0724 | Basement AI Shopper Checks Out, Memory Tools Launch, Solo Voice Apps

Mia: Welcome to ProductHunt Daily, part of the Bri podcast family. I'm Mia.

Milo: And I'm Milo. Today we're tracking an AI shopping browser that checks out for you, a pair of company-knowledge tools launching side by side, and three solo-built apps rethinking voice input on Apple devices.

Mia: We've also got open-source agent governance tools, a distributed filesystem for AI sandboxes, an AI podcast studio, a meeting copilot that stays invisible, an analytics toolkit, and a hardware-idea judge powered by GPT-5.6.

Milo: Let's get into it.

Mia: Let's start with the browser that does the shopping for you. It's called Basement, and it now includes agentic checkout.

Milo: So the AI shopper inside Basement finds the best price across the web and then completes the purchase. Every transaction runs on a single-use card that's scoped to exactly one order.

Mia: And that means your real card is never exposed, and nothing exceeds the budget you set. Payments land in one place, powered by Crossmint.

Milo: The browser also learns what you like from how you actually browse, and it keeps something called a Closet of everything you own. It's live now on iOS and Chrome.

Mia: The founder, Luis, says he built it because online purchases remain scattered across dozens of store logins, with tools that work for the store and nothing that remembers what you already bought.

Milo: Now a different kind of automation. Quaso is an AI agent that turns busywork into automations across all your apps and the open web. You just prompt it with what needs to get done and it builds and sets up the automation.

Mia: It connects to more than three thousand integrations, including Gmail, Slack, Linear, and Stripe. And when no native API is available, it uses a real browser to get the job done.

Milo: It runs on schedules, triggers, and on demand. You can fork templates or save any thread as your own reusable skill, and tag @Quaso in Slack so the whole team can use it.

Mia: Lucas, co-founder of the team behind Quaso, shared that they spent the past year building browser infrastructure for AI agents, and more than two hundred companies run their agents on it in production. They also released Anything API, which turns any site into a custom API.

Milo: Two products launched today tackle the same problem from slightly different angles: AI assistants that start every session with no memory of your organization.

Mia: The first is Pulse, built by solo founder Apoorv. He positions it as a permission-aware, proactive, agentic brain for a company. It connects Slack, email, docs, meetings, and twenty-five other tools into one searchable memory.

Milo: And here's the thing: every answer comes back with a citation on every line. The system captures decisions, commitments, and lessons as first-class items, rather than letting them stay buried in Slack threads.

Mia: Pulse also finds the right internal expert, runs agents only after user approval, builds reports from real data, and works by voice. It lives inside Claude, Cursor, and ChatGPT over the Model Context Protocol.

Milo: Apoorv emphasizes that he built this because existing company-knowledge tools often turned into employee surveillance, and because typical AI tools sound confident without providing anything verifiable.

Mia: The second product is PromptQL, created by the Hasura team. CEO and co-founder Tanmai describes it as the world's first AI-native workspace. It's multiplayer AI meant to replace Slack.

Milo: The product emerged by accident while the team was trying to build accurate AI that works with data and tools. They concluded that context is the bottleneck and that no single person on a team ever holds all the required context.

Mia: PromptQL provides shared threads with capabilities comparable to Claude or ChatGPT. Teammates can be tagged to review, correct, and refine answers without losing the underlying reasoning. It connects databases, SaaS apps, and coding environments.

Milo: Over on the desktop and mobile side, three new macOS and iOS apps take different approaches to on-device voice input. Each one was built by a solo developer responding to a specific friction they saw in existing tools.

Mia: First up is Megaphone. It's a free, MIT-licensed dictation app for Apple silicon Macs running macOS 26. The developer was prompted by a benchmark showing Apple's on-device SpeechAnalyzer API was more accurate and roughly three times faster than Whisper Small.

Milo: Megaphone runs completely on-device. No account, no API key, no subscription, no server. You hold the Fn key, speak, and the app types clean text into any Mac application. On-device Foundation Models remove fillers, fix self-corrections, adapt to app context, and handle inline voice commands.

Mia: Then there's Wispro, a Mac writing tool that turns spoken input into finished text across three modes. Basic Mode captures every word exactly as said. Smart Mode cuts filler words and rambling and formats the result into polished prose.

Milo: Command Mode turns a spoken instruction into a finished email, reply, or full draft. The developer built Wispro after trying several speech-to-text apps and finding that their style presets, things like a Slack tone or a Gmail tone, were fixed shortlists someone else wrote. Wispro instead lets users define their own tone and rules.

Mia: The third app is SMASH Voice to Invoice, and it targets tradespeople who work with their hands and dislike typing: cleaners, gardeners, plumbers. The developer built it after losing jobs because they had to stop working, remember rates, and type quotes.

Milo: SMASH lets a user speak quote details into their phone while still on site. It pulls from the user's saved rates and generates the invoice right there.

Mia: Finally, two recent open-source tools aimed at governing autonomous AI agents, but in very different ways.

Milo: HOL Guard is presented as a firewall for AI agents. It sits between agents and company systems to block high-risk actions, like deleting production data or exposing secrets, before they happen.

Mia: Michael Kantor, President of HOL, an open standards consortium with over twenty specifications contributed through LF Decentralized Trust and more than thirty-seven million transactions powered by its work, explained that the project was originally built for internal protection after HOL observed agents trying to bypass safeguards.

Milo: It's free, open source, and has already passed four hundred thousand downloads.

Mia: AgentLoop takes a completely different approach. It's built by a solo developer named Edward, and it prevents long-running agent chats from degrading. Each cycle starts a fresh Codex worker and a fresh critic that tests the result against a project rubric.

Milo: Failures become concrete fix notes for the next clean context. The system runs locally and sandboxed, with an observable dashboard and a polish mode that can continue until the critic issues a ship signal.

Mia: So Blaxel Agent Drive is a distributed filesystem, and the core idea is that multiple sandboxes can mount the same drive at the same time with read-write access.

Milo: That's a big shift, because normally if you've got, say, a coding agent and a review agent working together, you're moving artifacts through an object store or you're rebuilding context from scratch.

Mia: Right. With Agent Drive, those agents can share files, tool outputs, datasets, context—all through a normal filesystem path. One of the founders, Nico, demoed creating a drive, mounting it into two sandboxes, writing a file in one, and then reading that same file from the other.

Milo: And that's the kind of thing that makes multi-agent workflows feel much more seamless. You're not managing a separate transfer step.

Mia: Exactly. And then there's Cosyra 2.0, which takes a different angle. It provides a persistent Ubuntu workspace built for coding agents—and you can access it from your phone.

Milo: Adam, a co-founder, says he built it because he wanted an agent in his pocket. The idea was, he'd give a coding agent a task, walk away, and come back later to find it waiting for an answer.

Mia: So the new 2.0 workspace keeps repos, branches, files, installed dependencies, terminal sessions, and agent work all together across close and reopen. You're not losing state when you walk away.

Milo: That persistence across sessions is interesting, because it means the agent workspace starts to behave more like a developer's actual machine—always there, always ready.

Mia: Speaking of ready, PodcastorAI is tackling video podcast creation. It makes studio-quality video podcasts using AI hosts and natural voices.

Milo: And the pitch is speed—it turns scripts, links, PDFs, or even existing audio episodes into publish-ready videos in about fifteen minutes.

Mia: It supports both solo and two-host formats, and there's a digital twin feature that puts the creator on screen without cameras or a production team.

Milo: So you could be the face of your own podcast without ever setting up a shoot. That's a pretty sharp reduction in the friction to publish.

Mia: Meanwhile, Motionly is coming at motion graphics from a different direction. It's an open-source editor built for AI workflows.

Milo: And the key difference is, instead of generating final videos, Motionly's AI produces editable motion files that you can refine either visually or in code.

Mia: Right, so you get a visual editor combined with a structured animation format. That makes motion graphics versionable and reusable, which is a big deal for teams where developers, designers, and AI agents all need to collaborate on the same project.

Milo: And there's more coming—they've got plans for AI-powered animation generation from voice, websites, and educational content. So the scope is expanding beyond just the editor itself.

Mia: Let's move to the meeting space. Mufal is an AI meeting copilot that delivers live transcription and contextual answers during calls—but here's the twist: it doesn't join as a bot.

Milo: That's worth pausing on, because bot-join announcements can be distracting. Mufal instead presents an undetectable overlay that stays invisible in screen-share captures.

Mia: So nobody on the other end knows it's running. It generates automatic notes, action items, and what they're calling durable project memory. Runs out of the box on macOS, Windows, and iOS.

Milo: The founder shared that the product was built for in-the-moment help—understanding questions, recalling context, taking notes—instead of only working after the conversation ends.

Mia: And for advanced users, it supports bring-your-own API keys, plus there's a free plan with optional cloud sync.

Milo: Now Mnemcore approaches team memory from a different angle. It turns team video and timestamped notes into searchable organizational memory.

Mia: You capture notes tied to exact video moments, search across every recording at once, ask questions in natural language, and get evidence-backed answers linked to the original segments.

Milo: The founder, Jose, came from software engineering and competitive rugby. He described building Mnemcore to bring together scattered observations that would otherwise stay locked inside separate recordings and individual recall.

Mia: So where Mufal helps you during the meeting, Mnemcore is about making sure nothing is lost after the meeting.

Milo: Now Basedash has launched an AI Kit that exposes its analytics platform through an API—so you're getting the AI data analyst, automatic daily insights, automations, and dashboards, all powered by GPT-5.6.

Mia: And they're ranked number one on BI Bench. Max from Basedash announced it on Product Hunt, and the developer platform is designed for customer-facing analytics with a custom UI and per-customer query scoping.

Milo: So the idea is you embed analytics directly into your product, but each customer's queries are scoped to their own data.

Mia: Right. And then OpenCode Superapp brings Codex agentic capabilities to cloud, local, or self-hosted models within a native workspace that supports files, Git, terminals, voice interaction, and supervised Computer Use for Mac apps.

Milo: It's built with Codex and GPT-5.6, and a community question highlighted real interest in the self-hosted model experience—specifically around tool-calling reliability compared to cloud endpoints.

Mia: That's a practical concern. When you're running models locally, do the tools actually fire when they're supposed to? And there was also a question about how the supervised Computer Use implementation works for Mac apps.

Milo: So the community is already digging into the details around reliability and control, which makes sense when you're giving an agent the ability to operate your desktop.

Mia: First up this morning: canitbebuilt. You feed it a hardware idea, and it gives you a verdict.

Milo: Four possible outcomes: Buildable, Buildable with changes, Heavy lift, or Not viable.

Mia: And it doesn't stop there. You get a bill of materials costed at three volumes: one hundred units, one thousand units, and ten thousand units.

Milo: Plus subsystem risk ratings, certification gates you'll need to clear, and even a 3D concept model.

Mia: No signup required and the turnaround is about two minutes. The criteria behind all this, they say, is drawn from twenty-three years of hardware experience.

Milo: That's a lot of hard-won manufacturing knowledge baked into a tool. Now onto something for the SEO side of the house: CrawlRaven.

Mia: CrawlRaven connects your Google Search Console and Google Analytics, pulls in keyword imports from Ahrefs or Semrush, and runs a two-hundred-point technical SEO crawl.

Milo: Then it produces a single prioritized action plan, scored by ranking impact. It tells you what to write, what to update, and what to fix first.

Mia: And it includes cannibalization detection, so you know when your own pages are competing against each other in search results.

Milo: Also new today: Vevey brings an AI tutor called Matti to iOS. Matti helps you create 2D and 3D games.

Mia: Beyond creation, it offers game guides for daily puzzles, including Landmarkr and Wordle, and supports collaborative creation and publishing.

Milo: That was the full rundown. Let's tie it together.

Mia: The big story is Basement. It's a new shopping browser, and the headline feature is agentic checkout — its AI shopper finds the best price across the web and completes the purchase for you.

Milo: We also saw two company-knowledge products launch today, both trying to solve the same problem: AI assistants that actually understand your organization's information.

Mia: And three new macOS and iOS apps from solo developers, each taking a different swing at on-device voice input.

Milo: On the open-source side, two tools for governing autonomous AI agents hit Product Hunt — different approaches, same fundamental challenge.

Mia: Plus Blaxel Agent Drive, a distributed filesystem that lets multiple sandboxes mount the same drive concurrently with read and write access. Really clever infrastructure piece.

Milo: And PodcastorAI — it builds studio-quality video podcasts with AI hosts and natural voices from scripts, links, PDFs, or existing episodes.

Mia: Mufal caught our eye too. It's an AI meeting copilot that delivers live transcription and contextual answers during calls — without joining as a visible participant.

Milo: Basedash shipped an AI Kit that exposes their analytics platform — AI data analyst, automatic daily insights, and automations.

Mia: And finally, canitbebuilt uses GPT-5.6 Terra to evaluate hardware ideas. It gives you a verdict: Buildable, Buildable with changes, Heavy lift, or Not feasible.

Milo: That's the show for today. Thanks for spending the time with us. We'll be back tomorrow with everything new.