LongCat-2.0, Mira, Katalyst, AI Emaily, Dupely, Ogment AI, and More Launches

Mia: Welcome to ProductHunt Daily from the Bri podcast. I’m Mia.

Milo: I’m Milo.

Mia: Today we start with two big hardware claims: a massive open-source AI model trained without Nvidia, and an AI researcher that reads faces, voice, and eyes in 70 languages.

Milo: From there we move into self-editing screen recordings, AI agents inside your inbox and pipeline, and a new attempt to make brands visible to chatbots.

Mia: A startup just launched an open-source AI model that flips one of the biggest assumptions in the industry.

Milo: Say more.

Mia: It’s called LongCat-2.0. They claim it was trained entirely on custom AI ASICs—not Nvidia GPUs. It’s a 1.6-trillion-parameter Mixture-of-Experts model with 48 billion active parameters and a context window of one million tokens. And it’s released under an MIT license.

Milo: So this isn’t a small experiment. You hear “trained on non-Nvidia hardware” and you think maybe it’s a tiny model. A trillion-plus parameters is the opposite of that.

Mia: Exactly. The scale is what makes the claim worth watching. If this works at a reasonable cost, it starts to loosen Nvidia’s grip on the supply chain for frontier models. For now it’s a Product Hunt launch, not a peer-reviewed benchmark. The performance data isn’t public.

Milo: Right. We have the claim. We don’t have the proof. But it’s the sort of claim that moves money.

Mia: Now another launch that makes a big promise about reading people. Mia, you pulled this one.

Milo: This is Mira. It’s described as a full AI researcher—not a transcript generator. The pitch is that it reads faces, voice, and eyes to assess how a person feels during an interview or research session. It supports 70 languages.

Mia: So it’s multimodal, analyzing expression, tone, and eye movement together. And they’re not just claiming it works in English. Seventy languages is a huge breadth claim.

Milo: That’s the headline. But the Product Hunt listing gives no accuracy numbers, no bias data, and no detail on how it handles different cultural expressions or skin tones across that many languages.

Mia: So the gap between the promise and what we can verify is wide. Emotion inference from faces and voice is hard even in controlled settings. Applying it across 70 languages without published validation is a big leap.

Milo: It’s a signal to watch, not a product we can assess yet.

Mia: Let’s move to two AI agents that aim straight at repetitive business workflows.

Milo: The first is Katalyst. It’s pitched as an AI sales agent for teams on Salesforce. It handles post-call pipeline updates and automates sequence tasks across the pipeline. And the second is AI Emaily, which calls itself an AI-native inbox—a chief of staff for email that writes in your voice and can send replies on autopilot.

Mia: Both went up on Product Hunt in the same window. And both are positioned as agents that act, not just assistants that suggest. Katalyst touches a live pipeline. AI Emaily replies as you. The trust bar there is high.

Milo: The listings don’t tell us much about reliability though. No independent benchmarks, no error-rate data, and nothing specific about guardrails or human review loops.

Mia: That’s the recurring thread today. These tools promise to take over high-stakes desk work, but the early signal doesn’t yet answer how often they get it wrong.

Milo: there’s an interesting pair around trust and coordination.

Mia: Badge and Kadoink. Badge uses an AI agent to collect peer reviews and generate a proof-of-work score. The idea is to make professional reputation legible without chasing references. Kadoink does something else entirely: it’s a tool to instantly gather people by AI-generated text, video, or by ringing their phones.

Milo: So one automates proving you’re reliable; the other automates getting people to show up. The timing here is a coincidence—both popped in the same 24-hour Product Hunt window. There’s no connection beyond that.

Mia: Both are early-stage launches with minimal information. No user reception data yet. The question underneath is whether AI-mediated trust and AI-brokered gatherings feel genuine enough to catch on.

Milo: We’ve got one more trust tool, but this one sits in your browser.

Mia: Dupely. It’s a browser extension that bills itself as the trust layer for online shopping. Instead of hunting for coupons, it flags fake price drops. The tactic is called price anchoring: a seller inflates the “was” price to make the current discount look deeper than it really is.

Milo: Dupely detects that automatically. And it has a second feature: it finds identical products sold cheaper elsewhere so you’re not overpaying for the same item.

Mia: That’s a specific, hidden cost of e-commerce. You think you got a deal, but the reference price was manipulated. And you might not realize the exact same product is listed for less on another site.

Milo: One caveat: this is a Product Hunt listing, not an independent validation of how accurate the detection is.

Mia: there’s another Slack-native AI launch on the board.

Milo: Ogment AI. They want you to tag “O” in Slack like a colleague. The pitch is that it remembers decisions across the team and connects to a thousand tools. So past answers persist for the whole group.

Mia: That shared memory is what sets it apart from a one-shot Q&A bot. If the claim holds, it turns an assistant into persistent team infrastructure. But the only source is the maker’s own Product Hunt description. No independent review. No user verification of the thousand-tool count.

Milo: So for now it’s a pitch, not a proven setup. But the direction is clear: AI moving from answering prompts to being part of how teams remember.

Mia: A new startup is tackling a problem that didn’t exist a few years ago: AI-engine invisibility.

Milo: Scribble Network. Their argument is that customers are asking AI tools like ChatGPT before they go to Google, and most brands are completely invisible in those AI answers. Their response is a network of 50,000 creators who get paid to produce content that mentions brands, so AI models pick it up as source material.

Mia: The tagline is “The product that makes AI recommend your brand.” It’s a pay-to-be-cited model, which sits somewhere between influencer marketing and the old SEO playbook.

Milo: And it raises the same authenticity questions SEO did. If brands are paying to get inserted into the content that trains AI responses, how organic are those recommendations? No standard playbook exists yet, so Scribble is testing one.

Mia: A different kind of data play showed up alongside it.

Milo: Social Fetch. It’s a pay-as-you-go scraping API that targets TikTok, Instagram, YouTube, and more. Public profiles, posts, comments, videos. All from a single API.

Mia: Scraping walled gardens is a gray zone. Terms of service and legal enforcement can shift overnight. And this is a single Product Hunt launch—no independent testing, no uptime data, no legal review. So the promise of unified access is real, but the risk is unquantified.

Milo: Let’s pull one more thread: capturing notes from conversations that happen in person, not on Zoom.

Mia: Ellis. The tagline is “AI notes for in-person meetings.” It uses only your iPhone or Apple Watch for recording and produces a clean transcript. No extra hardware.

Milo: So it shifts AI notetaking from virtual rooms into the physical world. The open question is microphone quality and how well it handles multiple speakers in a noisy space. No published accuracy data yet. But removing the gear barrier is notable.

Mia: Switching to a free macOS tool that saves editing time.

Milo: Glideo. It’s a free screen recorder that auto-zooms toward your cursor clicks while you record. The result is a polished demo without manual zoom-and-pan editing. They’re targeting product demos and dev tools.

Mia: That’s a friction remover. If you make quick demo videos, the auto-focus on clicks means you don’t spend time in post-production tweaking frames.

Milo: Last one: a new entry in the no-code spreadsheet-database space.

Mia: Zoho Tables. It’s described as a hybrid with AI brains for organizing work and data. The promise is database structure without coding, plus some workflow automation. Details on the AI capabilities are thin in this listing, so how it differs from established tools remains unconfirmed.

Milo: It rounds out a launch day where most tools are making big promises—new hardware, new agentic behavior, new trust layers—but the proof is still early.

Mia: That’s the episode for today. Thanks for listening.

Milo: We’ll be back next time with another round of what’s launching.