Top AI Agent Projects : GenCode, Okara, Arc, Superhuman Go & Hemory 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 coding, marketing, and voice. 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 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, MENCP connectors by Databox. Ground your AI in trusted metrics. Databox MCP connects AI tools like Claude, Chat, GPT, and N8N to your business metrics and the logic behind them.

01:12 It solves the way assistants recalculate from scratch and guess at context whenever you paste data in. Through one endpoint set up in about a minute, your AI queries your real governed datab box metrics using your definitions, history and calculation logic, and returns visualized answers, dashboard breakdowns, and multi-account comparisons. You push new data in, run pre-built skills, and automate recurring reports.

01:36 Every connection uses OOTH, follows your permissions, and never trains models. It is built for agencies and data teams. Connect your AI and start free. Project number two, Statable Analytics. Privacy first web analytics with agent setup. Statable is a privacy-friendly web analytics platform and a simpler alternative to Google Analytics. It solves the cookie banners, consent friction, and complexity that G4 brings.

02:02 You add one small script or a WordPress plugin and it reports visitors, sessions, page views, traffic channels, devices, goals, and funnels in real time with no cookies and nothing stored on a visitor's device. You import your G4 history in one click, so you never start empty and connected, chat, GPT, cursor, or codecs over MCP, so your agent sets up analytics, goals, and funnels for you.

02:27 IP addresses count a visit and are discarded and data is EU hosted under GDPR. It is built for site owners and teams. Start free with no card. Project number three, Mianda Okara, an AI CMO running marketing agents. Okara is an AI CMO that puts marketing on autopilot with a team of specialist agents. It solves the cost and scatter of hiring a marketer, an SEO agency, writers, and social managers just to get traction.

02:55 It first researches your business and writes five strategy documents covering your product, positioning, competitors, brand voice, and content plan. Then every agent reads them before writing so nothing drifts off message. Agents for SEO, Reddit, writing, GOX, LinkedIn, technical fixes, UGC video, and influencers draft work you approve. Then publish to WordPress, Web Flow, Wix, GitHub, and your social accounts.

03:21 Once connected, pulling live search console and analytics data, you stay in control, approving before anything goes out. It is built for founders and lean teams. Enter your website and start free. Project number four, Gen Code, one coding agent, every frontier model. Genode is genark's AI coding agent that runs any model you choose from one account.

03:45 It solves the lock in and cost of coding tools. tied to a single provider and separate API keys. You pick Claude, GPT, Gemini, DeepSk, or an openweight model and run open models at a fraction of the usual cost with no key setup and enterprise model control built in. It plugs into your Genenspark skills, your authorized connectors, and the clawed code instructions, rules, and memory you already built.

04:08 So, it starts knowing how you work. You work in a chat style desktop workspace or a terminal UI with the same capabilities and track credits as you go. It is built for developers and teams. Get the desktop app or install the CLI. Project number five, harness router. Skip routing when the tool is obvious. Harness router is a tiny open-source decision layer that decides when an AI agent even needs to think about which tool to call.

04:34 It solves the wasted planner turns and cost of reasoning through steps that are already obvious. When the next action is clear, the agent just calls the tool with no routing overhead. When the choice is a real closed set, it asks a lowcost model called Jev for a structured pick with a confident score. When the next few moves carry consequences, it runs a bounded tree search over a side effect-free state graph and returns only the first real action, never executing simulated writes or commands.

05:03 It runs as a native stddio MCP server for codecs using your open router key and is MIT licensed. It is built for developers building agent harnesses. Install it and route less. Project number six, third router. Keep your coding tools, choose models. Third router is a routing layer that lets your coding tools run the models you choose through one endpoint.

05:25 It solves the lockin of coding clients tied to one provider and the guesswork of tracking usage across accounts. You point a supported client at a single router URL with a scoped key. Then pick models, set daily budgets on your API keys, and configure fallback chains for supported requests. It shows your codeex capacity across connected accounts with remaining weekly quotas, reset dates, and banked resets clearly marked when stale, and breaks down routed usage by model, date, and client.

05:54 You bring your own provider keys, and it never switches your local codeex account on its own. It works with codecs, pi, open code, and other open AAI compatible apps. It is built for developers. Check access and set up your first route. Project number seven, Gancha Ark. AI assistant that reads your screen. Ark is a free AI assistant that works on the screen you are already on.

06:18 It solves the copy paste and explain routine that ordinary chatbots force where you bring your situation to them. On Android, it is a floating sidebar over every app. And on Mac, a panel you open with control space over any window so it reads the current screen and acts on it. You summarize articles and PDFs, rewrite or reply in place. Hear any screen read aloud, extract dates, contacts, and codes, and turn reading into flashcards.

06:44 You save your own prompts as one-tap actions with hotkeys or install ones the community shares across more than 10 languages. It reads the screen only when you tap, disables itself in sensitive apps, and never stores screen content on its servers. It is built for anyone who reads and writes on a device all day. Get Ark free on Android and Mac. Project number eight, Sable Realtime AI co-pilot for live calls.

07:10 Sable is a realtime AI co-pilot that listens to your live calls and hands you the exact words to say next. It solves the pressure of highstakes conversations where the right answer to an objection arrives after the moment has passed. It transcribes both sides live, separates who said what, and surfaces one clean line to speak in about half a second, staying hidden from screen share and recording.

07:33 When the call ends, it writes a recap of what was decided, the objection that stuck, and what you owe them with a follow-up email already drafted. It remembers every call so you ask your whole history questions in plain language and see what is losing you deals. It pulls today's meetings from Google calendar. Here's an answers in over 10 languages and runs on Mac and Windows.

07:56 It is built for closers, founders, client teams, and recruiters. Download it free and try it for 3 minutes. Project number nine, PIP E. AI buddy that points where to click. PIP is an AI buddy for your Mac that sees your screen and points at exactly what to click. It solves the way people get stuck hunting for the right button in unfamiliar software since a chat box makes you describe your screen instead of just showing you.

08:20 You hold a key and ask out loud and it answers in a sentence while circling the exact control following you click by click even when a task takes five steps. It works in any app from system settings to video editors and the terminal where it shows the command to type and what it does. It points you click so you stay in charge and it can also do small local actions like timers or dark mode asking before anything is sent or deleted.

08:45 It looks only when you ask and screenshots are not stored. Built on the open-source Clickie, it is for anyone learning new tools. Install it free for Mac OS. Project number 10. Humalike X GTAP. AI NPCs that talk and remember. Humalike X GTAP brings AI powered NPCs to 5M role-play servers. It solves the empty world problem where meaningful roleplay stalls whenever other players are offline.

09:13 Each character has its own identity, personality, and persistent memory. So, it recognizes players, builds separate relationships, and reacts to conversations, clothing, masks, weapons, gunfire, and in-game events, deciding what to reveal, hide, or lie about. Static characters hold locations as informants or guides, while dynamic ones roam and can be stopped for a chat, and you can wire them into scripts as EMS patients, witnesses, hostages, or informants.

09:44 Connected characters share group memories, so one interaction can shift how a whole gang or faction sees a player. You design names, voices, knowledge, and appearance on one no code panel. Add the resource and deploy. Working with ESX, QBSOR, Qbox, and custom frameworks. It is built for 5M server owners get access free. Project number 11, GPT6 Soul and Luna.

10:05 Cheaper GPT6 models for everyday work. GPT6 Soul and Luna are two OpenAI models that bring Frontier intelligence to everyday work at different balances of capability and cost. They solve the mismatch where the most demanding projects need the flagship model, but routine tasks do not, yet still deserve strong results. Trained with methods similar to GPT6 Astra, they carry its gains in professional work, factuality, coding, computer use, and communication style into faster, more affordable models with sole tuned for

10:40 harder tasks and Luna for high volume, lowcost use. Both make far fewer factual mistakes than their predecessors and support adjustable reasoning effort. Improved prompt caching lets agents reuse more context and read cache tokens at a deep discount so long conversations run cheaper. They are available in chat GPT work and codecs and in the API as GPT6 soul and GPT6 Luna.

11:05 They are built for professionals and developers. Try them in chat GPT or build on the API. Project number 12, Superhum Go. AI assistant that works where you do. Superhuman Go is an AI assistant and agent platform that works inside the apps where you already write and work. It solves the tab switching of ordinary chat bots where you leave your task, reexplain context, and paste answers back.

11:29 Rather than waiting for a prompt, Go reads what is on your screen, and surfaces suggestions in line as you type, sorts your inbox and pre-drafts replies in your voice, and preps meetings and follow-ups. You curate a team of agents, including Grammarly, an email assistant, a knowledge checker, and a daily brief, or build your own with no code to run as you type, on a schedule, or after an event.

11:53 It grounds its help in your data through pre-built or custom MCP connectors and shows up as an underline, a side chat, a cursor menu in Slack, or the Go app. You control whether your content trains its models. It works on Windows, Mac, Chrome, Edge, iOS, and Android. It is built for professionals and teams. Try go free. Project number 13, Harmony AI service desk and Slack and Teams.

12:18 Harmony is an AI native ITSM and ESM platform whose agents resolve IT and HR requests inside Slack and Teams. It solves the ticket backlog and long weights of legacy service desks where routine requests pile up on human staff. Employees ask in plain language where they already work and the AI service desk agent handles most requests on its own from application access and password resets to onboarding and offboarding.

12:45 While proactive agents catch issues like low storage or expiring licenses before anyone files a ticket, it generates a knowledge base from tickets and context. Keeps a real-time inventory of assets and software and offers over a 100 ready-made agents plus a no code builder. Security runs in four layers with policy guard rails, human approval for sensitive actions, deterministic code, and lease privilege scoped access.

13:09 Over a 100 integrations. It is built for IT, HR, and enterprise teams. Book a demo. Project number 14. Hemorrhy. Always on memory wired into your agents. Hemorray is an always on listening app that turns everything you live through into a private, searchable memory your AI agents can use. It solves the way assistants forget what actually happened in your meetings and conversations, leaving you to reexplain context.

13:33 You start listening in manual mode or on a schedule, and what it hears settles into a timeline. Then you connect it to your agent over MCP in seconds. Once linked, agents search that memory to answer questions and generate real work like a monthly report deck from the month you discussed. A nightly journal or a full PRD from customer interviews it overheard.

13:55 Voice activity detection counts only speech so all day listening stays cheap. Audio is never stored in the cloud, processed as a stream and destroyed with raw audio kept only on your device. It works with clawed code, codeex, Gemini, Cursor, and other MCP clients on iOS and Android. It is built for professionals. Download Hemory and start your memory palace.

14:18 Project number 15, Niklira, conversational voice and video AI for apps. Ecle is a platform for building conversational AI agents with native voice, camera, and screen sharing. It solves the slow glued together stack teams otherwise build to give an agent realtime speech and sight. Voicetovoice means the agent replies at conversational speed and real-time vision lets it watch a live camera or shared screen while it talks all through one birectional session.

14:48 It reads text and frame through OCR recognizes objects in packaging and processes video at up to 30 frames a second under a sub800s latency budget. You describe the job in plain language in the builder or work through a versioned REST API. Then connect your own APIs, MCP servers, and over 3,000 apps with no custom integration code and deploy over telephony or a web widget.

15:12 It suits AI receptionists, answering services, lead qualification, and customer service. It is built for developers and businesses. Start building free. Project number 16. Good socials. AI social media manager for LinkedIn. Good Socials is an AI social media manager that writes and publishes your LinkedIn posts. It solves the cost and inconsistency of posting where a human manager runs thousands a month and most weeks still slip.

15:39 You sign in with LinkedIn and paste your website and it fills a board with drafts in about 90 seconds each grounded in one of three things. Market research, a deep dive or your own numbers read from connected tools like Stripe, GitHub, Post Hog, and Notion. It avoids generic filler and reports facts instead of selling. You approve posts on a board and it publishes five times a week.

16:01 And a note you leave on any card, rewrites it and becomes a lasting rule in your voice. So post sharpen week by week. It reads your data sources read only and posts as you with permission. It is built for founders, consultants, operators, and agencies. See your first post free. Project number 17. Pixver R2. Realtime world model. You can explore Pixver R2 is a real-time world model for continuous interactive video generation.

16:30 It solves the limit of ordinary AI video where a model returns a fixed clip and stops. Instead of a finished clip, R2 generates a world that keeps running while you are in it. Navigable in real time, holding its state as the session continues rather than resetting between actions. You enter an audiovisisual world and guide it with movement prompts, references, and audio.

16:50 moving a character with WD keys and steering the camera with the arrow keys. While prompts alter the terrain, characters, or how you move, its key change is that input persists. An earlier action stays true later. So, how a character behaves or a situation resolves builds on what you did. Characters keep a consistent identity and memory, responding live with voice, expression, and movement.

17:13 It is built for creators making interactive films and playable worlds. Enter a world and start exploring. Project number 18, Bleets Network. Startup agents pitch VC agents. Bleets Network is an agentto agent fundraising and scouting marketplace where a startup's agent pitches investors agents. It solves the warm intro lottery and spray and prey of raising where founders burn weeks on funds that were never a fit.

17:38 You drop in a deck or a few lines and it builds an agent that says only what you gave it. Then searches several thousand VC funds and keeps those matching your stage, geography, and thesis. Your agent pitches each short-listed fund and answers its questions over up to 10 rounds. And on a yes, you get the real contact people, while a no comes with the actual reason.

18:00 You read every conversation to see where interest dropped and sharpen your deck. Every fund starts with an agent built from public data, clearly labeled unclaimed until someone at the fund verifies and takes it over. It is built for founders and VCs. Build your agent free in beta. Project number 19, gas. Quiver GTM developer marketing run like an engineering system.

18:22 Quiver is an agentic developer marketing system that gives marketing the architecture engineers expect. It solves the way marketing reaches technical founders as vibes and disconnected tactics where the positioning sits in one chat, the plan in a dock, and performance vanishes before the next decision. It keeps product context, research, content, and results in one version source of truth.

18:46 Moves work through explicit states from draft to live and connects each shipped piece to the outcome it produced. Agents propose context updates, but a person approves what becomes true, so learning from one cycle sharpens the next. Purpose-built strategy, create, feedback, and analyze sessions run inside it. Or you connect an external agent over MCP, and a content API serves approved work as JSON.

19:09 You bring your own anthropic, open AI, Google, or open router account. It is built for technical founders and developer marketing teams, hosted or self-hosted under MIT. Start a trial or self-host on GitHub. Project number 20, House AI. Get your brand cited by AI. House AI is an AI visibility platform that tracks where your brand appears in AI answers and works to get you cited.

19:34 It solves the new discovery problem where buyers ask Chat GPT, Gemini, or Perplexity what to buy and get a short list that most brands never make. It tracks the full prompts your buyers ask across five AI surfaces, including Google AI overviews, refreshed every few days, and reports your visibility, sentiment, and share of voice against each competitor.

19:56 It traces every answer back to the pages it cited, so you see which review sites, forums, and wikis the models trust and aim your outreach there. Then a content agent turns each gap into a scheduled article in your voice and publishes straight to WordPress, Shopify or NextJS across many languages. Agencies run it across clients with white label reports. It is built for SAS, e-commerce and agencies. Scan your brand free in 30 seconds. Thanks for watching. See you in the next