0815 | AI News: Gemini 3.7 Flash, Freebuff, Port22, Suno Studio 2.0

Mia: Welcome back to ProductHunt Daily. I'm Mia, and I'm here with Milo — and today's show is packed with fresh launches.

Milo: Great to be here. We're looking at a brand new Google model, a free coding agent that's taking on the subscriptions, and a browser-based music studio that's getting a major upgrade.

Mia: Plus a Y Combinator-backed cloud tool, a mac motion design studio, and a real-time music visualizer. Plenty to dig into.

Milo: Let's get started with what's new on Product Hunt today.

Mia: Google has a new entry in its Flash line. Gemini 3.7 Flash launched this week, introduced on Google's blog by its senior director of product management, and it's being positioned as the most intelligent workhorse model yet for coding and agents. The Flash series is known for speed and efficiency, and this version is meant to push that beyond raw speed, into serious software engineering and agent-driven workflows.

Milo: And the community reaction here is telling. One Product Hunt commenter framed it for people building AI-powered products, coding agents, or automation workflows, emphasizing complex software engineering tasks, multi-step planning, and autonomous agents — while keeping the speed you'd expect from Flash. Worth flagging, though, that's an attributed community take, not verified performance data.

Mia: Right, and pricing is one of the few hard details we actually have. The introductory rate runs through the end of this year. Starting January first, 2027, it'll be a dollar fifty per million input tokens and seven fifty per million output tokens. What we don't have — no benchmark results, no context window figures, no rollout dates. So the capability claims here are all coming from Google or from the commenters, not from independent testing.

Mia: Now to a launch that's squarely aimed at the subscription fatigue in this space. Freebuff is a self-described free coding agent, pitched as the no-subscription alternative to Claude Code, Cursor, Codex, Replit, Bolt, Windsurf, Lovable, and Devin. The maker's pitch: no API keys, no lock-in, ad-funded access to what it calls the best open source models — pointing to a roster that includes GPT 5.6 Luna, DeepSeek V4 Pro and Flash, MiniMax M3, MiMo 2.5, and GLM 5.2.

Milo: So the business model is ads instead of a monthly bill, and the free limits are generous on paper — six one-hour sessions a day, with some models unlimited. The cost comparison is striking: zero dollars a year for Freebuff, versus seven twenty for Cursor, twelve hundred for Claude Code, twenty-four hundred for Devin per year. And they claim over eleven thousand dollars saved annually for five seats. But those are the maker's figures, not independent benchmarks.

Mia: Community reports back it up anecdotally. One user logged nearly six hundred messages and close to two hundred million tokens — with zero spent on model access, ads as the trade-off. Another said months of use felt equivalent to Claude Code but free, covering data analysis and research. But there are skeptics too — one commenter questioned the business model over a monthly Claude API bill, and another wondered when enterprises would trust free models. One even claimed the product started life as CodeBuff back in 2024.

Mia: Port22 is taking a different angle — it's an iOS companion for coding agents that run on a Mac. It works with Claude Code, Codex, or OpenCode, with Hermes, OpenClaw, and Pi coming next. The key design decision: it attaches to the terminal session you already started, rather than wrapping or reconfiguring it. On your phone, you can see which agents are working and which are stuck, read the live transcript, and get a push when an agent finishes or needs approval.

Milo: And those approval buttons aren't guessed keystrokes — they're the actual options read from the live session. The maker calls that trust detail the worst possible failure mode to get wrong. Pricing is simple: free for one Mac and two simultaneous sessions, fully featured and not a trial, or Unlimited at about twenty dollars a year to drop the caps. No account or signup — pairing is through the same Apple ID on both devices, or a six-character code over Wi-Fi for a work Mac on a different Apple ID.

Mia: Off the local network, traffic goes through an end-to-end encrypted relay the maker says can't read the code. One limitation worth noting: it supports open-source terminals like Ghostty, kitty, Alacritty, WezTerm, tmux, and Terminal.app — but not Warp or an IDE's built-in terminal. Community questions are open rather than settled — people asking about Windows and Android, and whether an agent would sit at the prompt indefinitely if your phone is off and the notification never lands.

Mia: Hoplite is a Y Combinator-backed cloud tool that wants to take your local coding-agent setup and deploy it to the cloud without reconfiguring anything. Onboarding migrates your sessions, your MCP servers, your dependencies, and your CLIs, so agents pick up where they left off — running in isolated sandboxes instead of crashing your laptop. The site talks about running hundreds of agents concurrently.

Milo: The workflow is four steps: connect a GitHub repo, import your local setup, start the agents, and verify via automatically prepared video recordings of new features before merging. There's also instant previews of agent output, prompting from iMessage, instant resolution from Sentry issues, one-click share links, and Linear and Slack integrations. The founder context is interesting — co-founder Ryan says it came out of internal tooling that let his team deploy a million and a half dollars in OpenAI credits effectively.

Mia: Community reaction is early but warm. One commenter with minimal usage called it one of the easiest cloud agents to onboard onto, compared to Factory and others, though they haven't fully switched from local yet. Another praised moving an existing local setup to the cloud without rebuilding everything. The co-founder projects concurrent agents growing from under a dozen now to thousands by the end of next year — but that's a stated belief, not a verified result.

Mia: Finally, OpenMotion is a free macOS motion-design studio aimed at founders, designers, and developers who want product launch videos without hiring a motion designer. The idea: you describe a scene in plain English, attach screenshots, logos, or a brand kit, and its AI agent generates editable motion, product launch, and SaaS explainer videos. Unlike one-click generators, the pitch is control — scenes, layers, timing, easing, colors, camera, and sound can all be refined on a real canvas before exporting.

Milo: Two concrete differentiators stand out. It runs natively on macOS with builds for both Apple silicon and Intel Macs, and it works directly with Codex and Claude Code subscriptions, so no separate API keys needed. But pricing is the open question — it's free while they're building, with one free plan today, and the FAQ notes a possible future Pro plan with higher limits, with some advanced features possibly moving to paid later. So availability and limits aren't settled.

Mia: The community evidence is mostly maker-shared reactions on X. One endorsement called it a really cool tool that creates polished motion graphics from a prompt and credited it with an official launch video. Another said it produced the motion design for their product. These are attributed endorsements, not independent verification — which sums up most of this week's launches. Strong pitches across the board, but with pricing, limits, and real-world performance still very much in motion.

Mia: Suno has dropped what its maker calls the biggest update to its browser-based digital audio workstation yet, and it's live now for every Suno Premier subscriber. The whole pitch is built around granular, hands-on control — the tagline is literally “Be a control freak” — so this is aimed at people who want to touch every detail, rather than just typing a prompt and getting a finished track back.

Milo: The headline additions are a pretty complete toolkit. There's MIDI support, so you can play chords, basslines and melodies on built-in synths with a physical controller, plus audio effects ranging from sidechain compression to convolution reverb. You can design your own effects and synths from scratch, there's a chat bar that builds sounds and plugins in your own words, and it will even batch-rename your tracks for you.

Mia: Beyond that, you can import audio or stems, edit tempo, volume and pitch, automate any parameter by drawing curves or recording movement, and the stem separation now splits audio out into vocals, drums, bass and more. Then you can export full songs, individual stems, or just the range you selected.

Milo: Early community reaction is thin but enthusiastic — one commenter called the update “actually insane” and wrote that it spells trouble for other DAWs, though that's an opinion, not a tested comparison. And there are still some open questions. The maker's own FAQ asks, without yet answering, how Studio 2.0 differs from rival workstations, how many downloads you can retrieve, and whether stem and multitrack downloads count against the Premier subscription's download limit. So the feature set is ambitious, but a few practical details are still being sorted out.

Mia: Next up is ChordViz, a real-time music visualizer aimed squarely at music teachers and creators. It's a standalone app for Mac and Windows that turns live MIDI and audio into on-screen visuals — chords, notation, a keyboard, music theory and audio-reactive graphics — all arranged into a dashboard you build once, save, and reopen whenever you want.

Milo: The core claim is real-time chord detection. As you play, it names the chord on the spot — including extensions, inversions and slash chords — reading the root, the quality and the bass note live. That's the kind of thing that makes it genuinely useful in a lesson: the student plays a chord and immediately sees what it actually is.

Mia: You get eighteen widgets to arrange — a keyboard, chord name display, circle of fifths, grand staff notation, oscilloscope, spectrum, metronome and more — plus dark and light theming and a fullscreen focus mode. Recording is built in too: video with audio at up to 1080p and sixty frames a second, stereo audio, and standard MIDI with tempo capture. There's even an always-listening fifteen-second buffer, so you never miss an unplanned moment.

Milo: For creators, it feeds into live performance setups — streaming to OBS, TouchDesigner and Resolume — and you can load your own virtual instruments if you want. It's free for seven days, then twenty-nine dollars one-time with lifetime updates, which is a pretty forgiving price for what it's packing.

Mia: Moving to the launch everyone's talking about today: BrowserAct Cloud, an AI agent that builds web-scraping bots from a plain-English description. The maker, a senior marketing operations lead at the company, says the whole point is that you stop thinking like a scraping engineer — no selectors, no browser steps, no pagination logic, no retry handling, no maintenance.

Milo: The agent opens the live website in a real cloud browser, explores the page flow, tests its extraction, and then hands you a reusable bot that outputs structured data — as CSV or JSON, through an API or webhook, or straight into automation tools like Make, n8n and Zapier. Managed infrastructure is built in: real Chromium, rotating or persistent residential proxies, automatic CAPTCHA handling, scheduling, and persistent identities for logged-in workflows. A sample run scraped the top twenty wireless headphone listings from Amazon in under three minutes.

Mia: On launch, you get fifteen hundred credits at signup and a seven-day free trial on upgrade. Audience-wise it's aimed at operators, founders, growth teams and researchers collecting structured web data. One commenter found it practical for agencies doing repeatable competitor and pricing research, and another called it the first AI scraping tool that worked for them without any fluff.

Milo: There's also real interest from heavier users — someone with more than a hundred and fifty production conversational agents asked about real-time scrape-then-reply integration, and a marketing researcher said they value raw filterable rows over summaries. Still an open question in the thread is whether the bot auto-detects layout breaks and heals itself, or fails silently.

Mia: Next, Basedash — the AI-native business intelligence platform — has launched Tasks in research preview. It's an AI operator that reads a company's real data, meaning revenue, churn, activation and pipeline, and converts it into a prioritized backlog of specific actions.

Milo: Each task comes with three components: “why now” evidence pulled straight from the actual data, an expected outcome with a rough metric target, and step-by-step instructions. Concrete examples include winning back the 214 trials that stalled before connecting data, and chasing twelve overdue invoices worth eighty-six thousand dollars.

Mia: The tasks can be copied as structured briefs into Linear, or pasted into agents like Claude, ChatGPT or Cursor, and the data work can also be executed directly by Basedash's own agent. Crucially, when you mark a task done, Basedash checks whether the targeted metric actually moved and feeds that result back into future recommendations — so the system claims to sharpen over time.

Milo: The founder says Basedash has been running on Tasks internally for the past few months, and that their backlog now starts from data rather than whoever spoke last in planning. That's a nice line, but the research preview status carries real caveats: the core loop of generating, executing and refilling tasks is ready, while impact-tracking and learning are early and still evolving, with rough edges expected. Setup means connecting data sources, enabling Tasks in settings and clicking generate.

Mia: Finally, there's oxpecker, an invite-only developer tool for teams that call third-party APIs. Its maker says he built it after the third time his team learned about an API change from a customer instead of from the vendor. The tool watches the API specs of 68 vendors, runs as a GitHub Action inside your CI pipeline, and reports which exact lines in your repo call a changed endpoint — surfaced right on a pull request ahead of the sunset date.

Milo: So it's not content to tell you that Stripe changed something. Oxpecker names the line — for example, a specific file calling an endpoint that's going away in March — and the site bills itself as Dependabot for the APIs you call. The scale of the problem is stark: across thirty vendors with verifiable spec history, the maker measured 917 breaking changes in twelve months, meaning one roughly every ten hours, with Cloudflare alone responsible for 368 of them.

Mia: The trade-offs are honestly stated. Roughly half the register has no public spec history, so that 917 figure is a floor, not a total. Access is invite-only while coverage widens, and users can request vendors by email. Pull requests add just one reviewable file — nothing auto-merges — and the maker says your source never leaves your CI. So it aims to give you certainty about exactly where an API break will hit, without ever trying to fix a call it can't prove.

Mia: Let's look at a new app for pet owners called Buply, which just surfaced on Product Hunt. The idea is a virtual ID for your pet, built by scanning its noseprint — the team compares it to a fingerprint, since every animal's nose pattern is unique. So instead of relying on a collar tag or a microchip, the animal's own nose becomes the identifier.

Milo: That's a clever concept, and they claim to be among the first pet biometrics apps using noseprints to reunite lost pets with their families. Owners can build out a detailed profile — breed, color, traits, even medical history. Then, if someone finds a lost dog, they scan the noseprint, pull up that profile instantly, and can chat directly with the owner. So the community itself becomes an active safety network.

Mia: Right, but it's worth being clear about where this stands. Dogs are supported right now, and cats are coming later. There's a TestFlight beta available for early testers. And the long-term aim is a single platform that pulls together pet safety, health info, adoption, and nearby services — a much bigger vision than just reuniting lost pets.

Milo: At the same time, we should note the caveats. All those capabilities and the roadmap come straight from the team's own Product Hunt launch comment, so they're maker claims rather than independently verified results. There's no third-party feedback, test results, or adoption data yet. And the team is candid about the open questions — will owners actually use a noseprint ID, what do they need first when a pet goes missing, and what feels confusing. No pricing has been announced either, so for now, treat it as a promising early idea with real unknowns still in front of it.

Mia: And that's our brief for today — everything from Google's new Gemini 3.7 Flash and Freebuff's free coding agent to Port22 and Hoplite for teams running agents on Macs, plus the creative angle with OpenMotion, Suno Studio 2.0, ChatGPT for Mac, and a fresh look at Claude's Opus model. Whatever you're building today, we hope something here sparks an idea.

Milo: We'll be back tomorrow with the next round of launches. Until then, happy shipping, and we'll talk to you soon. [pause ms="300"] Bye for now!