PH热榜 | 2026-08-16

@haydenbleasel  @fmerian Curious what "AI-ready" means concretely here, llms.txt and clean semantic output, or something more? That term is getting stretched a lot right now and yours seems to mean something specific.

@haydenbleasel  @fmerian Hayden is on one of those “apparently sleep is optional” runs.

Ultracite, and now Blume. Open source, zero config, AI ready, and coming straight for the docs stack.

At this point, we’re not even watching a maker ship products. We’re watching a maker build an ecosystem. 🔥

@haydenbleasel  @fmerian Congrats on the launch! "AI-ready" jumped out at me. I’m curious how you’re actually pulling that off. Is it just clean markdown that LLMs can chew through easily, or are you baking in structured data or schemas so AI assistants pull and cite docs correctly, not just scrape them? We’ve been down this hole ourselves on the content team, so I’d love to hear how you’re tackling it for a docs-first product.

@galdayan Means lots of things - llms.txt, raw markdown via negotiation headers, mcp servers, agent readability, skills. All in the pursuit of making your docs easier to read by agents. Check it out here: https://useblume.dev/docs/configuration/ai

@galdayan As for staleness, we have a few things to help here e.g. translations but hoping we can work on bridging the gap between product and docs more going forward, perhaps with skills.

@gapostolov I get a blank ehite page. Let me know when you get it up again?

@renchu_song For teams adopting this, what’s one non-obvious failure mode you’ve seen when swapping harnesses that the Unified Harness Protocol specifically protects against; so we know where the abstraction truly earns its keep?

@renchu_song The “OpenRouter, but for agent harnesses” framing is dangerously clean.

There’s something refreshing about admitting, “We kept rebuilding the same infrastructure, then realized the good stuff already existed.”

Going from weeks to hours while getting better agent performance is a pretty strong argument for letting the harness do the heavy lifting instead of reinventing it with another orchestration layer.

And the Unified Harness Protocol feels like the bigger bet here. If it actually makes harnesses interchangeable, builders can start treating them more like infrastructure than a permanent architectural decision.

Also love that you’re pushing beyond coding. The moment Codex, Claude Code, Hermes and friends become general purpose workers, the interesting question stops being “which agent?” and becomes “what ridiculous thing can I make it do?” 👀

Curious where you think the abstraction breaks first. That’s usually where the really interesting product opportunities hide.

@renchu_song Nice launch Congrats🙌 loving the clean UI and vision here, How are you guys handling data privacy and local vs. cloud processing under the hood?

@conduit_design Great question André! And yes, we've been watching the dsh launch closely (the Cordis paper is genuinely interesting work on revertible effects / hot-swap composability).

The way we think about it: dsh and HarnessRouter are composable at different layers. dsh makes the internals of one harness swappable: model adapter, tool registry, agent loop, all plugins. HarnessRouter makes harnesses themselves swappable: you integrate one API (UHP), and you can route each task to Codex, Claude Code, Hermes, and soon Pi, dsh and more, without rebuilding your backend. UHP is a versioned open spec (OpenAPI 3.1 + JSON Schema, 47 runnable conformance checks), so if you want to swap in your own harness implementation, or fork ours, the contract is public and independently testable. That's where the fork-and-customize story lives.

On plugin-level flexibility: it's always a trade-off between composability, customizability, and ease of use. Our bet is that the frontier labs and the best harness teams will keep their harnesses tuned with the latest best practices, and most application builders get better results building on those than reassembling agent loops from parts. It's an 80/20 thing, most use cases are best served by a strong common foundation; the domain-specific cases that need a 10-20% edge may want plugin-level control, and dropping down to something like dsh makes sense there.

And because we sit a layer above, dsh isn't a competitor for us, it's a candidate harness. We can absolutely support it behind the same API and let the market tell us which harnesses win, beyond the launch-week hype. Thanks for the thoughtful question!

@kuanzema  @abhiranjan_mehta Exactly the bar we set for ourselves. Community edition and Cloud implement the same versioned unified harness protocol spec and both have to pass the same growing list of check conformance suite, so an integration built on community edition moves to Cloud unchanged, and back again. Cloud sells managed infra and observability features, never API surface. Lock in through withheld APIs is a business model we specifically wanted to avoid.

@abhiranjan_mehta Exactly. We want users to have that sense of safety and freedom. Moving from local to Cloud is a choice, not a forced migration: our open-source Community Edition gives developers control over their own infrastructure, while Cloud handles deployment, operations, and scaling when they’re ready for production.

@marcin_michalak Thank you so much for your support, Marcin! I see a huge synergy between the HarnessRouter being an intelligent unit layer, and AgentX's multi-agent orchestration as a mesh layer on top, together we will be form a complete solution for Level 5 of AI: AI Organizations, which is the ultimate step in OpenAI's 5 level AI roadmap

@huisong_li Thank you for your support, Huisong! Your support means a lot to us :D

@charlie_titherley Hey Charlie, as a founder/CEO, I constantly need to build a lot of pitch decks and PPTs. I was so frustrated about all AI ppt products in the market and I tried to build one myself. When I first tried it last September, I find it so intimating to even get an agent loop (harness was not a term back then) able to understand a PPT template. A lot of trials and errors, a few weeks and no success. The only learning I had was agent able to generate EChart json config and be embedded into a slide page as an image, following the style.

This year, I retried the same challenge. With the harness being so mature, and a lot of taste skills, I was able to build a fully functional PPT agent within a day. We put our PPT agent implementation as one of our starter kits at https://github.com/HarnessRouter/starter-kit/tree/main/kits/slides

@galdayan Thank you for your insightful question! It's the classic abstraction-layer problem, and the trap you're describing (silent LCD) is what UHP is designed against.

First, more normalizes than you'd think. Claude Code's tool-approval flow, from the app's side, is just a lifecycle event: "run paused pending approval, approve/deny." Subagent delegation surfaces as structure in the trace. UHP models these in the common surface (sessions, streaming events, cancellation, structured errors), so harnesses that support them emit them natively.

Second, what doesn't normalize is explicit, not silent: harnesses declare what they support so apps can feature-detect and branch, and invoking an unsupported capability returns a structured error, same pattern LSP used (capability negotiation at init) to avoid LCD across hundreds of language servers.

Third, harness-specific config (permission policies, skills, MCP, model policy) stays typed at the harness-config layer rather than forced through the intersection. So the ceiling is per-harness, not weakest-common.

The common core is what's hardened by the conformance suite today; the capability surface will keep evolving as we add harnesses with weirder shapes. Spec's public and versioned, would honestly love your review on exactly this seam: https://github.com/HarnessRouter/harnessrouter/tree/main/protocol

@aureon Handcrafted explorable worlds that stay lightweight in the browser is a hard combination to pull off solo, most 3D-on-the-web dies on load time. How are you keeping the expeditions small, and roughly what does one weigh?

@aureon i completely love that you are solo + your own hunter too. let's make you #1 producthunt. just build cool stuff!!!

@marco_ciavarella Thank you for the kind words Marco. As you pointed out, my future vision is to have classroom passes (without login) so that kids can jump right in and enjoy the worlds. Also, whenever you have the time, do check out Ancient Egypt. I had so much fun making it and I learnt a lot of crazy facts haha!

@inna_fridrikova Hey Inna, thank you for supporting me that early and for your kind words. I appreciate it a lot!

@hejsfj Hi, Thank you. I have quite a lot of things brain dumped into future expeditions channel haha, everything from Ancient Civilizations (such as an underground/floating cities) to subsurface ocean of on Saturn's moon! But there is a community voting system live, which is what I'd consider next.

@thisiskp_ Thank you very much! I will check it out. Appreciate the invite!

@yosun I fail to understand. How would building something on your own be worthless? It’s like calling your baby worthless before it’s even had a chance to grow.

Do not ponder others' opinions. Take pride in your work, and your work shall speak for itself.

Ideas need time to nurture. Take good care of them.

@manavsehgal Mmhmm, I actually had my mind blown when I learnt they had to cut, quarry, ship and place it all every 2-3 mins. Those blocks weigh tons! And yet the internet sells us fringe theories, which is why I wanted to show the hard work behind it. As for the time, I'd say anywhere between 10-15 mins as a new visitor? I have lost count of it because I just click 'next' crazily seeing the same thing over and over again across months during development 🤣

@sina_tayebati Thank you, really glad to hear that.

@lorbes Thank you. I'm just passing along the feeling of how the web was back in the days to the next generation.

@alexandrmucha Hi, yes. The dioramas and the characters were modelled, rigged and animated by myself.

@etiennegarcia Took me 3.5 hours to sync the text with the audio haha. Glad you didn't skip it, lest you'd have faced the wrath of Cleo!!!

@zeng Thank you very much!

@kevin_amrelle Reading about that winter bike race where your heart rate capped at 120 BPM and none of your apps caught the trend really stuck with me—it’s wild how much we rely on tools that don’t actually talk to each other. What’s the first health signal you connected when you started building Vidaya?

@kevin_amrelle  “Thanks for the feedback! We really appreciate it and will definitely keep this in mind.”

@kevin_amrelle Health data has become incredibly fragmented, so bringing wearables, labs, nutrition, DNA, and medical records into one context feels like a genuinely useful step forward. The personalized action layer is the part I’d be most interested in trying.

@monir_ Thanks, Monir. "What should I actually do next" was literally the design question. Most of us do not need a tenth dashboard, we need the Tuesday morning answer. Every score and trend links to a concrete next step, and you can ask Vaya why a number is moving and what to do about it against your own data. Congrats on CheckYa as well.

@1mirul Thank you. Good and bad labels were the bar we wanted to clear. A red number without a reason just creates anxiety. Direction of the trend, the reason behind it, and a specific next action is what actually changes behavior. Appreciate you taking a close look at how the trends work.

@tanjum Thank you, Tanjum, that distinction is the whole product. A generic chatbot can only give you generic advice; Vaya answers against YOUR history and cites which of your sources each answer drew from (which panel, which device, which date range), so you can see the basis rather than trust a vibe. And when your data can't support an answer, it says so instead of guessing. That grounding is what makes AI actually useful for personal health, in our view. Appreciate you looking closely!

@hamza_afzal_butt Great question, Hamza! Transparency is something we designed for at two levels.

At the answer level: every Vaya response cites which of YOUR data sources it drew from (which lab panel, which wearable, which date range), so you can always see the basis for a recommendation rather than taking it on faith. If the data isn't sufficient, Vaya says so and tells you what's missing instead of guessing. And recommendations come with the evidence framing, target ranges, and the "why". This is drawn from a curated library of 1,000+ citations from medical literature, not free association.

At the scoring level: the scores are deliberately explainable rather than a black box. The VAI Score (data completeness) shows you the per-category breakdown, you can see exactly which categories are pulling it up or down and what to connect next. The Healthspan Score is broken into five labeled pillars (cardiovascular, metabolic, sleep & recovery, body composition, resilience), each traceable to the underlying markers.

And a boundary we're explicit about: Vaya gives you direction, not diagnosis, anything clinical gets redirected to a medical professional. Every response is also scored in production by our observability layer (9 LLM-as-judge evaluators at 100% sampling), so "is this answer grounded in the user's data?" is checked on every single response, not just at test time.

@abod_rehman Thank you, Abdul, that rigor is very deliberate. Our background is in banking, one of the most heavily regulated industries there is, so we came into health AI already thinking about accuracy, auditability, and what happens when a system gets something wrong. That shaped how we built Vaya: every response is scored in production by 9 LLM-as-judge evaluators and 9 monitors at 100% sampling, not just at test time. And rather than letting the model free-associate, we built a curated library of over 1,000 citations from medical literature that Vaya draws on, so answers are grounded in your data plus referenced sources, not vibes. Health AI deserves at least the same discipline as banking.

Great question, @imtiaj_ahmad , and yes, the foundation for this is already live. We built Squads, shared dashboards for small groups (originally athletes, pilots, military teams) where a captain, coach, or trainer can see the group's data, with privacy controls so every member chooses what they share.

Family is a natural next step on that same foundation, and managing a parent's health data is one of the use cases we hear most. An adult child watching a parent's trends is basically a two person squad with the right permissions. Children are a different story. Pediatric health data carries extra regulatory and ethical obligations, COPPA among them, so we are not going to rush it. If we do it, it gets done properly.

Is your use case more the parent side or the kid side? That genuinely helps us prioritize.

@istiakahmad Thanks, Istiak. That was exactly the design bet. A single day's number is mostly noise, but the same marker across 7d, 30d, 90d and 1y is where the real signal lives. Slow drifts, a resting heart rate creeping up over a quarter or a lipid trend across two panels, are invisible day to day and obvious over a year. Which metric would you most want to see a full year of?

@odeth_negapatan1 Great question, this is one of the specific failure modes we designed against. When Vaya doesn't have enough of your data to answer confidently, it says so explicitly rather than filling the gap with a plausible-sounding guess. It tells you what data it's missing and what to connect or upload to get a real answer (e.g., "I don't have recent blood work for you — your last panel was 14 months ago"), and every answer cites which of your sources it actually drew from, so you can see the basis yourself. We test this deliberately: our stress suite includes hallucination-trap questions that ask about data the account doesn't have, and in our May production audit Vaya answered "that data isn't available" correctly on all of them. And for anything urgent or clinical, it redirects to a medical professional instead of playing doctor. The honest answer to a health question is sometimes "I don't know yet, here's what I'd need", we'd rather say that than make something up.

@priyankamandal Thank you, Priyanka, you can actually do this today! Vidaya generates a downloadable PDF health report from your dashboard covering your yearly trends across all your connected domains, plus dedicated exports for blood panels (full marker tables with reference ranges, formatted so a doctor can read them at a glance) and DEXA results. It's exactly the use case we built them for, walking into an appointment with a year of your data instead of describing things from memory. We're continuing to expand what's included (raw CSV/JSON export of everything is on the roadmap). If you try it with your doctor, I'd love to hear how the conversation goes!

@raihanshezan hanks, Raihan! Dashboards were the easy 80%, the hard 20% is turning a moving number into "here's what to do on Tuesday." That's the problem we actually wanted to solve, since most of us already had plenty of graphs and no answers. Glad it landed! 🙏

@kate_ramakaieva Thank you, Kate! You nailed the core idea, a single timeline is the unlock. Any one reading (a sleep score, one lab value) is just a snapshot, but put wearables, labs, nutrition, DNA, and medical records on the same axis over years and the trends start explaining each other, that's where the "oh, THAT'S why" moments live. Congrats on everything you're building at FuseBase too! Out of curiosity, which of those data types would you connect first?

@german_merlo1 Thank you, Germán! Love hearing you've started paying closer attention to your health, that's exactly the moment Vidaya is built for, when you start caring and immediately discover your data is scattered across five apps. If you give it a spin I'd genuinely love to hear what you connect first. Wishing you the best right back!

@suyash_kr Thanks!!

@rajesh_gupta__ We encrypt user information so that it is never exposed

@vimal_singh6 Use cases could be anything from remote support to field service, telehealth intake, or guided onboarding!

@garygao I can already see this being a game-changer for remote tech support—no more back-and-forth trying to describe a weird UI glitch. Could this eventually handle more complex scenarios like live diagnostics or repair walkthroughs?

@garygao Congrats, seems pretty cool. Are you planning to expand to WhatsApp or other channels, since Facetime is pretty exclusive?

@mcarmonas Thanks!

@maurya_abhiranjan Thank you!

@abhiranjan_mehta Thanks!!

@arun_pratap_singh7 Yes, we have plans to expand to WhatsApp messaging + video calling in the future!

@abhi_singh4000 Yes, visual identity verification could actually be more robust than a lot of the current identity verification infrastructures over text, email, or calls today

@priyankamandal Our latency is really good. Feel free to give it a try on our website here: https://www.trychert.com/facetime

@manavsehgal Thank you — really glad that part resonated with you!

The wake window is actually one of the core ideas behind Air Alarm. Instead of asking you to pick the exact minute you want to wake up, you just set the latest time you absolutely can’t sleep past. Based on your sleep cycle, Air Alarm then chooses the best moment within that window to wake you when you’re more likely to wake up naturally.

So ideally, no more lying there calculating “if I fall asleep now, I’ll get X hours…” — you set the deadline and let the app handle the rest.

And yes, keeping things private and lightweight is very intentional. There’s no complicated account system right now, your sleep history stays on your device, and I don’t really plan to turn it into an account-heavy app in the future either.

@build_with_aj Thanks for asking! I don’t have plans for an Android version at the moment. Air Alarm relies on the AirPods feature that detects when you fall asleep and automatically pauses audio, and as far as I know, there isn’t an equivalent capability I can access on Android right now. If that changes in the future, I’d definitely consider it!

@micahdoes Construction has somehow been asked to choose between software from 2005 and AI that confidently invents a $40,000 mistake.

So yeah, “speed with control” feels like the right problem to solve.

I especially like that you’re not trying to replace the expertise that makes construction complicated. The interesting opportunity is using AI to remove the tedious work while keeping humans firmly in the loop when the numbers actually matter.

If CostLogic can make estimating faster without making people nervous about trusting the estimate, that’s a pretty compelling sweet spot.

Curious to see how far you can push that balance. 🔥

@marupelkar Seeing Fable and Remotion actually pay for graphics through Vaaya really brings the vision to life—it’s one thing to talk about agentic commerce, but watching it power actual creative work is next level. What’s one tool or workflow you’re most excited to see agents tackle next with these multi-step recipes?

@omri_ben_shoham1 Fair challenge, and you're right that a GitHub profile is not a hard identity. We don't treat it as one.

The thing that blunts your specific attack is that the GitHub line is card-gated. Nothing from it is spendable until a card is on file, and that card settles whatever gets drawn. So renting an aged high-star account gets you a bigger number on the screen, not free money. On top of that the amounts are deliberately small. We're underwriting tens of dollars, not thousands, so acquiring and burning an account costs more than the payoff.

That sizing isn't an accident. I've spent time in alternate lending, where thin-file underwriting always has exactly this problem. The answer there was never a perfect signal, it was small limits, fast feedback, and tightening as you learn. Same playbook here.

On weighting, it isn't stars or age. Those are the cheapest things to manufacture so they carry the least. I won't pretend the current model is more than a starting point though, and we're actively improving it.

The real unlock is spend history. Once accounts have a track record of drawing and repaying, that becomes the primary signal and GitHub drops to secondary. That's effectively a FICO equivalent for a developer profile, and it's what we're building toward.

@alexander_benz Let's see if I've got what you're doing here before I go diving into this repo. First, open-source? You're already "plus ten" in my book. Ok. So, I often have long, detailed, conversations with ChatGPT when she it tied into my github repo. I have codex running in the app. So I totally get the notion of having ChatGPT do all the brain power and then what I have her do is give me the detailed, maybe one-shot, prompt to get stuff done in Codex with those prompts sometimes going for 20 minutes of execution. Then ChatGPT, via looking at Github, can review the work and create the next codex prompt. Codex is arms n legs, ChatGPT is the brain. It let's me use cheap model in Codex and the latest in my 20 dollar ChatGPT subscription.

So, are you essentially saying, why do the codex part, have chatgpt do what I do now but THEN kick it over to the MCP server to execute on whatever? Do I have that right?

@codeandsea Thank you so much! Yes, the app will be available on all Apple devices very soon. The iPad version is almost ready, we just need to polish it up a bit. The macOS version will be available very soon as well.

@tori_seidenstein yes, agreed. Streaks feels weird in journals

@tori_seidenstein Thank you so much! glad you like it :)

@nicolas_carmont Thank you! means a lot!

@heena_kohli Hi, thanks for giving Yap a try and I do appologies for that experience. Sadly i haven't seem to record that issue but that's no excuse! Can you please reach out to the support email (mine) so we can connect, I'd like to understand more closely what failed.

@daniel_smidstrup Exactly, that's the idea! The barrier to entry is not code anymore, so there are plenty of ideas that can be executed, but not all of them are worth it!

Thanks for your comment, Daniel! :)

@farid_sukurov Yes, I was inspired by all the people who are currently creating their products from scratch; I don’t want them to make the same mistakes that others of us may have made with our first projects.

Thank you for your comment Farid! :)