Top AI Agent Projects : Fluree AI, Pushary, Notte, OpenCode Superapp & Pulse 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 see what matters in one place. You will discover trending AI tools and projects worth your attention without wasting time. Let's get started. >> Before we jump into today's project updates, here's a quick announcement for everyone.

00:23 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. Subscribe now to get weekly videos, in-depth guides, and real-time project breakdowns.

00:49 The link is right there in the description. Don't miss it. All right, let's get into today's video. >> Project number one, Flory AI, trusted knowledge graph layer for enterprise AI. Flory AI is a hosted enterprise data platform that turns scattered data into a trusted, queryable knowledge graph. It solves the problem of AI agents reasoning over messy, disconnected data they cannot verify.

01:13 You drop in anything, including PDFs, databases, spreadsheets, documents, and images, and Flory connects the entities and relationships automatically, giving your AI a governed place it can trust. You then ask questions in plain language, build dashboards and apps, and deploy agents that watch for events and act within your policy, staging fixes for approval with every step logged and reversible.

01:36 Permissions are enforced on every request, and answers trace back to each fact. Claude, OpenAI, Gemini, Ollama, or any MCP speaking agent reasons over the same graph built on the open-source FloryDB. It is built for enterprise data and AI teams. Start free and make your data AI ready. Project number two, Pushory, approve your coding agents from your phone.

01:58 Pushery is a control panel for AI agents that sends their permission requests to your phone. It solves the wasted time when an agent freezes mid-task waiting on a single yes while you are away from the desk. When an agent hits a step it should not take a loan. Your phone buzzes and you approve, deny, choose an option, or type a reply from the lock screen and the work continues.

02:19 You set the rules once so safe reversible steps run on their own and only risky actions like deleting, deploying, or spending money reach you. It works with Claude Code, Codex, Cursor, Gemini CLI, WinSurf, and any MCP agent. And the same question can land in Slack or in any third-party workflow. Your source code stays on your machine since only the question travels and every decision is logged.

02:41 It is built for developers running agents. Set it up in one command and start free. Project number three, Harness Router. One API to run agents inside your app. Harness Router is a back-end service that brings coding agents into your own product through a single API. It solves the months of work needed to build agent infrastructure yourself including sandboxes, runtimes, sessions, streaming, files, permissions, and cost controls.

03:08 Your app sends a task, Harness Router runs the agent, and finished reviewable work comes back to your interface so your users never open a terminal. Results arrive structured and renderable as code diffs, generated files and documents, images, or confirmations of real tool actions in GitHub, Slack, or Notion. You name the harness and config choosing Codex, Claude Code, or Hermes and swap between them with one line while your integration and output stay the same.

03:33 To wire it up you drop its instructions file into your coding tool, describe the feature, and add your key. It is built for product developers shipping AI features. Get an API key and build in minutes. Project number four, Free Solo Flash. Agent-driven post training for small custom models. Free Solo Flash is a managed post training package that your coding agent drives to produce a specialized model.

03:56 It solves the trade-off where small models are cheap and fast but lose quality, while frontier models are costly to serve on routine tasks. You point an agent like Claude Code, Cursor, or Codex at it, describe what you need, and approve one clear fixed price quote instead of a per token meter or GPU hour guesswork. It runs supervised fine-tuning and reinforcement learning on your data, so a sub-10B model tuned on your task can beat a frontier model on it cheaply enough to retrain often.

04:23 Every run returns downloadable weights in standard formats that you own and serve anywhere. Your data stays isolated and encrypted and pinned configs and seeds keep runs reproducible. It deploys as an OpenAI compatible endpoint. It is built for engineers shipping production agents. Point your agent at flash and train your first model. Project number five, Prost.

04:44 Turn your published archive into a book. Prost is an AI book service that assembles a creator's existing content into a publish-ready book. It solves the barrier that a book normally means writing from scratch, hiring a costly ghostwriter, and waiting months. You paste a newsletter, podcast, or video feed, or drop in transcripts and notes, and its pipeline, called the Inkwell, transcribes everything and structures it into a full manuscript using patterns drawn from many bestselling nonfiction books.

05:11 You then read every chapter in your portal, leave comments, and request rewrites, while a built-in reviewer checks voice, flow, and cohesion with each note linked to the exact passage. A source map color-codes every paragraph so you see which words are yours and which are added connective tissue. You export a print-ready PDF, EPUB, or DOCX and sell it directly.

05:30 It is built for creators who already publish. Count your archive and start assembling your book. Combat Cypher and Doel Adristics. Come at file it so I will do a do a testicle. Project number six, Prompt QL. Multiplayer AI with a shared team brain. Prompt QL is a multiplayer AI agent that gives a whole team shared threads and a shared brain. It solves the second job nobody wants, keeping context current, so AI answers reflect how your business actually works.

05:58 It starts from what is already scattered across Slack, docs, tickets, CRM, and warehouse tables. And on every task, it shows its work, listing the sources it pulled and the assumptions it made. You correct it once or pull in the teammate who knows, and that correction becomes shared context, a reusable skill, a piece of team knowledge, or a fix to the semantic layer.

06:20 Corrections turn into suggested edits to a cited Wikipedia-style wiki with revision history and scoped access. So, knowledge compounds from real work instead of rotting. It is built for teams across data, support, and go-to-market. Point it at your context and get started. Project number seven, Ask Cody, one gateway to many AI models for developers.

06:39 Ask Cody is an AI coding platform that gives developers one gateway to many large language models. It solves vendor lock-in and juggling separate subscriptions, since you pick the right model per task through a single account. You chat with models like GPT, Claude, and Gemini to generate code, write tests, refactor, debug, and document, and compare their answers side by side.

07:00 You build custom agents from a declarative config that bundles prompts, reasoning modes, guardrails, and tool integrations, then reuse them across chat, IDE, and API. An OpenAI-compatible API works as a drop-in by changing the base URL, and it plugs into editors like VS Code, JetBrains, Cursor, CLion, and Continue. A UI builder generates React components, and usage runs on tokens with real-time analytics, so you pay for what you use.

07:25 It is built for developers and small teams. Start free and pick your model. Project number eight, Moxie Docs, code-based docs that stay true to code. Moxie Docs is a documentation service that generates code-based docs from your GitHub repo and keeps them true to the code. It solves documentation drift, where docs quietly fall out of sync as code ships faster than anyone can write about it.

07:45 You install a Scope GitHub app on the repos you pick, and the first index produces architecture overviews, module walkthroughs, and convention guides, each cited back to source files. On every merge, it rechecks affected pages, flags gaps on the pull request while the fix is cheap, and regenerates stale docs as reviewable docs-only pull requests that never touch your code or auto merge.

08:07 It also exposes an MCP server, so agents like Claude Code, Cursor, and CodeX pull verified conventions instead of recrawling the repo, cutting wasted tokens. A weekly Friday cleanup batches small fixes into one pull request. It is built for engineering teams shipping fast. Connect a repo and run the first index. Project number nine, Wise Pro, voice to polished text on Windows.

08:28 Wise Pro is a desktop voice-to-text assistant for Windows that turns natural speech into clean, ready-to-use writing. It solves the slowness of typing and the cleanup that ordinary dictation leaves behind. It runs quietly in the system tray, and a hotkey drops text straight into whatever app you are using, with no switching or copying. It offers three modes.

08:49 Basic captures every word exactly as spoken. Smart cuts filler and rambling into polished prose. And command turns a spoken instruction like "Draft an email to John" into finished writing on the spot. It recognizes the app you are typing into and adapts tone and formatting, learns your names, acronyms, and jargon, and saves voice snippets for reusable It detects over 100 languages, keeps your keys encrypted locally, and does not store your recordings.

09:15 It is built for anyone who writes at a keyboard all day. Download it and start talking instead of typing. Project number 10, Notone, browser infrastructure that runs AI on the web. Notone is a browser infrastructure platform that lets AI agents operate the internet reliably at speed. It solves the messy stack developers otherwise assemble to run web automation in production.

09:37 Browsers, auth, proxies, and deployment glued together by hand, it gives you managed cloud browser sessions that scale to thousands of concurrent instances. AI browser agents you drive with a plain language goal in a few lines of code, and serverless browser functions that run your logic next to the browser. Agent Vault store credentials encrypted and fetched at runtime.

09:58 Agent identities supply real emails and phone numbers to pass OTP and 2FA, and session profiles save login state across runs. It handles captchas and residential proxies, is model agnostic across OpenAI, Anthropic, and Gemini, and drops into Playwright, Puppeteer, Selenium, or browser use with one line. It is built for developers automating the web.

10:18 Sign up free and run your first session. Project number 11, Battery Autopilot, AI that runs your Mac's battery. Battery Autopilot is a macOS app whose AI runs your Mac's power and charging on autopilot. It solves the slow wear that comes from leaving a laptop plugged in at full charge, and the fact that manual battery care never sticks. It watches your power app and charging patterns on device, then tries power saving actions and measures the real wattage change on your own Mac, automating only the ones that prove they

10:47 help, while leaving your foreground apps untouched. It guards charging at 80% through the Mac's power controller, self-heals if the system overrides it, and keeps a fail-safe that charges when the battery runs low. It also learns when you usually unplug and tops up to full just before you leave. A language model helps plan strategy, but only decisions that pass local safety guardrails are applied, and daily results show as extra minutes gained.

11:09 It runs on Apple silicon Macs and keeps learning on device. Download it free and see the minutes it adds. Amrit Minda may call my eating your room and yambic H mod. Project number 12, Open Code Superapp, local first AI coding workspace for Mac. Open Code Superapp is a native Mac workspace built on the open source Open Code runtime. It solves the tradeoff between the power of a cloud coding agent and keeping your projects and model traffic private and local.

11:35 You run cloud, self-hosted, or fully local models, switching provider per task, while your credentials stay in your own open code configuration. And an integrated Ollama engine discovers models already in your library over a private loopback endpoint. Projects, parallel sessions, files, Git, work trees, terminals, and search sit together in one place, and you can add commands, right panel apps, skills, MCP services, and your own widgets that the agent helps you build.

12:02 A paid one-time unlock adds browser automation, native macOS computer use, private dictation, real-time voice, and encrypted remote to continue a session from your phone, with each consequential action permission gated. There is no account, and data stays on your Mac. It is built for developers who want local control. Download it free and start building.

12:23 Project number 13, Kazaira 2.0, cloud coding terminal with agents on your phone. Kazaira is a mobile cloud terminal for iPhone and Android, with coding agents pre-installed. It solves the problem that your dev environment is stuck on a laptop you are not carrying. So, work waits until you sit back down. On first launch, it provisions an isolated Ubuntu container in the cloud that is yours alone, with cloud code, Code X CLI, open code, and Gemini CLI already installed alongside Node, Python, Git, tmux, and Vim.

12:52 You paste your own provider key once, then point an agent at your repo, and watch it edit, test, and ship from a real shell on your phone. Because the agent runs in the cloud, not on your hardware, a task started at your desk keeps going when your laptop sleeps, and pro sessions hibernate when idle and resume with state intact on any device. Your code stays in your container, and keys are encrypted on device.

13:14 It is built for on-call engineers, commuters, and laptop-free developers. Install it and ship your first change in minutes. Project number 14, Basement AI my sidekick on every product page. Baseman is a free shopping browser for iPhone and a Chrome extension with an AI shopper on every page. It solves the tab hopping ritual of checking whether a price is fair and a seller is legit before you buy.

13:38 Open any product page and your own AI called a baseling reads it, hunts for a lower price, and watches for restocks while the real takes from Reddit, Trustpilot, X, and Blue Sky for that exact page are stacked in the header. Other buyers are in the room with you, so you can talk it out and loop in a friend before you commit. A live score weighted by the whole room reads whether a listing is trustworthy.

14:00 Once you approve, the baseling places the order and keeps you stocked on a schedule, and your closet gathers every purchase from across the web in one place. It is built for collectors, resellers, and anyone who refuses to overpay. Download it free and shop with the room. Project number 15, Framer, design agent that builds your site live. Framer is an AI website builder with design agents that work right on your canvas.

14:23 It solves the scattered process of designing a site, wiring up content, and writing code in different tools. You type what you want, and the design agent creates and refines it in place, keeping every change visible, editable, and easy to undo. A CMS agent structures your collections, so content and layout stay in step, and a code agent adds custom effects and interactions from a description.

14:45 You can also steer your site from outside tools, letting Claude code, Codex, or Cursor push changes from Slack, the terminal, or a pull request. Hosting, SEO, localization, analytics, and AB testing come built in, so one place carries the site from idea to launch. It is built for designers, teams, and creators who want polish without stitching tools together.

15:05 Start free and publish your site. Project number 16, Base Dash AI Kit, AI business intelligence grounded in governed metrics. Base Dash is an AI native business intelligence platform that turns plain English questions into trusted charts, dashboards, and reports. It solves the slow SQL heavy setup of traditional BI and the risk of AI answers that quietly get the numbers wrong.

15:29 You describe what you want to track and it builds a dashboard in minutes. Or you ask about your business and get an answer traced back to real data. A built-in semantic layer lets teams define each metric once as approved SQL. So chat, charts, dashboards, and automations all reference the same govern logic instead of drifting. It connects to over 750 sources including Postgres, Snowflake, Salesforce, and Stripe or uses its own warehouse and exposes an MCP server so any AI client can reach your data.

16:00 Access controls, audit logs, and single sign-on are built-in. It never trains on your data and it runs in the cloud or inside your own network. It is built for teams that want fast, reliable answers. Start free and ask your first question. Project number 17, Megaphone, free, private, on-device dictation for Mac. Megaphone is a free, open-source Mac dictation app that runs entirely on your device.

16:24 It solves the privacy and polish gap in voice typing where ordinary tools send your audio to the cloud and still leave messy text. It uses Apple's speech analyzer to transcribe live as you speak. Then on-device foundation models clean up filler, fix wording, and format the result at your cursor in any app. Smart cleanup turns rambling speech into finished sentences.

16:43 A personal dictionary keeps your names, products, and technical terms spelled right. And screen awareness reads the active app locally to match tone. Staying concise in Slack and formal in mail. Saying, "Hey Megaphone, passes your instruction and context to the local model." So you can ask for a rewrite or bullet points with no API key or cloud request.

17:00 It detects several languages and learns the choices you repeat. It is built for anyone who wants private, hands-free writing. Download it and start talking. Project number 118, Agent Loop, unattended coding loops on your machine. Agent Loop is a local tool that runs coding agents in an unattended loop until the work meets your standards. It solves the tedious job of being the relay between planning and code, passing messages from planner to worker to reviewer and back.

17:27 You plan once in chat GPT, send Agent Loop a bounded goal, and write your definition of done in a guidelines file. Each cycle of fresh Codex worker starts with clean context, edits the files where your code actually lives, and runs your tools. Then an independent critic in a separate process grades against your rubric. Failing feedback carries forward into the next cycle, and passing tests count as evidence rather than permission to stop, so the critic still catches defects a green test suite misses.

17:54 A dashboard shows every cycle, verdict, transcript, and cost, and you can check status from chat GPT over MCP. It runs on your machine with Git and the Codex CLI, is open source, and keeps its engine pluggable. It is built for developers who want to delegate work they would otherwise baby sit. Clone it and set your first goal. Project number 19, Blacksail Agent Drive, shared cloud file system for AI agents.

18:17 Agent Drive is a distributed cloud file system from Blacksail that lets AI agents collaborate by sharing files across sandboxes in real time. It solves the duplication and orchestration overhead agents hit when they need the same context, data sets, or dependencies. You create a drive in one call and mount it into any sandbox as an ordinary path. Then multiple agents read and write the same files at once.

18:40 A POSIX compliant fuse client built in house makes it behave like local storage, and it is tuned for the small file high concurrency pattern agents actually produce, rather than repurposed object storage. Storage grows automatically with no capacity planning. Agents hand off small JSON artifacts and notes, share long-term memory that a new sandbox can resume from, cache code dependencies once for instant reuse, and fan work out to parallel workers that collect results in one place.

19:06 It is built for teams running fleets of agents. Join the private preview and mount your first drive. Project number 20, Hulse, permission-aware company brain for teams. Hulse is a company brain that remembers your team's decisions, commitments, and hard-won lessons, then answers in plain English. It solves the daily task of knowledge scattered across 10 tools and one person's memory, re-explained to every AI from scratch.

19:31 It connects to the stack you already run, including Slack, GitHub, Notion, Linear, Drive, and Jira, and reads everyday work into a connected graph rather than a flat search index. So, you go straight to a decision and the reasoning behind it. It mirrors your existing permissions, showing each person only what they could already see, and every answer sites its sources and marks a confidence score tuned weekly from your own feedback.

19:55 It surfaces stuck work and stalled decisions before you ask, and drafts the next step like a ticket or a message into an approval inbox that never sends without you, and gives 5 minutes to undo. It works inside Claude, ChatGPT, and Cursor through an MCP server. It is built for teams of 5 to 500. Try the live demo with no sign-up. Thanks for watching. See you in the next video.