Top AI Agent Projects : Clears, TinyFish, Attyn, Kane CLI & Human Behavior 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, web automation, and analytics without wasting time. Let's get started.

00:22 >> 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, clears turns your backlog into reviewed pull requests. Clears is an autonomous execution layer that carries software delivery from stories to reviewed pull requests across your repos.

01:10 It solves the backlog drag that keeps engineers on routine tickets instead of the work they want. Background flows run continuously to root cause, scope, and assess risk. So thousands of raw tickets become decisions handled on their own or escalated to your team. A task board wires every ticket to its live agent session. So you watch it go from requirement to merged pull request.

01:33 Steer when it drifts and parallelize the whole backlog. A shared context layer gives every agent the same memory so there are no cold starts and over MCP your terminal commands the whole board. It is built for engineering teams. Start a free trial and clear your backlog. Project number two, same page artifacts connected writing surface for product managers.

01:54 Same page artifacts is a contextaware writing surface where a product manager's documents draft themselves. It solves the hours PMs lose writing PRDs, release notes, feature briefs, and weekly updates from scratch across scattered tools. An AI agent that knows you and carries your full product context, generates drafts and has them waiting, and signals surface important insights you turn into an artifact in a click.

02:19 You write with AI built into the surface, start from your own data or a blank page and publish straight out as feature requests in linear or documents in Confluence. It connects natively to tools like Slack, notion, and ASA and processes your data privately under GDPR and CCPA. It is built for product leaders keeping their org aligned. Sign up and let your drafts write themselves.

02:43 Project number three, Bulb Things AI co-pilot for asset tracking and maintenance. Bulb Things is an asset management platform with an AI co-pilot that tracks, maintains, and manages your equipment. It solves the scattered spreadsheets and lost maintenance requests that make asset tracking a chore. You build an inventory in a few clicks. Add photos, locations, and documents, and label gear with QR codes your phone scans to look up or update on the spot.

03:09 It automates booking and checkout workflows, centralizes repair requests from start to finish, and schedules recurring maintenance and inspections with smart reminders based on usage and history. The Copilot works in plain language across an asset or the whole workspace. So, you snap a photo to add an asset, make a booking, log a repair, or pull a report, and it stays optional and fully controllable.

03:31 It is built for small teams managing inventory, maintenance, or fleets. Sign up free and add your first asset. Project number four, Tiny Fish. Everything AI needs to use the web. Tiny Fish is enterprise web infrastructure that gives AI agents everything they need to use the live internet from one API. It solves the messy stack teams otherwise build to search, read, and act on pages that constantly change and block bots.

03:59 Four products share one key. Search returns fresh, structured results rendered from real pages. Fetch turns any URL into clean markdown or JSON. Browser runs stealth cloud sessions that get past login walls and antib-bot checks. An agent navigates, fills forms, authenticates, and returns structured results. The platform decides which to use per task.

04:22 So you write no routing code and it finds elements by meaning rather than brittle selectors. So a site change does not break it. It is MCP native across claude cursor chatgpt and more. It is built for teams automating real web work at scale. Get an API key with free credits to start. Project number five. Framer. Build and run sites with design agents.

04:44 Framer is an AI website builder where design agents do the work on your canvas. It solves the scatter of designing, wiring content, and coding a site across separate apps. You describe what you want and the design agent creates and refineses it in place. Every change staying visible and reversible. A CMS agent structures your collections so content and layout move together.

05:05 And a code agent adds custom effects and interactions from a plain description. You can steer the site from outside too, letting claude code, codecs, or cursor push changes from Slack, the terminal, or a pull request. Hosting, SEO, localization, analytics, and AB testing all come builtin. It is built for designers, teams, and creators who want polish without stitching tools together.

05:28 Start free and publish your site. Project number six, Harness Router Community Edition. Self-hosted agent harness backend, one API. Harness Router Community Edition is open-source infrastructure that plugs coding agents into your product through one agent API on your own machines. It solves the months of back-end work each harness needs around the model call, sessions, streaming, files, artifacts, cancellation, and failure handling.

05:55 You pull one docker container and the gateway, runner, and console start together with no managed database, vault, or cloud account. So your provider keys stay in local SQLite and workspace files stay on infrastructure you control. You integrate once and pick codeex clawed code or hermas per task switching without rebuilding because they share an open versioned contract called the unified harness protocol that ships with runnable conformance checks.

06:26 A starter kit gives you working demos instead of an empty repo and you can move to the manage cloud later. It is built for developers shipping agent features. view it on GitHub and run the container. Project number seven, Ch AI video agents for FaceTime calls. Ch is a platform for building AI video agents that answer and place FaceTime calls in a few lines of code.

06:47 It solves the blind spot of voice only agents. When a problem is visual, another question rarely helps. Because it runs on video, the agent reads the caller's camera to see what they are showing and guide them one step at a time. So it fits remote support, field service, teleaalth intake, onboarding, and inspections. You draft the agents instructions, model, voice, avatar, and framing.

07:09 Test it in a browser preview to check behavior, then publish it to a provisioned FaceTime line for bounded inbound and outbound calls. The FaceTime product runs as a controlled private preview, and live calls need explicit authorization. It is built for developers adding video agents to their product. Book a demo or FaceTime an agent to try it. Project number eight, Cost logic, construction takeoffs, estimates, and invoices together.

07:35 Cost Logic is a browser-based construction platform that runs takeoffs, estimates, and invoices as one AI powered workflow. It solves the retyping and tool hopping that slows contractors from plan set to payment. You drop in construction PDFs, and pages are named automatically while the scale is read from the title block. Then you measure linear area and count quantities right on the drawing or let Auto Room detect rooms for you.

08:00 Those measured quantities flow straight into line items with markup, waste, tax, and deposits. And its built-in assistant, Onyx, finds rooms, drafts the estimate, and answers questions about your jobs. An approved estimate becomes an invoice in one click, emailed as a branded PDF with payments tracked and a CSV export for your bookkeeper. It runs on Mac and PC in the browser with no install.

08:23 It is built for contractors and estimators. Start a free trial and finish your first takeoff. Project number nine, GLM 5.3 open coding model for long horizon agents. GLM 5.3 is Z.AI's flagship openweight model built for agentic coding and long horizon work. It targets the way ordinary models lose the thread across many files and long tasks. It handles a 1 million token context, so it reasons over whole code bases in one pass.

08:54 And its reasoning is always on with low, high, and max effort levels you pick to trade depth against speed. Heavy post training exposed it to far more long horizon environments, simulating the full software life cycle from finding bugs to writing fixes, running tests, and shipping. And it does this while spending fewer output tokens than its predecessor.

09:14 Strong tool calling lets your app validate and run each requested action. You use it now through the GLM coding plan with agents like Claude Code, Klein, Open Code, and Codeex with the API and open weights following after safety review. It is built for developers running coding agents. Subscribe to the coding plan and start building. Project number 10, Big Mike texts you mispriced sports betting lines.

09:40 Big Mike is an SMS first sports betting agent that finds mispriced lines and texts you play. It solves the problem of watching odds all night and still missing the moment a number turns in your favor. It scans 12 sports books around the clock across NFL, MLB, college football, NBA, and WNBA. And when a price beats the true odds, it lands in your iMessage with the fair price, the EV percentage, and the reason it qualifies.

10:06 You reply to ask about any line and it answers like a person while the iOS app holds the full board, your record and your squad. Line moves, injury alerts, and start sit intel arrive before the number shifts and its whole graded pick history stays public to audit. It is built for sports betterers who want the edge where they already text big Mike and start 7 days free.

10:29 Project number 11, Zedi, personal intelligence agent that tracks everything. Zedi is a personal intelligence agent that reads the internet around the clock and briefs you only when something you named actually moves. It solves the doom scrolling and duplicate heavy feeds that bury the few updates you care about. You describe what matters in one sentence and a tracker goes live in seconds watching news, podcasts, blogs, GitHub, research papers, and newsletters with sources it finds plus any you pin.

10:56 When a story breaks, it folds many reports into one card that says what happened, why it matters, and every source, and it surfaces connected developments beyond what you follow. Its memory learns your preferences from plain language feedback and stays open for you to edit or delete. Briefings arrive as a feed, push, newsletter, RSS, or audio. It is built for anyone who wants to read less and know more.

11:20 Download the app and name your first tracker. Project number 12. It tin personal AI at your cursor. It tin is a personal AI for your Mac that works at the cursor. So you write, speak, and act inside the app you are already in. It solves the round trip of opening an AI tab, reexplaining context, and copying the answer back. Four ways in share one layer.

11:42 Inline assist rewrites selected text in place. Realtime dictation drops spoken words where you type. Screen assist answers about anything you circle on screen. And Blackboard draws a visual answer. Because it acts wherever a field accepts text, it needs no integrations and works across tools like Gmail, Slack, Notion, and Docs. You set up personal or business profiles with their own context, memory, and approval rules.

12:07 And pick the intelligence, a provider key, a TIN credits, or a local ondevice model for private everyday work. It is built for professionals, founders, creators, and students. Download it for Mac OS and bring AI to your cursor. Project number 13, Nenpace, an extended mind, not storage. Nspace is a personal thinking space that fuses an AI model with the room where you catch and cultivate your thoughts.

12:34 It solves the splitmost tools force. Filing cabinets like notion store but do not think. While chatbots think for you but barely remember. Here the model nen1 lives inside one space that holds your working memory, tasks, notes, habits, and log book. It runs a sift that reads your open thoughts and suggests where each belongs. And nothing moves without your okay.

12:56 On any page, you ask nen one about what is there, and it is tuned to question rather than flatter, pointing at what you wrote and handing the thinking back. It keeps your working memory light by holding the rest. It is built for people who want to return to their own thinking. Enter Nspace free on web or iOS. Project number 14, 9. Outcome turns your content into personalized lead magnets.

13:18 Outcome is an AI funnel builder that turns your content into a personal action plan for every viewer. It solves the flat lead magnet that teaches everyone the same thing and leaves each person to figure out how it applies to them. You paste a video, post, or framework, and it grounds every answer in your material. Then a short quiz surfaces each lead's goal, situation, and obstacle.

13:41 It runs your questions, knowledge, and instructions as chain steps. So the result reads the way you would advise one-on-one, delivered in seconds. You stack blocks like a written analysis, a score, a checklist, live web research, or a custom page. And each report ends on the next step you choose, whether a call, a course, or a community. Capturing the lead's email as it goes.

14:06 It is built for creators and coaches growing a list. Paste a link and build your first funnel free. Project number 15. Munder Difflin. Digital clones of you and your team. Mundifflin is a free open-source multi-agent harness that turns the CLI agent you already use into a digital clone that works around the clock. It solves the bottleneck of an agent that only runs while you sit at it, one task at a time.

14:28 You install a node that wraps agents like Claude Code, Codeex, Grock, or C-Pilot on your own laptop, and it learns your workflow, tooling, and knowledge. So, it reviews pull requests, fixes bugs, answers questions, and ships small work in your style. An orchestrator plans and routes tasks to specialized agents, each in its own isolated work tree, escalating only the calls that need a human.

14:53 On the team's plan, each teammate's clone runs on a dedicated cloud sandbox, and they hand off work over an end-to-end encrypted node to node network. Code and keys stay local, and it is MIT licensed. It is built for individuals and teams in any role. Download it free and clock in your clone. Project number 16, Min, the IM that builds itself. Min is a CRM that builds itself continuously from your meetings and emails.

15:20 It solves the data entry and stale notes that make ordinary CRM a chore nobody keeps current. You just do your work, take the calls, send the email, run the one-on- ons, and it sits invisibly on all of it, remembering the promise, the deadline, and the off-hand detail you would have lost with no bot joining your meetings. Every relationship stays distilled into a living capsule, always current with three sharp sections, where you stand, action items with deadlines, and the whole history.

15:48 and each line expands into its source. Before a call, you open it and get prepped in a minute. After you ask it what you are missing and let it draft the followup on what was actually said. It works with Gmail, Outlook, Calendars, and recorders like Zoom, Meet, and Teams, and your queries stay private. It is built for anyone whose work runs on relationships.

16:10 Download it for Mac OS or Windows and set up in 2 minutes. Project number 17, Bay Stash AI business intelligence that answers in plain English. Bay Stash is an AI native business intelligence platform that turns plain English questions into trusted charts, dashboards, and reports. It solves the slow SQLheavy setup of traditional BI and the risk of AI that quietly gets the numbers wrong.

16:34 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 governed logic instead of drifting apart. It connects to hundreds of sources like Postgress, Snowflake, Salesforce, and Stripe, and exposes an MCP server so any AI client can reach your data.

17:03 Access controls, audit logs, and single sign on are builtin, and it never trains on your data. It is built for teams that want fast, reliable answers. Start free and ask your first question. Project number 18. Con's Kain CLI. Natural language browser testing in your terminal. Con's CALI is a terminal native testing agent from Test Mu AI that runs endtoend browser flows described in plain English.

17:29 It solves the brittleleness of selector heavy playright and selenium scripts that break whenever a page changes. You type a request like verify the checkout flow on staging and it drives a real Chrome browser to do it with autohealing, smart waiting and dynamic querying builtin. So no scripts to maintain. It first reads your requirements, tickets and existing suites to map every promise your product makes.

17:51 Then measures coverage strictly, counting a rule proven only when a test asserted it and captured evidence. Every run ends in a portable tamper evident bit evidence pack with screenshots and a replayable trajectory. It works from your shell or CI and plugs into agents like Claude Code, Cursor, Codeex, and Gemini and can export native playright. It is built for developers, QA engineers, and coding agents.

18:14 Install it from npm and run your first flow. Project number 19, Ido AI code review that runs your app. Ido is an AI code review agent that builds and runs your app on every pull request. It solves the blind spot of reviewers that only read the diff and miss bugs that appear only when the code runs. When a PR opens, Edido reads the change and its description, focuses on the user flows it affects, and builds a fresh single-use copy of your app from source.

18:43 It then navigates the app like a real user, exercising the happy path, edge cases, and adversarial scenarios against your real backend, so it catches logic, data, authorization, and concurrency bugs that static tools cannot. Every finding lands in the PR with a video replay logs the exact lines responsible reproduction steps and a severity rating and pushing a fix reruns the test until it passes.

19:08 It is framework agnostic, needs no test scripts and runs in isolated encrypted sandboxes. It is built for engineering and QA teams. Connect your repo and start reviewing PRs. Project number 20, human behavior agents that read every user session. Human Behavior is a product analytics platform with agents that read the user sessions nobody has time to watch.

19:29 It solves the hours teams lose scrubbing replays and hunting for the session behind a bug. One SDK captures each visit and replay error tracking and agents all read from that same recording. Replays are real DOM captures with console network and errors on one timeline and inputs are masked in the browser before anything is sent. Every exception is fingerprinted and grouped with the breadcrumbs and a replay attached.

19:55 So there is no session ID to chase. You describe an agent's job in a sentence and it runs on a schedule reading stacks breadcrumbs and recordings. Then files the issue and post the finding cited to a moment straight to Slack instead of a dashboard. It is built for product and engineering teams. Book a call or view the demo to get started. Thanks for watching. See you in the next video.