private with zero retention

02:02 and no training. It is built for agent builders and research teams. Start building for free. Project number two, Hello Inbox, AI deliverability stack for email senders. Hello Inbox is an email deliverability platform that pairs an AI expert with nine tools to get your campaigns into the inbox. It solves the hidden problem where carefully written emails land in spam, quietly costing a big share of revenue.

02:27 You connect a sending domain in a couple of minutes and build a sender profile that documents your DNS, authentication, audience, and postmaster data, giving the AI the context to answer your exact situation rather than generic advice. It sends a test email and analyzes authentication, content, and reputation with suggested fixes. Verifies lists to catch invalid addresses.

02:49 Flags spam trigger words and rewrites them. monitors DM Arc and runs inbox placement tests across Gmail, Outlook, and Yahoo. Alerts warn you on blacklisted IPs, spam spikes, or reputation drops. And the assistant keeps memory of your profile and past chats, so you never repeat yourself. It works with any email provider. It is built for marketers and serious senders.

03:13 Start testing free in 2 minutes. The page is JavaScript rendered, so I only have the title. Let me find the official details another way. I have solid official details from Slashy's own launch and pages. Project number three, Slashy Assistant. AI native email client that acts. Slashy is an AI native email client and assistant that does the inbox work for you.

03:34 It solves the two failures of bolt-on AI writers. Drafts that sound like a press release and tools that hand you text but never notice the follow-up you forgot. It drafts replies in your actual voice by learning from your sent mail. triages what matters and clears the noise and tracks who still owes you a reply so nothing slips because it connects to your email calendar and meeting notes.

03:59 It briefs you before a call, schedules meetings by finding open slots and pulls real context instead of generic filler, getting sharper each time you correct it. It carries the speed, keyboard shortcuts, and command pallet of a fast client, plus a native unified inbox, and you can fire off a draft or summary from iMessage or Slack without opening the app.

04:19 MCP clients like Claude Desktop, Cursor, and Codeex can act on your inbox and calendar, too. It is built for founders, operators, executives, and sales teams. Start a free trial and reach inbox zero. Project number four, Oats. free open-source meeting notes on device. Oats is a free open-source meeting notes app for Mac from Oriso that records right on your machine.

04:42 It solves the privacy and lock-in problems of notetakers that send a bot into your call and park your recordings in their cloud. You hit record and it listens to any audio your Mac can hear across Zoom, Meet, Teams, a phone call, or a hallway chat with nothing joining the meeting. When it ends, you get a quick digest, a full summary, action items with an owner each, and a quality score drawn from what was actually said.

05:07 Those action items collect in a Toto's tab, so ticking one off in the notes closes it, and follow-ups get followed up. A searchable library lets you ask questions across your whole history, and you run everything on device with a local model, so no audio leaves your machine or connect for richer features like speaker tagging. Local notes are plain marked down in an Obsidian vault and every line is on GitHub to fork or self-host.

05:33 It is built for anyone who lives in meetings. Download it and record your first meeting. Project number five, Elva. Discover, govern, expose APIs to agents. Elva is an API management system built by Theno that discovers every API in your code and exposes it to developers and AI agents. It solves the way APIs scatter across repos with no spec, no owner, and no safe path for agents to call them.

05:58 You point it at a repo and it reads root registrations, validators, and serializers to find every endpoint, generating an open API spec with no spec required, then scores each on design, security, and AI readiness, and fixes gaps like missing descriptions and O with an agent. API contracts decide what ships. You pick an audience, endpoints, and fields.

06:21 exclude PII and ELVA diffs every commit against that promise blocking breaking changes before they reach a partner from a contract. It generates a hosted MCP server behind an O gateway with scoped keys per tool authorization and full call logs so claude cursor chatgpt and partners call your APIs safely. A playground tests it and the agents calling it report which tools confuse them.

06:45 It is built for engineering and platform teams. Sign up free and turn a repo into a catalog. Project number six, Aside, a browser that does real work. Aside is an AI browser rebuilt so both you and agents can get complex work done across your logged in sites. It solves the failure of other AI browsers that stall when the work gets real and keep saying they cannot do it.

07:08 Instead of relying on integrations, Aside uses websites and accounts directly the way you do, so you can hand it almost any task. signing into email, dashboards, and internal tools, handling comments, replies, and follow-ups. Working with files on your computer, and carrying out payments. It turns your browsing history into memory, so it knows which sites to use, and you never repeat context, and that memory stays local on your device.

07:33 Its vault autofills credentials into sites without ever exposing them to the AI. and sensitive actions like payments, posts, and messages wait for your approval with every credential use logged. Tasks and data stay on device, sandboxed and encrypted, and you bring your own chat GPT cla or API key. It is built for operations, hiring, sales, and research teams.

07:58 Download it and hand it your first task. Project number seven, Try Case. Open a PR, get demo videos. case is a visual end-to-end testing agent that tries each pull request in a real browser and posts the proof on GitHub. It solves the blind spot of code review where a diff looks fine but nobody sees whether the feature actually works and where writing and maintaining a browser test suite is its own chore.

08:21 With your repository connected, opening a pull request starts it. It reads the change, picks the user journeys affected, then opens your app and clicks through each one like a person. Back on the pull request, it leaves a verdict, a captioned video, and a screenshot. So, you watch the discount apply or the checkout complete before you merge with no test scripts to write.

08:42 It runs on the AI you already pay for, connecting codecs with your chat GPT subscription or your own open router key. It is built for developers who want to see what changed before shipping. Connect a repository and test your next pull request. Project number eight, Ozrain. Shared brain. Your agents read and write. Ozbrain is a hosted knowledge base that every AI agent you use can read and write.

09:05 It solves the way contact stays trapped in separate chats, so you and your teammates paste the same background again, and the current version of a plan is wherever someone last saved it. You point every agent at one brain, and what you work out, everyone's agents already have. It stores your project's decisions and research as structured, nested articles, so an agent pulls the exact slice it needs.

09:28 The email body, not the whole launch plan, which makes answers faster, cheaper, and less prone to invention. When newer thinking lands, it marks the old as replaced and points to the latest on its own. And every change is staged, attributed to an agent with its reasoning, so several agents work at once without overwriting each other. You add it from the connector menu in Claude, Chat, GPT, Cursor, Claude Code, and Gemini with no code, and your data is encrypted, never trained on, and exportable as markdown anytime.

09:57 It is built for solo builders and teams. Start free and give your agents a brain. Project number nine, Juggler visual workbench to see coding agents. Juggler is a visual workbench for AI coding agents that lets you see and control what a model is doing to your code base. It solves the way a scrolling terminal transcript hides tool calls and buries context, so you cannot tell what the agent actually received or did.

10:24 Conversations are persistent trees rather than log files. So you branch at any point, send a tangent or competing approach to a subthread and return only its result to the parent instead of flooding your main context. Every tool call opens into its own view. And any model transaction can be inspected down to the system prompt, messages, tools, tokens, timing, and stop reason with a context surgeon to fold, move, and undo pieces of history.

10:51 You launch the desktop app for local work or run juggler on the machine where the code lives and open the same live session from a browser or your phone with the session saved to disk so it survives quits and reconnects. It works with clawed code, codeex, copilot, Gemini, Olama, and other providers on your own subscription or keys. And its core is open source with no account.

11:13 It is built for developers doing hands-on agent work. Download it and open your project. Project number 10, Resurf local context library you hand to AI. Resurf is a fully local personal context library for Mac, iPhone, and iPad that captures what you want to keep and hands it to AI. It solves the friction of saving that makes you not bother. And the way notes, links, and files scatter where no assistant can reach them.

11:39 You capture links, notes, images, voice, and documents with a shortcut or drag and drop from Mac, Chrome, or iPhone, and everything lands in one inbox to organize later into spaces. You browse captures as visual cards, search and filter by type, read articles in a distraction-free reader, and view PDFs natively, all navigable from the keyboard. A side AI pane answers from your own captures, pulling in the relevant ones and the contents of files you saved, running on your own open AI key, so you see exactly what it

12:10 costs. And you also hand that saved context to agents through MCP or the CLI. Your data stays on your Mac, works offline, needs no account, and syncs through your own iCloud. It is built for anyone who collects ideas and references. Download it and start capturing. Project number 11, Perplexity Hybrid Compute. Splits agent tasks between cloud and Mac.

12:34 Project number 11. Perplexity hybrid compute. Splits agent tasks between cloud and Mac. Hybrid compute is a feature of Perplexity's computer agent that splits a single task between cloud models and a local model on your Mac. It solves the trade-off where Frontier Reasoning lives in the cloud, but your sensitive files should never leave your device. You give computer a task and it runs research and reasoning in the cloud while your Mac handles private files and apps locally, then brings both into one result.

13:02 An ondevice PII classifier reads each task before anything is sent, swapping names, addresses, and account numbers for stand-ins and restoring them when the answer returns. And you can force any file to process only on your Mac. You can start a task from your iPhone and have your Mac run the protected steps. And it works across your Mac apps, local files, and connected apps like Slack and Mail.

13:24 You set up a local model once choosing Gemma, Quinn, or a Perplexity model in one click with no Alma or runtime setup, and local work uses no cloud credits or API key. Perplexity has open- sourced the ondevice PII classifier. It runs on Apple silicon. It is built for professionals handling sensitive work like finance and legal. Download the Mac app and run your first hybrid.

13:47 Number 12, Visibi. See, and win AI search citations. Visibi is an AI visibility platform that tracks what Chat GPT and other engines say about your brand and tells you how to win. It solves the blind spot where buyer research has moved into AI assistance. Yet, rank trackers and analytics never sample those prompts or explain why a model recommends a competitor.

14:10 It continuously runs your prompt universe across chat GPT, Claude, Perplexity, Gemini, and Google AI overviews, then reports per engine citation share, the exact prompts you win or lose, and the adjectives you own versus those owned against you. A citations explorer ranks every URL the engines quote. And a recipe heat map shows which onpage elements like headings, FAQ blocks, and tables each engine extracts from so you know what to publish.

14:39 Every week, it fuses your brand prompts, citation gaps, and competitor recipes into one prioritized priced action plan with ready to ship content briefs delivered to your inbox with no analyst meeting. It also folds in traditional search performance alongside the AI data. It is built for marketing operators and agencies. Get a free visibility report in 60 seconds.

14:59 Project number 13, Ghostriter by My Handler. Reads your screen, writes the reply. Ghost Writer is a writing feature of the My Handler desktop app that drafts from whatever is on your screen. It solves the daily grind of rewriting the same reply where describing what you want is the very work you are avoiding. You put your cursor in any text field, double tap the hotkey, and your handler reads the context around it, the thread above, the question beside it, the half-finish sentence in it, and works out what kind of

15:32 writing this is before typing a word, a reply owed, a sentence to continue, a blank box to answer, or a message the situation calls for. It maps each message to its sender, so it answers the other party, not your own words. And if a draft proposes a time, it checks your calendar first. The text appears at your cursor in your voice and you read, edit, and send it since it never sends anything itself.

15:54 Screen capture and your vault stay on your PC with the assembled context sent to a zero retention cloud model to write. It is built for anyone who lives in their inbox and chats. Download it free for Windows. Project number 14, Epude Notetaker. Meeting notes on Mac, no bot. Epilude Notetaker is a Mac meeting notes feature that records on your machine with no bot joining the call.

16:18 It solves the way meeting bots sit in the participant list and change how people talk. And the scramble of writing up notes by hand afterward. It captures your microphone and the meeting audio from your side. So any call your Mac can hear becomes notes across Zoom meet teams, Slack huddles or a conversation across the table. The transcript builds live and when you stop, you get a summary shaped to the meeting type since it classifies each call and pulls the decisions, blockers, or commitments the conversation actually

16:48 produced with key points cited back to the transcript. Action items become to-dos with the owner and due dates set out loud, linked to where it was agreed, and it never invents an owner or guesses a deadline. It offers to record with one click when a call starts, recovers a recording if your Mac restarts, and saves each meeting as a markdown file in a folder you choose.

17:08 Searchable and ready for Obsidian, Notion, or PDF. It is built for anyone who runs on meetings. Start free with 10 meeting notes. Project number 15, NeoPress. Build and grow a website with agents. Neopress is an AI website builder where agents design, publish, and grow your site through conversation. It solves the way building a site scatters across a page editor, a CMS, SEO settings, and an analytics tool that never talk to each other.

17:35 You describe the page you want in plain language and refine layout, copy, and style by chatting with no templates or design tools to learn. A CMS agent drafts and organizes content in collections for you to review and publish. And an analytics agent explains your traffic and suggests what to improve so you decide what to apply. Every page ships as server rendered HTML with site maps, structured data, canonical tags, and an autogenerated LLM.TXT, so search engines and AI crawlers read real content instead of an empty

18:06 shell. It counts human visits and AI citation crawls as page views, tracks which models crawl your pages, and supports custom domains, multilingual publishing, forms, and MCP. You can migrate an existing site as is, and you own all your content. It is built for startups, agencies, professionals, and local businesses. Start a free trial and build your site.

18:29 Project number 16. Yuki finds who's ready to buy. Axe. Yukti is an agentic outbound platform that turns account knowledge and live signals into revenue actions for every deal. It solves the way outbound scatters across prospecting, research, sequencing, and CRM tools, leaving reps guessing which account to work and what to say. Its agent, Arya, builds a whole campaign through conversation.

18:54 You describe the flow in plain English, and it sets the signal triggers, persona matching, cadence, and outreach rules, updating the config live as you talk. It monitors signals across your market, funding, hiring, leadership changes, lawsuits, and competitive moves, and for each one gives the reasoning, the business impact, and a timing window rather than a bare alert.

19:16 Every morning, reps open one cockpit of prioritized accounts with matched personas and written sequence hooks, push outreach in a click, and each account enters a track journey with healthc scoring and AI generated next steps. It scores accounts on readiness and ICP fit. Writes AB email variants grounded in real intelligence and pushes to outreach, sales loft or lem list, syncing your CRM automatically.

19:39 It is built for sales reps, AEES, and RevOps teams. Get started and execute from day one. Project number 17 bashing Q Apilot MCP for Android. Test Android apps in plain English. Q a Pilot MCP is an Android test automation server you drive by talking to Claude Cursor or any MCP client. It solves the barrier of writing and maintaining Appium code just to run a mobile test flow.

20:06 You install the CLI, add the MCP config and connect it to real devices or emulators over Appium. Then describe a flow in plain English like tap search, type a query, select the first result, and verify the detail page loads. The AI builds a structured plan and runs each step on the device in real time with autogenerated readable step titles and a live preview URL.

20:28 So you watch the run in your browser. Once a happy path passes, you accept the steps as a saved test case, then replay tests by ID, in batch, or from an Excel sheet and generate a report with step results. screenshots, errors, and timing, including a Girkin feature file on pass. Everything runs locally against your own device. It is built for QA engineers and mobile developers.

20:52 Install the CLI and record your first flow. Project number 18, Abrush, AI image generation inside Photoshop. Abrush is an AI image generation plug-in that lives inside Adobe Photoshop. It solves the constant switching between Photoshop and separate AI services, which breaks a creative workflow and scatters your work across tools. It brings generation into the app you already use, so you create images from a prompt, edit them in your document, and upscale results without leaving your canvas.

21:23 You work with over 20 top models, including stable diffusion, flux, and Quinn image, and pick the best one for each stage. Then edit only the region you need and refine details by hand in Photoshop. Professional controls cover control net IP adapter references, Laura styles, and generation history and presets save prompts and settings so you reuse your best results and share them instead of starting over.

21:49 You own every image you generate. It is not used to train models and commercial projects stay private. It is built for artists and creative teams. Try it in Photoshop. Project number 19, Loqua. Contextaware voice typing that understands you. Loqua is a contextaware voice typing app for Mac and Windows that understands what you mean, not just what you say.

22:11 It solves the slowness of typing and the cleanup ordinary dictation leaves behind. You invoke it with one shortcut in any app, terminal, Slack, notion, email, or a code editor. Speak naturally and polished text lands at your cursor with filler words removed, repetition cut, and phrasing refined in real time. It hears the structure in your speech and builds lists, headings, and hierarchy on its own and translates into nearly a 100 languages so you speak yours and send theirs.

22:41 Beyond dictation, you capture any part of your screen and ask about it to get an answer, analysis, or summary without switching apps. Highlight text and speak to rewrite it on the spot and trigger actions like reminders, opening apps, or sending a text by voice. It also reads selected text aloud for hands-free listening. It is built for anyone who would rather think than type.

23:03 Download it and try it free. Project number 20, Wizree. Clone winning e-commerce ads with AI. Whizree is an agentic AI growth platform that finds the ads already winning in your market and rebuilds them for your brand. It solves the creative bottleneck in paid acquisition, where testing properly means fresh concepts every week, and most teams wait through briefs, freelancers, shoots, and edits for a few variations that may still flop.

23:28 You paste your website URL, and it reads your site to build a brand DNA. Your positioning, products, tone, colors, and logo, so everything it makes stays on brand with no brief. A research agent searches ad libraries for ads worth cloning and ranks them with the evidence behind each, like how long one has run and how many variants sit behind it. You pick one and a cloning agent rebuilds its structure, not the words around your product as static and video ads with AI presenters, B-roll, captions, and music.

24:00 A built-in editor lets you ask an agent to add hook overlays, restyle captions, or translate. Then it launches to Meta and Google and keeps optimizing for rorowaz. It runs on creative credits with the cost shown before a run orchestrated across leading models. It is built for DTC and e-commerce brands running their own ads. Start your first campaign run. Thanks for watching. And see you in the next video.