PH热榜 | 2026-07-23

@thys_beesman Brandon, this is exactly the kind of complexity we built Teable for. Migration preserves the structure and flags anything that needs attention. Automations also surface issues when the underlying data changes. We’d love for you to stress test it.

@thys_beesman We usually don’t promise a “100% migration.” We aim for 200%: not merely reproducing the old Airtable system, but making it significantly more capable.

Moving rows is the easy part. The real value lives in the relationships—linked records, references, lookups, rollups, formulas, permissions, and the business logic accumulated around them. Teable’s relational model can preserve that connected structure while giving you more powerful ways to reference, query, and operate on it, all backed by real PostgreSQL.

But migration is only the starting point.

Once that relational foundation is in Teable, you can build almost any AI workflow or custom app directly on top of it. Agents can understand and act across customers, contracts, owners, approvals, renewals, and communications—not as disconnected text, but as related business context. They can generate reports, request approvals, send reminders, update records, and power purpose-built applications without forcing you to scatter logic across automation platforms, scripts, databases, and app builders.

So where exact one-to-one legacy behavior makes sense, we reproduce it. Where the old system was constrained by Airtable’s limits, Teable gives you the freedom to rebuild it better.

That’s what we mean by 200% migration: your existing relational knowledge comes with you, but the system that emerges can be far more adaptable, automated, and intelligent than the one you left behind.

Give us your hardest linked-record workflow—not the clean demo base. That’s where Teable’s relational foundation, freely customizable AI workflows, and custom apps begin to unlock possibilities that a conventional migration tool simply cannot.

@artem_fedorovich That’s exactly the problem Teable is built to solve—but we’re much more than a database-spreadsheet.

Teable is a complete environment for agents to operate in. Your data, apps, automations, permissions, audit trails, and collaborators all live in one place, backed by real PostgreSQL and mature, production-grade infrastructure. Instead of critical business context being scattered across spreadsheets, automation tools, internal apps, and disconnected AI agents, Teable brings it together into one trusted system.

That means the AI doesn’t work in a vacuum. It understands the live data, permissions, and workflows your team already relies on—and can act across them seamlessly. Complex configurations and automations that once required several tools, integrations, and months of custom development can now be built and operated through one coherent experience.

Our belief is simple: teams shouldn’t have to choose between the flexibility of a spreadsheet, the reliability of a database, and the power of AI agents. Software should adapt to the way your business works, while keeping every change controlled, traceable, and trustworthy.

So yes, Teable can absolutely be your real source of truth—but that’s only the foundation. What we’re really building is the place where your data and agents work together to run the business.

I strongly recommend giving it a try with one of your real team workflows. That’s when Teable truly clicks.

@artem_fedorovich Yes, Teable is designed to be the source of truth. Permissions and audit logs stay with the same data your team uses for apps and automations, so everyone works from one trusted system.

@artstavenka1 Great question—and this is exactly why the underlying architecture matters.

Every step is logged and traceable, individual actions can be rolled back, and agent permissions can be tightly scoped to control what each agent is allowed to read or change. If something fails mid-run, the agent can inspect the failure, attempt a fix, and continue—instead of leaving you with an unexplained, half-finished result.

And inside a team, multiple agents can collaborate across the same connected data and workflow, each with its own responsibilities and permission boundaries.

So the goal isn’t just to let AI write to your data. It’s to make agent work observable, controllable, recoverable, and collaborative enough for real business operations.

@artstavenka1 Every step is logged and traceable. If something fails along the way, the agent will inspect it, try to fix it, and do its best to finish the job.

@ansari_adin Ansari, it covers them all. Teams describe the business process they need, and Teable AI works across the data, automations, integrations, and custom apps needed to make it run. A lead workflow, for example, can include importing data, enrichment, routing, follow-up emails, and a dashboard in the same workspace.

@alex_watson2110  Yes, you don’t have to choose between powerful software and control of your own data. Full capability, full ownership.

@alex_watson2110 Exactly. For teams that care about data ownership and infrastructure control, being able to migrate to Teable and self-host it is a powerful option. Thanks for the support!

@avery_thompson2 Absolutely. Teable is designed for shared workspaces, not isolated one-person AI sessions. Team members collaborate around the same live data, apps, and workflows, while agents can have distinct responsibilities and permission scopes. Every action is logged, so the team can see what happened, who initiated it, and what the agent changed. The goal is shared AI context without sacrificing control or accountability.

@avery_thompson2 Thank you! Shared workspaces are central to Teable. Team members work with Teable AI around the same live data, apps, and workflows, with permissions and a clear history of every change.

@sabber_ahamed Permissions are part of the system the AI builds within—not an afterthought bolted onto the generated app. Agents operate inside defined access boundaries, and every action is logged and traceable.

You still retain final control and can review or refine access rules for each app.

@sabber_ahamed Sabber, Teable AI can handle the access setup as part of building the app, and you can review or adjust it for each app.

@kellyops  Kelly, that’s a sharp read. We’re bringing the data layer, workflows, and custom apps into one place, so teams don’t have to stitch Airtable and Retool together.

@sophialgrowth Sophia, absolutely! That’s a great use case for Teable. You can tailor the pages, actions, and data each group sees based on their role, while built-in login handles sign-in and app access. Everything runs on the same Teable data and workflows.

@ryancheng Thank you! And yes—the AI goes beyond moving rows. It can understand messy source data, infer schemas and field types, and help map relationships before building the workflow. You can still review and adjust everything, but you’re no longer starting from a blank mapping screen.

@ryancheng From an ops perspective, this means less setup work and fewer mistakes. The AI handles most of the mapping first, and you just review and adjust anything that needs attention before moving the data.

@ryancheng Great question! Teable AI can identify schemas and field types even when the source data is messy, then map them automatically. You can adjust anything if needed.

@wys1010 Exactly 😂 Teable gives the spreadsheets superpowers—and finally lets the ops hero take a vacation.

@wys1010 We’ve all seen that setup 😂 Hopefully Teable can give the ops hero a little backup

@wys1010 We know that ops hero well 😄 The goal is to turn their spreadsheet setup into a system the whole team can run together.

@nico_mandera Just ask GPT-5.6 sol to do it, 0 manual job! check then you can start building a new fully automated AI workflow in Teable !

@nico_mandera it’s more of a quick check than a cleanup job 😄 Linked records and attachments come along for the ride!

@nico_mandera Great question! Nico. No manual work is needed, unless you’d like to give everything a quick check before getting started.

@3rdmatter Exactly—your database shouldn’t just store the work, it should finish the work. That’s the shift. 🚀

@3rdmatter That’s a great way to put it 😄 Less switching between tools, more getting things done in one place.

@3rdmatter That’s the idea: an approachable database you can actually run your business on.

@galdayan Multi-agent collaboration can’t simply mean pointing two independent agents at the same table and hoping they don’t collide.

In Teable, you can separate agents by responsibility and tightly scope what each one may read or change. Every action is logged at the step level, so overlapping writes remain attributable and inspectable, and individual actions can be rolled back if needed. Agents also work from shared, live context rather than isolated copies of the data.

We don’t want to disguise concurrency as “collaboration,” though. Permission boundaries prevent many conflicts; logs and recovery make the remaining ones visible and correctable. More explicit coordination policies for genuinely overlapping agents are an important part of making multi-agent systems trustworthy at scale.

@galdayan Gal, every step an Agent takes is recorded and traceable, while the Agent keeps working to complete the task. If two Agents touch the same data, the change history makes it clear what happened.

@alexander_shishkov1 Thank you—you’ve identified the exact trust gap we care about.

AI may help build the app or automation, but the result isn’t just a hidden prompt rerun from scratch each time. The workflow is persistent and inspectable, every execution step is logged, and individual actions can be reviewed and rolled back. You can also tightly scope what each agent is permitted to read or change.

For more complex work, multiple agents can collaborate within the same team while keeping clear responsibilities and permission boundaries.

Our principle is simple: AI can be flexible in how it reasons, but its actions on business data must remain visible, controlled, and recoverable. Real PostgreSQL provides the trusted data foundation; Teable adds the operational layer required to let agents work on it safely.

@rusutraiancristian 

That’s a completely fair question. If your spreadsheet is personal, relatively small, and mainly used for calculations, you may not need to move it.

The problem appears when the spreadsheet starts running the business: multiple people edit it, records need relationships, access must be controlled, changes must be audited, and reminders, approvals, reports, emails, or custom interfaces have to be built around it. At that point, teams usually glue together more spreadsheets, scripts, automation services, and internal tools—and eventually nobody fully trusts or understands the system.

Teable keeps the familiar table experience but puts real PostgreSQL underneath it. Then it adds relational data, permissions, auditability, automations, custom apps, and AI agents in the same environment.

So the value isn’t “another place to store spreadsheet rows.” It’s turning business data into a trustworthy system that can actively run the workflow around it—without losing the flexibility that made the spreadsheet useful in the first place.

@rusutraiancristian Rusu, I get where you’re coming from. If your spreadsheet already does everything you need, there’s no reason to move it. Teable becomes useful when that same data needs to support a growing team, controlled access, automations, AI workflows, and custom apps.

@madalina_barbu Thanks for your support, Madalina!

@1mirul Amirul, thank you! Making AI genuinely useful in day-to-day work is exactly what we’re aiming for. We want Teable to become a natural part of how teams work with their data. Would love to hear how it goes when you try it on a real project.

@talhakhalidmtk Muhammad, good question. In Teable, each source can update on its own schedule while everything flows into the same workspace. Each run is tracked, so teams can keep their data current without coordinating every source manually.

@mathias_barboza Hi Mathias, thanks! Teable’s Authority Matrix lets you control access at the table, field, and record levels, including specific actions. Record history shows who changed what and when, while audit logs keep key operations traceable as more people work in the same base.

@buzzy_jade Yes! The table is the data foundation—and Teable can turn it into a real app with interfaces, workflows, permissions, and automations on top. Same data, no rebuild.

@buzzy_jade You can start with a table, then turn it into something your team actually uses—like a CRM, landing page or customer portal. Everything stays connected to the same data.

@buzzy_jade Yep 😄 Your table can power the app directly, and you can shape the interface around how your team actually works.

@erok_chen That’s exactly the point. Teable help get the work done. Contracts, approvals, reports, emails: real operations, not AI party tricks.

@erok_chen Glad that came through! We picked these examples because they’re the everyday tasks teams actually spend time on. If there’s another workflow you’d like to see, let us know.

@erok_chen Appreciate that. The real test for us is whether AI can take a recurring business task all the way to completion.

@andy2026 You absolutely get it. Teable is the all-in-one workspace where everything data work needs—data, apps, workflows, and AI—comes together and actually runs.

@andy2026 Marketing's hard part isn’t getting more data, but it’s keeping campaigns, leads, and follow-ups connected. That’s where Teable can really help.

@andy2026 Exactly. Teable brings data, apps, workflows, and AI into one system, so marketing teams can run campaigns, emails, approvals, and reporting in one place.

@tammytan516 You have to try it! Turn that customer onboarding idea into a working system and see how surprisingly effortless it feels. We think you’re in for a refreshingly smooth experience.

@tammytan516 Customer onboarding is a great use case 🙌 You can keep customer info, tasks, approvals, and follow-ups all in one place. Would love to see what you build!

@tammytan516 Love that this got you thinking about a real workflow right away. Customer onboarding would be a fun one to build in Teable. Let us know how it goes!

@itsluo You really get what we’ve been building. ❤️

Three years, thousands of customers, and countless workflows later—Teable’s new AI can now create the kind of custom apps and automations we once thought were impossible.

And the best part? Tables are still at the heart of it all.

Getting work done has never been this easy. We’re so glad to have you back with us on Product Hunt! 🚀

@itsluo This means a lot coming from someone who’s been self-hosting Teable for two years 🙌 Users like you helped us get here. Would love to hear what you’re running on Teable today!

@itsluo Thanks for sticking with Teable for the past two years. Hearing that this evolution feels natural from a longtime self-hosted user means a lot to us.

@peng_wood Thank your support !!

@peng_wood Really appreciate it, Wood! We put a lot into this launch and can’t wait to hear what you think after trying it 🙌

@peng_wood Thanks, Wood! It means a lot to have your support on launch day.

@kevin  @tanmaig Nice idea!

@karthik_venkateswaran it's also been so cool to see how many more people in the company (across functions) now "build" product!

@nk_kapur  wooot! ❤️

thanks for the feedback:

• recurring workflows section coming soon!

• just added a good usage section to get started with both for users and admins. you should see it on the sidebar

• artifacts can be made editable, just need to ask promptql to make it editable. tables being editable coming soon!

• external teams connect v1 is live! you can add external guests to private threads or rooms already which makes it super nice to collaborate with external consultants, vendors, customers, lawyers etc
  
  

@nk_kapur Thanks for being an early supporter and pushing us forward continuously! Have really valued all the feedback from you and implemented most of it :)

@dan_harvey and that's also the reason it feels like the rate of changes to the wiki (shared context) seems to be steadily increasing. which i found very surprising. I would have expected that the rate settles down to a small steady rate once all the shared context has been taught.

but exactly to your point, why would I stop after teaching it one thing. I'll move on to the next thing. done right i think it's like a continuous acceleration. too weird to comprehend what that means....

@galdayan Wikipedia style!

So promptql suggests disambiguation entries whenever this happens. It also has articles on each user and team as well. So that way depending on who's asking and why they're asking it's able to choose the right definition (like a human would) without getting confused.

These kinds of multiple definitions and disagreements are actually super common and honestly the point of the wiki is to capture all of that in one place and then use it well!

@nextmark changes are never automatically written back. changes are only suggested. a human always has to approve (or edit or reject) the changes. these changes are a precise set of bullets not a giant wall of text.

but most importantly the control mechanism is like wikipedia. so people who care about certain domains "follow" those topics. And that way any changes they get notified on. and if it changes then they can swiftly revert.

the last boss is that you can lock a page entirely, but that's usually not necessary. the more open the context is the more self-correcting it becomes.

And given the fact that every change has a human name attached to it, people don't want their names to be associated with sloppy or wrong info. The team and social dynamics keep things naturally clean :)

@zach_gold collaboratively going from question/answer to agent is my favourite.

Zach: "How many active users did we get from the PH launch"
AI: "hmm..not sure about active user definitiion. maybe @/tanmai knows"

Tanmai: "yeah, for social campaigns we usually do a has credit card on file and atleast one interaction in the first 6 hours"

AI: "ok, here's what that looks like"
Zach: "awesome, can you make that live refresh every hour and in case someone crosses more than 5 threads in the first 6 hours, help me draft a personal email to set up a call with them. ping me whenever this happens"
AI: "Done"


Collaboratively building an agent on the fly is just too much fun :)

@tmaleh_ Hey! Great question - so connecting to your tools and setting up on PromptQL is a 5-15 min task.. so PromptQL helps you ease over at the pace you want to. Also we don't move data - so you can connect to these tools and PromptQl works with them as it needs to.

The champions / our first users in a team - are usually the ones who push all work via PromptQL in that first week and start seeing the wiki once they connect their tools of choice. And once you're set up, you can also access PromptQL in Slack - and start tagging it in. The way you start off where the team mostly is and slowly start moving to promptql for deep work.

Does that make sense? I'd love to show you a demo of how we work in case you'd like to try it out for your team.

@honeymittal It was accidental! We started out building AI that'd be accurate when connected to company data.. and it soon became obvious that the missing layer was shared context. As we started to build out the system that would capture shared context- aka multiplayer AI, we suddenly realised we're built an AI native slack!

@rajoshi_ghosh  @suryansh_tiwari2 thanks! exactly right -- not having an AI that has amnesia everytime you talk to it is kind of nice :)

@tiru_selvan it's kind of cool how it makes you feel closer to everyone else in the team. even though you're using AI. Because you're like: 'wait, how did you know, oh because X taught it, so cool'.

@hasui0310 awesome! Try it out and let us know if you need any help getting set up.

@msethu Thanks for being an early supporter. I hope you've been able to unlock more usecases with PromptQL

@sabber_ahamed We hear this from customers all the while. There's a lot of really good shared knowledge, but some information should be need-to-know.

The wiki is also able to be scoped so learnings from scope-restricted threads end up in secured parts of the wiki. Only users with that level of access are able to pull them into context

@felix_josemon That's awesome to hear!

@abhijeet_singh4 yeah, so much of engg work ends up being the before (what should we build, why) and after (is it done, what should we do immediately after). the shared thread way of working really surfaces that in a whole new way.

@tanmaig  @nikhil_moorjani Thanks Nikhil! Appreciate the support :)

@omri_ben_shoham1 Yes, it gets folded back in. Everything durable worth learning in a thread is actively learnt and it is also shown to the user for review before "committing" it to the shared context.

You can see the entire change log of the shared context and tie it back to the originating threads as well.

@jackbogdan  @chrismessina This means a lot coming from you, Chris 🙏 And yes — 'keep the banter, outsource the visages' is exactly the point. Whenever you and Jack are ready, send a photo each and we'll build your twins so your next episode is basically just the two of you talking, minus the camera setup. Would love to see how it turns out.

@jackbogdan  @chrismessina Thank you so much for hunting us, Chris! 🙌 Since you’re experiencing the video podcast workflow firsthand, your feedback will be especially valuable to us. We can’t wait to see what you and Jack create with PodcastorAI—please send any questions or honest feedback our way as you try it!

@jackbogdan  @chrismessina Hi KARAN, 👋

Congratulations on your Product Hunt launch! 🎉

I checked out PODCASTERAI, and I really like what you're building.

I help early-stage startups increase their Product Hunt visibility through genuine upvotes, real user engagement, and organic app downloads—never bots or fake traffic.

Beyond Product Hunt, I can also help you reach a wider audience by connecting your product with relevant content creators and influencers from my network who have strong and engaged communities.

I believe your product deserves more visibility, and I already have a few growth ideas that could help.

If you're open to it, I'd love to share my proposal. It will only take 5 minutes, and I think you'll find it valuable.

Looking forward to hearing from you! 🚀

@parsons_wu_real I Karan, 👋

Congratulations on your Product Hunt launch! 🎉

I can help your product get 1,000 genuine app downloads and real Product Hunt upvotes—no bots, no fake traffic.

I also work with genuine creators who have audiences ranging from 100K+ subscribers, with many regularly reaching 1M+ views per reel/video.

My price for 1,000 genuine downloads is $1,000.

If you're interested, I'd be happy to share my growth plan. Looking forward to hearing from you! 🚀

@rhinogo You're reading our minds 👀 We've actually been building an MCP server for exactly this — pipe a transcript straight from your tool of choice into Podcastor, no copy-paste tango. Consider your idea officially on the roadmap (and mostly already there 😏)

@rhinogo Great suggestion! This is exactly the kind of seamless workflow we want to build—moving content from the tools you already use directly into PodcastorAI. Which tools would you most like us to support first?

@layla_carter1 That’s a fair concern, and it’s something we think about a lot. Our goal isn’t to give every creator the same AI voice or format, but to preserve what already makes them distinct: their own voice, digital twin, ideas, scripts, pacing, and preferred episode structure. Podcastor should handle the repetitive production work without flattening the creator’s personality.

@henry_habib Thank you! The host’s lip movements stay closely aligned with the pacing and intonation of the audio throughout longer episodes. We also use intelligent alternating shots to keep long-form videos visually engaging, and we’re developing a new model with context-aware emotions and body language to make the digital twin feel even more natural over extended conversations.

@henry_habib Appreciate it, Henry 🙌 Because the avatar tracks your actual audio, it stays in sync across the full length — and we support single videos up to 30 minutes, so longer episodes are covered. Thanks for the kind words 🙏

@auula_ Thank you! That’s exactly the practical problem we wanted to solve: creators shouldn’t have to re-record an entire episode just to make it work as video. With Podcastor AI, they can turn the content they’ve already created into a video podcast while preserving their voice and presence.

@joviechen Not yet, Jovie — no preview before the final render at the moment 🙏 But it's a great suggestion, and one we'll take away and evaluate. Really appreciate you raising it 🙌

@omri_ben_shoham1 That’s a very real challenge, especially in long-form content. We’re not claiming the digital twin is completely indistinguishable yet, so we built an intelligent alternating-shots system to vary the framing and keep longer episodes from feeling visually repetitive. We’re also developing a new model that adds context-aware emotion and body language, making the host feel much more natural and expressive over longer stretches.

@ryancheng Hi Ryan! Thank you so much! The digital host’s lip-sync closely matches the pacing and intonation of the original audio, so uploaded episodes feel natural on screen. We’re also developing our next-generation model to bring more emotion and expressive body language to digital twin hosts, which should make the final videos even more engaging.

@ryancheng Appreciate it, Ryan 🙌 The avatar is animated straight from your uploaded audio, so lip-sync and pacing track the original episode instead of being re-timed — pauses, emphasis and all. It's the kind of thing that's easier to feel than to describe, so I'd genuinely love for you to test it with one of your own episodes and tell us how close it lands.

@mingyouagi Yes, absolutely! You can upload your existing recording, keep the original audio, and add an AI host as the video layer—there’s no need to record or regenerate everything again.

@fayann Exactly! We built this workflow for creators who already have great audio but don’t want to record everything again for video. They can reuse the original episode and turn it into a publish-ready video podcast with an AI host.

@codeandsea Thanks Brent — and it's a totally fair ask 🙏 Subscription fatigue is real, and 'buy credits that don't expire' is something we've been actively debating internally. No firm date yet, but you're not the only one asking, so it's very much on our radar. In the meantime it's free to try, really appreciate you taking the time to think it through with us.

@colton_hayes2 Not yet. We currently support individual accounts, but team accounts and multi-seat access are already in development. We know Podcastor becomes much more useful when an entire content team can collaborate, so this is an important part of our roadmap.

@os_ishmael Thanks, and that’s a great point. If you use Podcastor’s TTS, you can precisely adjust pacing and add pauses to shape the delivery. For uploaded NotebookLM audio, we can extract the script so you can revise the spoken content rather than simply accepting the original output.

Fine-grained control over host reactions isn’t available yet, although the current lip-sync is already very strong. We’re now developing digital twins based on real recorded video, along with a new model that drives emotions and body language from the tone and context of the content. This is especially important for educators and experts, where preserving the creator’s authentic delivery matters as much as visual polish.

@erok_chen The biggest difference was not simply faster coding, but making our entire development loop more reliable. GPT-5.6 Sol with Codex became especially effective at understanding requirements and navigating complex, cross-module changes, so we could move from planning to implementation and verification with far less back-and-forth. We also use it in production monitoring: agents inspect failed tasks and logs, trace issues across the database and codebase, then prepare fixes for our team to review. That has helped us iterate faster while catching risks much earlier.

@ea_z Absolutely. For research-heavy formats like true crime, creators should be able to focus their energy on accuracy, storytelling, and responsible reporting, not spend another full-time job producing the video. That’s exactly the production burden we want Podcastor AI to take off their plate.

@huisong_li Thank you so much, Huisong! We really appreciate your support and kind wishes. It means a lot to our team on launch day! 🙌

@aidan_codefox Thanks! We completely agree that chemistry and friction are what make a two-host conversation worth listening to. That’s why we offer three formats: Deep Dive, where both hosts unpack a topic together; Debate, where they take opposing sides and challenge each other; and Storytelling, where one leads the story while the other asks questions and keeps it moving. So if you want real disagreement rather than two hosts simply taking turns agreeing, Debate is built specifically for that.

@buzzy_jade Thank you! That’s exactly the idea: creators already have so much valuable content sitting in their libraries. We want to help them give it a second life in video, without having to start from scratch every time.

@eric_t_ Exactly! Creators should be able to spend their time shaping ideas and telling great stories, not losing an entire week to editing. Our goal with Podcastor AI is to handle the production work while keeping the creator’s voice and presence at the center.

@eric_t_ that's the dream we're chasing — you keep the ideas and the storytelling, we take the editing week off your plate. Try it with your next episode and tell us if it delivers 👀

@new_user___1802026258df93399c32311 Thank you so much for actually trying PodcastorAI and sharing such thoughtful feedback! It means a lot to hear that the audio quality and customization exceeded your expectations. And the cat AI host absolutely made our day—we’d love to see it if you’re comfortable sharing! 🐱💜

@madalina_barbu Thanks Madalina 😊 Would love for you to give it a spin — bring any script, link, or PDF and it'll come back as a full video episode hosted by your digital twin. Curious what you'd make with it!

@fei_li5 Exactly, Eric — the "everything afterward" is where most creators stall out. That gap is the whole reason we built Podcastor. Thanks for getting it 🙏

@xiao_zhang9 Thank you! We didn’t want creators to be locked into a single format. Some ideas work best as a focused solo episode, while others become much more engaging through conversation, questions, or debate between two hosts. Podcastor is designed to support both, so the format can follow the content rather than the other way around.

@xiao_zhang9 Thanks Xiao 🙌 Exactly — real podcasts aren't one-size-fits-all, so we wanted both the solo monologue and the two-host back-and-forth to feel natural. The host handoffs in two-host mode are the part we're most proud of. Would love to hear which format you'd reach for 👀

@tammytan516 Hi Tammy, that's the whole reason we built it this way 🙏 Most tools erase the creator; we wanted the opposite — your face, your voice, your channel, just without the camera setup. Glad it landed for you.

@tammytan516 Exactly! We want creators to stay focused on their ideas and storytelling, while Podcastor AI takes care of bringing the content to life visually, without losing the creator’s presence.

@charlenechen_123 Thank you! We currently support episodes up to 30 minutes, so we don’t have a reliable render-time benchmark for a full one-hour podcast yet. Longer generation is already on our roadmap and should be available next month. Once it’s released, we’ll share accurate timing based on real production runs rather than guess today.

@ayushtweetshere Thanks Ayush!

Been in SEO space for 10 years now.
My workflow moved between duplicating and changing filters across tab, joining multiple Sheets just to get the right insights. I have longed for something like this. The existing tools never changed my workflow, but this one did.

Super excited to share it with the amazing guys at Product Hunt!

Looking forward to your feedback and support

@ayushtweetshere Hey, pulling GSC and GA4 into one place is the boring problem nobody wants to solve. When I've worked with SEO agencies the crawl reports were always 200 issues long with no sense of which five actually moved traffic. Does CrawlRaven rank findings by likely impact, or is prioritizing still on me?

@neelptl2602 Thanks for the support.

@axebarma  It discovers new keyword opportunities based on the content published, what keywords it already rank for and what can potentially rank for more. The analysis becomes even richer if you add your target keyword list with data from tools like Ahref. With time, you get more data and hence new opportunities are identified. Let me know if this answers your question.

@axebarma GSC data syncs from Google automatically every day, so findings update on their own as your changes take effect. No need to re-run anything. Just note Google reports search data with about a 2 day delay, so give fixes a couple of days to show up.

@hannesh Thanks. It looks at positions 4 through 20, so the bottom of page 1 and the top of page 2 both count. The closer you already rank, the higher it lands on your list, since a jump from 6 to 2 wins more clicks than a climb from 18. We skip anything past 20, since those need real content work. And we want at least 50 impressions so the term is one people actually search.

Where would you draw the line? Some folks want to reach past 20 for niche terms that still convert. Others want the list tight around fast wins. Which fits how you work?

@alex_watson2110 Thanks Alex!

@aidan_codefox The ranking uses fixed rules, no AI. It looks at three things from your own Search Console data. How much search demand a keyword has, how close you are to page 1, and how many clicks you should be getting but aren't. Since it runs on your data, your list is yours alone.

Tell me how you'd want it though. Would you rather set the rules yourself, or have it watch what you changed and whether your rankings went up?

@ansari_adinThe crawl is what we're building next, so I won't pretend a 200-check spec sheet exists today. That number comes from our v1, which was a full technical SEO crawler before we rebuilt around Search Console data.

I'm not trying to beat Screaming Frog or Sitebulb on check count. We're building the part they leave to you: ranking each finding by whether it hits a page that actually gets traffic and sits close to a win. A broken canonical on a dead page and one on your top traffic page look the same in most crawlers. We tell you which to fix first, using your GSC data.

So what we catch that they miss is the impact ranking. Happy to send you the real breakdown once it ships.

@vollos We only request the readonly scopes, webmasters.readonly for Search Console, and analytics.readonly for GA4 (asked separately, only if you open the GA4 tab). No verify, edit, or property-management access at all. We literally can't change anything in the client's Search Console, only read the performance data.

On disconnect, the access lives in the client's Google account, so if they revoke it there (or your Workspace admin does), our token stops working immediately and the sync halts. We hold nothing that survives that. If you want your data removed on our side too, account deletion wipes the stored tokens and history.

@nishant_modi Ahrefs looks outward, competitor backlinks, the full keyword universe, what everyone else ranks for. CrawlRaven looks at your own Search Console data and tells you what to do with it. Which pages sit one push from page 1, where two of your pages fight over the same term, which pages are quietly losing traffic.

Most people keep Ahrefs for research and use us for the weekly "what do I work on now" list. You can even import your Ahrefs keywords straight in.

@sabber_ahamedYou hit Resolve, and the detectors do the rest. They re-run on every daily sync. If your fix held, the item stays resolved and drops out of your Attention list. If the problem comes back later, it doesn't quietly reappear as a new finding. It comes back marked Returned, so you can see it regressed and the history stays attached. No guessing whether it's the same issue twice.

Note: the crawl and GA4 are rolling out next, so today this runs on the Search Console side. The same Resolve and Return logic will cover the crawl fixes when they are released.

@mdshadabalam3 Thanks for the support.

@denitsapenchevavaltchanova Thanks. Right now the list leans toward quick wins, because that's what Search Console shows best. A page stuck at the bottom of page 1, or a keyword you rank for but never gave its own page, rises to the top. Small effort, and your own numbers already prove the upside.

The heavier technical fixes come from the crawl audit, which we're adding next. Once it's in, the plan will rank both together

@irahimiam You can add multiple sites on one account. You can also import multiple site by connecting to multiple search console. The current plans allow you to add 3 sites for $49 and 10 sites for $99.

@clemente_lopez1 I agree with your use case. I have similar experience. Highest intent terms often show very minimal data in GSC. We have plans to add controls for the user where they can decide how they want to prioritise. Consider this added to roadmap.

@roguetink You will be able to see ranks, positions and pages for any given keyword. It recommend how to target keywords and in what priority

@roguetink Thanks. Cannibalization was one of the original itches, my cofounder's agency was doing exactly that export-and-cross-reference routine you describe.

On stale keywords, the list isn't taken as given. Every keyword is matched against your live Search Console data, which refreshes daily, so one that stopped mattering shows its real current numbers. And you can mark any keyword as ignored, it drops out of your working list without being deleted, so the list stays clean.

An automatic "this keyword is dead" flag isn't there yet, but it's a really good idea. Would love to hear what it finds on your site this week.

@omri_ben_shoham1 Good question. GSC is the source of truth for positions, since it's Google's own impression-weighted data from real searches, while Ahrefs/Semrush positions are spot checks of one location on one day. So when they disagree, we trust GSC.

Your Ahrefs/Semrush export contributes what GSC can't tell you: search volume and difficulty for keywords you don't rank for yet. GSC says where you stand, your keyword list says what's worth chasing, and the ranked list combines both.

@gamifykaran Thanks. Yes, no other tools needed. It pulls keywords straight from your own Search Console data and shows you the ones you're already appearing for but haven't written a page for yet. Adding your own keyword list is optional.

Don't have a sample report to link right now, but happy to share a few screenshots of what the suggestions look like. There's also a free plan, so you can connect your Search Console and see it on your own data.

@shivam_vyasThanks. It's ranked by traffic potential relative to how close the win is. We look at your real impressions, your current position, and the gap between what you get and what you'd get with a small move. So a page sitting at position 11 with high impressions outranks a bigger but more speculative opportunity.

Note: Search Console is live today, GA4 and the crawl audit are coming next.

For GA4 google verification is pending.

@parthkoshti Yes we do. Available on top tier plan.

@shreyans_assistiv Really like that AskCodi combines an AI coding assistant with an AI CTO that can plan, delegate, and choose the right model for each task. The BYOK support and OpenAI-compatible API make it a flexible option for teams and developers. Good luck with the launch!

@shreyans_assistiv Congtrats on an awesome launch! Excited to try the new version of AskCodi.

Both routers and sandboxed openclaws are all the rage atm. Seems like you've put them together in a single, inbox-style interface, which seems really cool. But where does it go from here? Can this really become like an AI CTO, that designs your infra, picks the tools, innovates on project concepts, etc?

@zaczuo Currently inbox is for blockers but the feature map of each project stores the decision made by codi + every tasks details board also highlights them but they definitely also have a place in the inbox. Thank you for the feedback, definitely going up in the next update.

@zaczuo this is a good feedback. We agree that important decisions deserve a dedicated view in the inbox too. We’re adding this to an upcoming update, thanks!

@artstavenka1 Not boring but suitable. Our algorithm uses public data to see which models perform well for which task. It is about reaping the cost benefit without loosing the accuracy too much. Plus the systems in place (skills and supervisor) boost the accuracy instead of dropping it. It is not a pure llm router, llm just classifies the task as to what kind of task it is and then we choose the best cheapest model for it. Hope that cleared it up a bit.

@artstavenka1 To add to Shreyans’ answer: routing isn’t a one-way decision. The engineer’s output still has to satisfy the task’s acceptance criteria, tests, and review. If the result exposes unexpected complexity or fails validation, Codi can treat that as evidence that the task was misclassified and escalate it to a more capable model. The savings come from starting efficiently, not from being committed to the first model choice.

@andrasczeizel You captured our thought process so well. There are an initial set of ruleset like problems including monetary and security - those are the ones escalated to human. Rest Codi answers as a supervisor. This mainly abstracts the technical decisions away.

The PRs are small enough to do no major harm :)

@andrasczeizel Codi works against the project charter and feature map it builds with you at the start. That gives it a clear definition of what is in scope, what is out of scope, and what success looks like for each task.

The work also stays contained in isolated git worktrees and produces reviewable diffs. Tests, acceptance criteria, and a separate review pass have to succeed before Codi considers a task complete. If the work changes product scope, introduces a security or spending decision, or conflicts with the original charter, it comes back to you through the inbox.

We want unattended work to mean "useful progress waiting for you," not "surprise, your product changed overnight." Thanks for raising this because that trust boundary is one of the most important parts of what we’re building.

@raunak_bhansali Thanks! 🙌 Great question. Isolation is at the worktree level: each task gets its own branch and git worktree, so a feature, a bug fix, and a refactor never share a working tree or step on each other. Right now Codi runs them isolated and coordinates them rather than firing all three at the exact same instant, and true simultaneous same-repo branches are on the near-term list. Across projects it's already fully parallel.

@piyush_bhansali2 The system has been tested to build out complete applications from start and we have seen certain behaviours from agents that is preventing it from drifting 

1. Codi truly acts as a supervisor - reviewing not just the PRs but also the plan

2. Agents are escalating to human users when some request doesn't align with the vision or user experience

3. Daily briefings shows what is built and what is left at any given time. Purposefully built to always stay updated of what agent is working on and correct as fast as possible

Agents are problematic in that format but alignment is core of the product we have built.

@sm0r3ll It works with claude code, openai codex and rest of the models on the market are served through askcodi gateway at provider api prices. BTW loved your reaction :D

@shreyans_assistiv  @jayant_surana1 Yes, it is true multi-project orchestration. You can run multiple Codi sessions across different projects simultaneously, while approvals and questions from every session arrive in one inbox. You do not need a separate terminal or agent manager for each repo.

@omri_ben_shoham1 Codi works in bounded tasks, not one open-ended overnight run. Each task has acceptance criteria, tests, and budget limits. If validation repeatedly fails or the work drifts from the project charter, Codi pauses that branch and surfaces it in the inbox instead of continuing down the wrong path.

@adam_maceachern1 Kimi K3, GLM 5.2 and Sonnet 5 as the more effective worker models. Fable or Sol for supervisor. The system is absolutely killing it :) Thank you for your support!!

@scott_davidson_jr It would be still per project, you would have to still manage atleast one agent per project. Fine at one, starts causing issues as you start working on more. This is not just multi tasking, this is a freelancer on steroids. Plus we have built in systems which actually remove you as a bottle without going off rails. Of course you could build out this system will skills, but skills are still suggestions for agents in long loops.

Do you currently run a multi agent auto model system in Cursor or Conductor? Last time I saw someone's stats, autorouting was still 98% opus. Would love to learn more.

@jernej_jan_kocica by actually dividing the whole system into supervisor and worker, we are eliminating long runs. The entire project is divided into features and modules. Agents are fired our features -> agent comes back with a PR -> supervisor reviews and merges. After every 4-5 PRs, board and codebase is re-synced so that supervisor doesn't go off plan.

@arun_prasad06 Thank you for giving us a chance. Most of the tools are designed for the 200 USD plan, we are aiming to optimise for a market where everyone doesn't have 200 dollars every month to spend on AI

@ankita_singh27 Codi adapts, checks if any task are dependent on the change and updates them accordingly. It acts as a true supervisor.

@aditya_harish_2002 It's actually pretty good at mapping repos, I have personally tried mapping 3 of my messiest projects with multiple repos and it covered what I built better than what I could explain

Do give it a try and let me know what you think of it :)

@sabber_ahamed Definitely, it is tricky to design failsafes mid task. So how we have do it is each agent works with the task in breakpoints with Codi supervising along the implementation. Plan and PRs are already reviewed by Codi.

@korbinian_abstreiter It is a combination of LLM Routing + benchmarks + public dataset of model performance over time. We keep experimenting and this adaptive routing is currently showing the best performance.

@byalexai So all you need to configure are the providers - anthropic models through claude code, openai models through openai codex and rest of the models on the market through askcodi gateway.

Model selection is auto by codi based on the best model to complete the task with as much saving as possible. Routing is always of today's performance of the model. So on the day of launch Sonnet 5 would definitely be better than Opus 4.8.

@byalexai To add one detail: your provider setup defines the pool Codi can choose from. Within that pool, Codi looks at the task type, required capabilities, cost, and current model performance. The selected model still has to satisfy the task’s acceptance criteria and review checks, so routing is not based on price alone. If a task turns out to be more complex than expected, Codi can move it to a stronger model.

@ankita_singh27 We have a few things built in that improve output quality while reducing cost. Codi acts as both a supervisor and reviewer for the other agents, and it also routes each task to the most cost-effective model using our routing algorithm. The routing doesn't rely only on benchmark scores - it also considers recent model performance and task difficulty. So instead of defaulting to Fable or GPT-5.6 Sol for everything, it'll often choose models like GLM 5.2, Kimi K3, or MiniMax M3 when they're sufficient. Across our benchmark suite, we've seen roughly 50% lower inference costs with only about a 0.5% drop in task completion.

@adamkamaneh At the moment, we're focused on building the core experience. Integrations are on our roadmap, and we're actively working on expanding them.

If there's a specific tool your team relies on, we'd love to hear it.

@derek_julian That’s a great point. We don’t want AI agents to become another source of noise 😄

Agents don't act on their own. They only do what they're assigned, inside the room and role you place them in.

So an agent never wanders into a thread to offer an opinion nobody asked for - if you want a moment to be human-only, it just is.

Think of it less like a bot listening to everything, and more like a teammate who only works on the tickets you hand them.

@dustin_warren Exactly! This was one of the biggest pains we experienced ourselves.

@kyle_bennett6 Great Question !

Pricing's the hardest part and we're being careful - we're shaping it with our current users right now, so I'd genuinely welcome your take too.

Our intent is - never charge people who are just participating - clients following along, chatting, leaving light comments stay free. We only charge for the heavy lifting (calls, storage, intensive Board use), which means just a small core of each team pays, priced to actual use rather than per head. So as a studio grows, most of your people cost nothing - and the total stays at or below what you already pay for just one of the tools in your current stack.

@owen_parker4 Thanks Owen!

Calls are built right in - no Zoom or Meet hookup needed. They run natively inside a room's Board, so you jump on a call exactly where the work already lives: you're talking and interacting with the same files, tasks, and media in real time, instead of screen-sharing a window from another app. That's the whole idea - the call happens where the context is, not in a separate tab.

@priya_kushwaha1 Priya, thanks for the question! 👋

Versioning is decoupled from the Board - it's just a canvas that references files, not a version history itself.

New versions get created either when a task/stage is finalized (each retake/approval bundles into a version), or when someone manually uploads a new overall version at the room level (e.g. "Episode 3 complete").

Iterating on the Board itself doesn't fork history - only finalize/approve or an explicit new upload does.

@ansari_adin Fair challenge Ansari— and honestly, "agentic" here is more literal than the examples you gave.

It's not an invisible AI quietly flipping task statuses in the background. Our agents are actual teammates you assign work to — they join a room, take a task, do it, and hand it back through the same review loop a human would. You're not babysitting an assistant; you're delegating to a worker.

Aurora is the clearest example — a manager you just talk to. Tell her "set up shots SH001–SH010 and split them across the team," and she'll batch-create all ten tasks, round-robin them to your performers, each with its own description and deadline. She can spin up rooms and stages, invite members, reorganize and reprioritize tasks, set up review chains, and pull deliverables back into the conversation. Two things keep it safe: she only ever acts with your permissions (never more than you could do yourself), and every change she proposes shows up as a preview you approve or reject before anything actually happens.

Then there's Iris, a scriptwriter — give her a brief, she drafts and revises. Lan, a concept artist who creates and iterates from your feedback. Plus quieter helpers doing translation and transcription, so notes and scripts cross languages without leaving the room.

So "agentic" isn't a buzzword for us — it's the whole point: AI teammates that actually do the work, accountable to the same review cycle as everyone else. Some are live today, more landing right after launch. Early, yes — but real. 🙂

@amraniyasser Thank you for your feedback! 🙌

That’s exactly how RunEvr was born - from our own daily struggles managing creative projects.

We’ve been using it internally for months, and AI agents became a natural part of our workflow.

Excited to see how others use them too!

@omri_ben_shoham1 Really sharp question - this is exactly the gap we obsess over. You're right: "make it pop more" dies the second you flatten it into a ticket.

So we don't. That feedback lives right where the work is - a comment or voice note pinned to the exact frame, so the vibe and nuance stay intact instead of getting squeezed into a status field. There is a light structure underneath (task → take → review → new take), but it's just there to track where things stand - whose turn, needs another pass - not to define what the feedback means. When an agent picks up a revision, it's working from the real note on the real asset, then hands a new take back for review. Human-in-the-loop, not a checkbox.

So yeah - there's a skeleton under the hood, but it's deliberately thin. It keeps the messy back-and-forth moving, not sanitized.

Honestly though, you've given us a really good angle to keep chewing on - it's exactly the kind of thing we want to keep thinking deeper about. Thank you for that. 🙏

@better_shaya Thanks so much Shaya, glad that resonates - that context-switch fatigue is exactly why we built RunEvr this way.

On real-time collab: multiple people can be on the same asset in the Board at once, live cursors, comments, and reactions land in real time, no screen sharing needed - it feels more like being in the same room than passing a file back and forth.

Would love to have you try it and hear what you think.

@deepaksingh09 Thank you so much!

And yes, the “losing context, not files” problem is exactly what pushed us to build RunEvr.

The most surprising use case so far was actually outside of creative work 😶

One of our 3D animators started using Aurora (our AI coordinator) not only for project-related tasks, but also to create a weight loss plan, organize his routine, and track his progress. And the funny part - it actually helped him lose weight! 😂

It was a great reminder for us that AI agents can become personal assistants in many areas, not just work.

@aidan_codefox Thanks Aidan, appreciate you asking this!

We're not chasing all-in-one to beat specialists - the bet is context preservation. Our Forum and Drive aren't trying to out-Slack Slack or out-Dropbox Dropbox, they exist so your people and assets live in one room instead of scattered across dedicated tools. The one area we're pushing hardest to be best-in-class: the real-time Board - creatives reviewing media together live, no screen sharing, actual data sharing.

@busmark_w_nika Thanks Nika! ❤️ Really happy you like it!

It has a bit of the Slack feeling, but our idea was to go beyond chat.

With one click on a project, everything flows to your screen - tasks, media and materials, chats, calls, reviews, and all the context your team needs to keep creating.

@nadiia_prokofieva Thank you! ∞

The biggest challenge was that we were building RunEvr while running a full Animation Sudio.

RunEvr actually came from the pain of managing real production processes. We were solving our own problems every day, while at the same time building an entirely new environment around them.

Balancing Product development with running the Studio was definitely the hardest part - but it also meant every feature was tested in real production before becoming part of RunEvr.

@adriancia Exactly! I want Moxie Docs to be unobtrusive but deliver real value. The passive context + conventions & doc actions our MCP gives agents means docs are also being improved / maintained passively during the week by anyone using AI tools, the Friday Recap is the last line defense closing that gap.

@paul_crinigan Absolutely - and for the MCP when we index a repo we pull out and identify conventions (these get updated periodically as they can also change over time) and any agent requesting doc info gets the documentation that exists itself, it can provide or identify any source / cited files related to the doc (for either continued examination or for identifying if work it did may have affected a doc in the repo), and any codebase context that is relevant! You can check out more on our MCP here if you're curious: https://moxiedocs.com/mcp

@rachid_abadli Every opened PR triggers a pass / check! Along with any AI agent utilizing the MCP, and routine checks every few hours. If docs drift from code changes - we find, fix, and surface that!

@galdayan Great idea! We do not check for this specifically yet, but I am working on it right now 🫡 we will flag RFCs / ADRs / Plans and detect when the implementation is done, then flag them for either removal or rewrite / combine with a relevant doc surface.

@vollos You can choose what repo docs get written to (where we open the PRs to) and that can be either public or private - and separate from the main repo or alongside your code. From there we have a layer of publishing before the doc goes live.

So the workflow is we propose a doc, you review / approve, any approved you click a button and we open a PR adding them, once you merge the last step is just clicking publish. At any time you can revoke published docs (they’d still be in your codebase though, just taken off the site).

@nextmark We add guidance to the AGENTS.md for any repo connected (we propose the changes via PR - you merge), expose MCP tools, and an agent skill to instruct agents to update any docs in the repo that they changed the behavior of, or to add new docs (or remove) if they add new features or fully remove described ones!

@omri_ben_shoham1 We run via the checks in GitHub as well as leave comments in the PR - so from the GitHub side you can configure a passing moxie docs check to be required. But our Friday recap is meant to solve that directly as well, every Friday we open a PR automatically that fixes, deletes, updates, etc any docs that may have drifted from recent changes or weren’t already documented.

@masterbek we don’t make any changes on your codebase! All documentation updates, changelog entries, and public knowledgebase pages are opened via a PR request to your connected codebase, or require a user to 1-click to publish (for changelog / knowledge bases).

We understand AI can make mistakes and the only thing worse than no docs is wrong docs. We surface suggestions but leave the power to decide when they ship in the hands of maintainers.

@grace_lee26 that’s exactly what our focus is! Everything runs continuously but nothing happens automatically - the control to review and merge is 100% in your hands. Moxie Docs does the heavy lifting surfacing opportunities, creating changelog items, opening PRs updating docs, and aligning / checking engineer PRs, etc but the final step of merging is always in the hands of repo maintainers.

The only thing worse than no documentation, is incorrect documentation. We take this an extra step with the new public Knowledgebases since it’s an entry point for consumers - you first merge our proposed docs into your repo (no vendor lock in), then 1-click a button to publish them live on the knowledge base.

@dhiraj_patel5 indexing starts immediately and begins surfacing insights / opportunities immediately! Within minutes the MCP will have codebase context and conventions start getting populated.

For a full index depending on repo size could be up to 5/10 minutes - but the value is that it’s never a one time thing, we continually improve / re scan over time - code changes faster than ever with AI tooling so we directly connect to the tools where they work (AGENTS.md, skills, MCP, etc) so our detection has to keep up pace with any repos development speed.

@irahimiam We run checks as GH native tie-ins so you can set it as a requirement they pass! Otherwise it will flag via comment + the failing check with details.

Moxie Docs can also be configured to automatically rewrite any PR description to match a standardized format so it’s easier for humans to review / orient any PR.

@aidan_codefox really love this question! Moxie Docs doesn’t propose code changes so at this time we scan and index code only to understand conventions, overall context / behaviors, and capabilities (which get fed into documentation generation and our MCP), and since code is what’s running - that’s the source of truth for docs. I personally use & suggest code review tools that would catch (probably in the same PR as a doc update) that the code may have bugs or be incorrect.

What we would help with though is that reverse side - if you have your agent fix a bug, a normal flow would mean it fetches codebase context and scans similar files and tries to figure out conventions (all token burn) but then fixes the bug - maybe after some follow up prompts. With our MCP it gets your conventions and context upfront, when it fixes the bug it can automatically fetch what docs already exists, scan those only to identify if it needs to update them, and then update them for you in the same session.

@codeandsea great question! We do offer free support for qualified open source projects, but a big chunk of the value is the MCP itself. We categorize your code based conventions, docs, context and surface it all directly to agents via skills and an MCP - so even if docs only update periodically (which often is not the case) the MCP will continually provide benefit by giving AI agents direct context and conventions.

If you’ve ever had to say “follow our conventions” or ask AI to update docs it immediately caused to drift from truth, that’s what we solve for.

@uussamaab  Looks solid, having custom prompts per app instead of rigid presets makes a lot of sense.

@sabber_ahamed Good question! Command Mode isn't guessing from context, you trigger it explicitly with its own hotkey, so the app already knows "this is an instruction" before you even start talking. That's what separates it from the other two modes: Basic Mode just transcribes exactly what you say, no editing at all, and Smart Mode cleans up rambling and filler words into polished prose, but still treats your speech as content to write down. Command Mode is different, it treats your speech as an instruction and generates the output for you.

@nishant_modi No Mac yet, it's on the roadmap. And free because it runs on your own Groq API key, their free tier is generous enough to cover daily use without hitting limits.

@omri_ben_shoham1 I actually built Wispro with personalization and flexibility as the first driver, specifically to solve this. You can set custom writing styles with your own instructions, so Smart Mode isn't just generic cleanup, it reshapes your rambling into whatever style you've personally defined. Would really love your honest feedback when you try it, and whether it actually solves this for you

@swebliss1 Yes! Wispro is powered by Whisper large-v3, which supports a wide range of languages, including Swedish

@sergbmw Thank you! That's a smart addition. Noting this down.

@busmark_w_nika Thank you! Yes, exactly what Command Mode is for, just say 'write me a tweet about XYZ' and it drafts it on the spot 🙌

@Quaso  @giordano_lucas "just describe something you're tired of doing by hand" - this

the web is UI over-load, context-switching uses mental bandwidth, Quaso frees it 🕊️🕊️ just start with a prompt

@Quaso  @samatnotte 

How can I buy or try it ?

@Quaso  @samatnotte quaso-maxxing

@lak7 hey! quaso prefers picking up integration if they exist and your task is feasible with what they expose through their APIs/MCPs etc. if this is not possible, browser enters the play! both can work together on a single automation as well

@noctis06 , great question.

If an integration exists, the agent picks that up by default, it's faster and more reliable. If the integration can't do what's needed, or none exists, it falls back to browser automation automatically. You can also tell the agent to prefer one path over the other for finer control, it's not purely automatic if you want to steer it.

For auth, we use Notte profiles: you log in once, browser state (cookies, session) gets saved, and future runs reuse it instead of logging in fresh each time. If a session expires, the agent will prompt you to log back in rather than silently failing.

On the browser side, we're not starting from scratch, that's the whole reason Notte exists. A year of building stealth browser infra for agents (residential proxies, fingerprinting, session persistence) carries straight over into Quaso.

It's still not bulletproof. Heavy bot detection sites (aggressive fingerprinting, active challenge-response) can still trip it up, that's honestly the harder 20% of the problem. But it's a much stronger starting point than agent tools building browser automation for the first time on top of a launch.

@noctis06 Great questions! Think Lucas responded it all. Meanwhile, did you get the chance to try?

@parthkoshti Thanks Parth! Looking fwd your feedback!

@ogandreakiro pronounced qwah-so 🥐

hey @aryan787544, I think the first sticky workflow is the one that closes the gap between conversation and execution.

We live in Slack all day, and that's where bugs, feature ideas, and "we should do this next" moments naturally surface. Quaso turns those messages into structured work, tracks it in Linear, and, because it's connected to our codebase and the same tools our team uses, can help close the loop by starting on fixes and features.

Beyond that, we use it daily to be reactive to a bunch of recurring stuff: customer support, Sentry errors, payment issues in Slack, automatic replies for churn signals, meeting prep and enrichment, user behavior analysis across Supabase and PostHog. All things we used to do manually and just offload to Quaso now.

That's made us dramatically more reactive to ideas and incidents: Quaso is present right where work already happens, rather than being another destination the team has to remember to open

@aryan787544 Following up Lucas response - did you get a chance to try? :)

@alexander_shishkov1 Great question. Honestly the answer on the quality of local models varies a lot between models since you can run something really basic as a small language model or some of the newer more capable models but it depends on your hardware.

About the the permission/sandbox boundary the great thing is that it works exactly as Codex, you make the call if you allow things at the OS level, or at the app level, if you want to allow every single time, or if you want to go full Auto mode. You make the calls.

@arnav_salkade Yes, that's the great thing. It runs on the same harness as OpenCode and with offline local models!

@noctis06 Thanks. It is more than just STT, since it needs to understand pauses, utterance, tone and interruptions. It uses realtime voice apis and it has access to tools to launch/edit/read real task threads and manage it exactly like you would.

@jogindertanikella Congratulations on the launch!!

@omri_ben_shoham1 Thank you for your comment. Linking live supplier/distributor pricing would be a natural progression once there's some traction. Right now, it's a rough estimate and will vary based on quantity.

@arnav_salkade Thank you for your comment. Right now, it's just to give a rough idea of what building hardware entails.

@ayda_golahmadi Thank you!

@chilarai Thank you for your comment. I can help out with the next steps too. :) It's part of my day-job as CEO of T-Works, India's Largest Protopying Centre and Manufacturing Incubator.

@thys_beesman Thank you for your comment. I wanted the reports to feel less like they came from a spreadsheet and more like they came from an experienced hardware engineer. So I used my 23 years of experience to shape the prompting and the evaluation rubric. The judgment isn't arbitrary, it's grounded in practical engineering gates. As more engineers challenge the reports, I'll keep refining that rubric.

@galdayan Thank you for your comment. If you say a Nuclear Power Reactor, it's going to show as Not Viable. :) So all are not having the same certainty.

@adamroman Really impressive, Adam! This solves a pain point I run into all the time. Having to stay glued to my laptop just because an AI coding agent might need my input breaks the whole workflow. Being able to monitor and continue agent tasks from my phone is a huge quality-of-life improvement. Looking forward to trying Cosyra!

@adamroman Great, now I can't even use "I'm away from my laptop" as an excuse to stop working.

@terminal_candy Yes. We don't have paywalls for features, only for the compute, because that is a real operational cost for us. 10 hours are included in the free trial, or you can use it without a credit card. There's a free version that includes 1 hour of compute per week. You can use the app as much as you want, though.

@terminal_candy That's exactly the gap we were after. When an agent needs input the prompt comes to your phone and you answer it right there, no walking back. And yes, it's the full terminal, not a cut down view. Same shell, same TUIs, scrollback and all.

@ethan_cheng I think I optimized for the social aspect more than levelling. That does make sense though I was considering adding cosmetics to Squishy with a currency that you get after taking better care of Squishy.

@jared_salois Hey Jared, there are ways to increase Squishy's happiness with the focus break (timed session to stay off your phone) and adding more ways in the future as well. The happiness also resets every day so depending on the difficulty could start with hard (50) or start easy (100).

On your other note, it is also simple to start fresh with someone else. Simply unpair on settings then run through a pairing code with someone else.

@kate_ramakaieva Thanks Kate! Android is coming out soon, but currently on Apple and Chrome :)

@omri_ben_shoham1 It mainly is a daily summary if Squishy gets too low, because I've realized when testing that I actually do make Squishy sad (and back to happy) a lot when using it with my friend (and that notification might be a bit annoying). Perhaps I can add a toggle for notifications of people who may want that real time status update.

@austin595 Great app concept with an insanely smooth onboarding flow. I would want a lifetime purchase option (tired of subscription based ones)

@pooja_phillips Most likely go down the personalization route first!

@terminal_candy Thanks Peter! I was thinking of potentiall making Squishy shared between 2-4 people, but with only 100 Happiness it can be quite difficult to manage with multiple drops at once. In regards to one Squishy but with other people, I haven't thought of that yet, I think the sync might be a bit more difficult to keep up with, but I appreciate the idea :)

@omri_ben_shoham1 That makes sense. I was a bit hesitant with how many notifications but it does make sense since this is an accountability app at the end of the day. Thanks for the suggestion I'll add it to my list!

@busmark_w_nika Thanks!

@perronef5 Really like that Basement doesn't just help you find products—it actually completes the purchase while keeping your payment details protected with single-use cards. Combining price comparison, secure checkout, and AI personalization makes for a compelling shopping experience. Good luck with the launch!

@hooni_tri You can chat with shoppers directly on any page 🤘

@galdayan As of now you need to confirm before it buys it for you. We aim to ship a more autonomous one with certain guards in place for derisking agent behavior.

@galdayan For sure. Any potentially dangerous actions default to requiring explicit approval from the user, so there's always a human in the loop before the agent does anything destructive. Once you feel confident, you can enable auto-approve.

@mia_qiao Payments is the whole point, the category spread is just tagging. The first target user is a developer selling a code-shaped product to other developers: a CLI, a template pack, a course that lives in a repo, a paid tool. Concrete workflow: fork the template, edit one config file, run the init script with a restricted Stripe key and it creates the product, price and payment link for you, turn on GitHub Pages and the store is live. Your buyer clicks buy, pays on Stripe, types their GitHub username at checkout, and a scheduled Action invites that username to your private product repo. The storefront I sell from is this exact engine unmodified, so the demo is the source.

@john_heaton06 Thanks! Usually within minutes, always within a few hours in the default polling mode, and there's an opt-in webhook mode if you want it near-instant. The email question is the nice part: emails never have to match, because delivery doesn't key on email at all. Checkout asks for the buyer's GitHub username as a custom field and the invite goes to that username, so their Stripe receipt can go anywhere. And since GitHub expires an unaccepted invite after seven days, the engine re-issues it before that happens, a buyer who missed the notification doesn't lose what they paid for.

@leo404 No manual permission editing. The free engine keeps your books in a ledger and delivery is one scheduled Action, so revoking is removing the collaborator, one call or one click, and the reconcile pass will tell you if paid access and actual access ever disagree. Pro automates the refund case end to end: a refund guard watches Stripe and revokes the buyer's repo access on its own when a refund lands. On license keys, you're right that this is where 'just Stripe' usually grows a backend, which is why I kept keys out of the delivery path entirely. Access IS the repo invite. If your product itself needs keys, Pro ships an ed25519 license module: keys signed in CI, verified offline in your app with drop-in snippets for JS and Python, still no server

@kuberwastaken Really like that Megaphone keeps everything on-device while turning natural speech into clean, polished text. The app-aware voice commands and native macOS integration make it feel like a fast, privacy-friendly productivity tool. Good luck with the launch!

@terminal_candy absolutely, especially a terminal prompt, I love it on my codex and claude code sessions

@mrtoner great question!

the one apple uses for built in dictation is called DictationTranscriber and is much dated
SpeechAnalyser is very new

Megaphone also uses the inline AI to fix tones and has application context so in an email window, it types like an email, on a slack thread, it types like a slack message !

@aryan787544 absolutely ! It's likely a mixture of both :) Megaphone is for everyone and Open Source

@galdayan that's actually a really good suggestion, maybe we can do that wish JSONs

will ship later today
Edit: there's a new import and export dictionary button in v1.1.8 !

@hi_i_am_mimo SpeechAnalyser API and Apple's on-device foundation model and 100% swift :)

@omri_ben_shoham1 Nothing automatic. The critic only grades what's written in GUIDELINES.md, so rubric-clean but wrong gets a pass.

Polish mode is the closest thing to a catch. After the plan passes, leftover cycles switch the critic to an open question: highest-impact improvement, or ship. Nothing bounds that to the rubric, so off-rubric problems surface there. Opt-in, and it can only suggest, never fail a run.

Past that it's noticing and rewriting the file. Same as adding a test after a bug slips through, except you can't write the test until you've seen the bug.

@yusuke_matsuba Bug fixes, easy, not even close. If I can write the pass condition as one line in a checklist file, the critic has something real to check and it just converges. Features work too, but only as well as I break the plan into steps.

The context thing you're describing is actually why I built this the way I did, instead of one long chat holding everything, each cycle is a clean session that just reads the current state from a file, so it never drags the old context along with it.

@galdayan No check, the worker just does what the fix line says. A bad FAIL costs a cycle same as a good one.

That's on purpose though. The whole point of the split is the builder doesn't grade its own work, and letting the worker argue with the critic brings that right back.

What keeps a bad call from spreading is the fix line only lives one cycle. The next critic is a fresh session, it reads PLAN.md, GUIDELINES.md and the actual files, never what the last critic said. So if the phantom fix broke something the rubric covers, the next critic catches it in the files.

If it broke something the rubric doesn't cover, nothing catches that, no. Same limit I told Os about, the critic can only hold the line on stuff that's written down. To me a critic that keeps failing correct code is a rubric problem, tighten that line in GUIDELINES.md once and every run after gets it. CI is the same deal, a check that fails good code gets fixed, nobody gets to skip it.

For the damage side I want per-cycle snapshots with a diff view, so you can see what a bad fix touched and toss it. Same list as the scope cap.

Only time I've actually hit this, my rubric file was the problem. Cost one cycle, run still passed. Have you run into it live or is this preemptive?

@irahimiam Thanks! No rollback, partial edits stay where the worker left them. Cancel ends the whole loop too, not just the cycle.

It's a hard kill, so a file can get caught half-written. Same as ctrl-c'ing out of any coding CLI.

Git is the checkpoint. Commit before you start a loop and you're safe.

@os_ishmael Thanks! And yeah, project managing two AI tools is exactly what pushed me to build this.

Scope is directory-level right now, not file-level. The project path gets realpath'd before a loop starts and has to sit inside the daemon's root, so symlinks can't sneak out. Then every worker and critic runs with cwd pinned to that project, workspace-write sandbox, network off inside it. One folder is the whole blast radius.

Inside that folder it's wide open though. No per-cycle file budget, no allowlist. PLAN.md keeps it narrow in practice since the worker only takes the next incomplete increment, but that's just the prompt, nothing actually enforces it.

Git is the real seatbelt. Learned that when a loop overwrote work I hadn't committed yet.

A proper per-loop scope cap is next on my list.

On the architectural stuff, it really comes down to whether you can write the goal as rubric items the critic can check. If you can, size doesn't matter much. Bug fix in a huge repo works fine. Architectural change with no tests, the critic has nothing to check against and it just thrashes. If you point it at something bigger I'd want to hear where it breaks.

@dipankar_sarkar Yeah, you're poking at the right spot.

They don't accumulate. Only the last critic's fix line gets injected into the next worker prompt. STATE.md is the thing that carries, and the worker rewrites it every cycle. The critic doesn't even read STATE.md, it only sees PLAN.md, GUIDELINES.md and the actual files.

That's on purpose, and it's why I haven't hit the flip-flop. PASS ends the loop, so finished code never gets voted on twice. With polish mode on, the polish critic can only return IMPROVE or SHIP, there's literally no FAIL in the grammar, so it can't take a PASS back. If something regressed it comes back as IMPROVE, restore whatever broke.

Cap is hard. maxCycles 1 to 10, default 3. Polish doesn't get its own budget, it just spends whatever cycles are left after the PASS.

Your point still lands though. Stateless critic means nothing stops it asking for X in cycle 1 and not-X in cycle 3. Objective GUIDELINES items are the only thing holding that line, and if the rubric is loose your two-pass rule is probably the right call.

Was yours free-form or a fixed checklist? I suspect that's the actual variable.

@thys_beesman right now, last write wins. We're working on session-based locks!

@brandon_davenport3 Thanks Brandon! Looking forward to your feedback & thoughts

@yaroslav_stelmakh Yup! Nothing leaves you device and every action is stored in "evidence" along with great analytics so you can understand where your Agent is making risky tool calls.

@galdayan That’s exactly the failure mode we designed around. A security product that constantly interrupts normal work eventually teaches people to ignore it.

Guard does not treat every unfamiliar action as a block. The default Balanced policy lets known-safe work continue, blocks clear high-confidence threats, and reserves approval for actions that are consequential, changed, or genuinely ambiguous. We also use structured command parsing rather than simple keyword matching, so quoted examples, searches, and dry runs are not treated the same as execution.

We do not yet publish a credible global false-positive percentage. The 400K figure is package downloads, not 400K instrumented runtime decisions, and Guard is local-first by design, so we do not collect everyone’s commands, prompts, files, or approvals to manufacture a dashboard metric. Internally, we've been leveraging HOL Guard for months now and have blocked hundreds of thousands of commands that could be deemed as critical or high risk.

What we do have today is a large regression corpus with paired benign/destructive cases, explicit false-positive release gates, side-effect-free decision inspection, and local receipts that make every interruption explainable. As real team deployments grow, we’ll publish measured results when the sample is meaningful.

@adriantomkins thanks, Adrian!

@thys_beesman Great questions.

Our view is that high-stakes security decisions should not depend on another model making a probabilistic judgment in real time. You can use AI to generate adversarial cases, find bypasses, and improve coverage, but eventually something has to be the final authority.

In HOL Guard, that authority is local, deterministic policy. The runtime combines structured command parsing, provenance, trust state, severity, confidence, and the user or organization’s chosen security posture. Clear threats can be blocked, trusted actions can proceed, and unclear high-impact actions can require approval instead of forcing Guard to guess.

That is how we balance false positives and false negatives. We are not tuning toward “block everything unfamiliar.” We are tuning toward making consequential actions explicit before they run, while allowing teams to choose how conservative they want Guard to be.

GPT-5.6 is not making the live allow/block decision today. We use models heavily to attack the system during development: generating edge cases, searching for evasions, and challenging assumptions. But once Guard ships, enforcement remains reproducible, inspectable, and explainable.

And because the project is open source, the rules, parsers, tests, and decisions can be challenged and improved in public rather than hidden behind an opaque model response.