PH热榜 | 2026-07-17

@philip_kubinski the line i'd underline in the pin is 'save context from any claude chat'. i keep md context files across four repos and the same fact lives in all four — nothing tells the other three when one changes. and that's the tidy layer: yesterday i pulled a price into a single constant, and the stylesheet next to it was still describing the old one in a comment. storing the file somewhere better wouldn't fix that. the file being a byproduct instead of a chore might.

@philip_kubinski does Unabyss rank memories by relevance based on the user's current task, or does it expose the full context to the model? Intelligent context prioritization seems like it could have a huge impact on both response quality and token efficiency.

@philip_kubinski Congrats on the relaunch 🚀

I really like the shift from "another memory app" to portable context that follows you across AI tools. That feels much closer to how people actually work today, jumping between Claude, Cursor, GPT, and everything else.

I'm curious, what's the most surprising workflow your early users have built with portable memory? Was there a use case that made your team realize, "We didn't originally design it for this, but it works brilliantly"?

Wishing you and the team another fantastic launch. Can't wait to see how far portable AI memory evolves! 🔥

@jakub_witowski yes, next week. I'll let you know as soon as it's ready! But you can already onboard your team and we'll merge your accounts into one org in a few days.

@jakub_witowski, we are multitasking - as we speak, the team plans are being developed :D.

@abdullah_javaid3 appreciate your feedback!

1. Wrong facts -> we have conflict resolution in place, so incorrect information won't be retrieved from memory. Facts are cross-checked against other, more recent memories before they're retrieved.

2. Source tagging is bulletproof. Permissions for sensitive/confidential data are handled by the agent, so I can imagine edge cases where things don't work exactly as intended. Agency use-case is very specific and we're launching the agency context architecture soon. Memory silos will be fully isolated, making this 100% secure. Until then, I'd recommend using source-level (connection-level) permissions.

Happy to update you when 2) is live!

@abdullah_javaid3 About the wrong facts and conflicts, we have the conflict resolution engine that works in the background. But if it misses something, you can always adjust it in any agent the Unabyss is connected to or in our in-app context chat. Fixed in one place - fixes it everywhere.

Our permission layer is overprotective. So it is more likely you will get less confidential data than it will return something that it should not.

If you could find time to share your feedback from using Unabyss, it would be great - we ship every day, so if something does not work perfectly, it should be fixed in a few days ;).

@artstavenka1 thanks! We have a complex conflict resolution engine. When some memories conflict, we don't rely only on the date but also on the source, repetition, who the author is, etc. If automatic resolution raises concerns, we are adding a note explaining why this data changed and when.

@kyzo :D why not - this sounds like a valid use case

@kyzo yup, and if you share your MCP with your match, you can skip dating. You'll already know everything about each other!

@greenparrotnow we're shipping local memory next week. I'll share closed beta with you via DM!

@greenparrotnow already in progress ;) local version is launching soon

@conduit_design first of all, we want to work in the background. Once you set up, there is no need to change how you interact with any agent. Unabyss does its thing, and you don't even know about it - you just see better results from agents and fewer questions asked.

About the name (thanks!), it comes from a combination of 2 words: "un" + "abyss" - as you can see in the screen below, when you hover over the logo in the footer, you will see the full etymology.

@philip_sorensen Thanks! About your question: every time a new item is fetched from the connected app, it is analyzed against the current context. Unabyss checks for conflicts, dates, and identities - if new data is better used to update an existing entity than to create a new one, the update will happen. This ensures that the context is always up to date.

@aleksandra_dabrowska02 yes! you can connect it to our Notion board with the client list, Gmail, and Calendar to get the full context, and then use it in Claude to get whatever you need!

@therealkaczor thanks! We think we know how to crack it ;)

@yaar58863951762 

Thanks! That's one of the changes we've introduced after feedback from our users - glad to see it improves your experience :)

@yaar58863951762 thanks! feel free to share your feedback anytime ;)

@henry_habib extremely easy :)

if something gets saved by mistake, you can ask Unabyss to edit/remove it wherever you use it (Claude, GPT, Cursor, or in Unabyss itself :))

and you want to know what Unabyss knows about specific topic - you can talk it through. And (on Monday) we'll release Context View, where you'll see every information that Unabyss has!

@henry_habib super easy, there are 2 main ways in which you can do it: talk to our context agent in the app or do the same in your Claude, ChatGPT, or Hermes. Once you adjust it, it will stay fixed in your context.

@priya_kushwaha1 thanks!

We worked very hard to improve sync times :)

@amanda_baldwin1 

Everything happens via our "store" tool in memory. You can save every conversation to Unabyss - just call it manually (e.g. "save it to Unabyss") or adjust your e.g. Claude preferences so that every chat is saved to your memory.

Some users choose the 1st option, some the latter - depending on how much information they want to transfer to Unabyss.

And yes, it works for OpenClaw too! In general, it works with every AI tool that has MCP :)

@amanda_baldwin1 when you are onboarding a new agent to Unabyss, there is a short setup/onboring flow that allows you to pick what you want to save. Based on that, you get global instructions that you can paste in your preferences - this way, context always flows between agents effortlessly.

@shivangit26 totally agree! Unabyss executes conflict resolution just before retrieval, so any potentially incorrect fact is checked against other memories, with recency taken into account. Outdated or incorrect information are simply filtered out before being returned.

And btw this is a core part of our memory architecture, and we're improving it every day. Literally!

@karolina_uchacz of course, you can give an access to your context to your brother for example! :))

@florian_hofmann2 thanks a lot!

@tarqiya_forgah Thanks! We have a lot of power users, but interest from teams is strong too. We will be launching team plans next week to accommodate the interest ;)

@mystoryland Hi Olga, in Unabyss, you can see what the source origin is. In terms of conflicts, we have a resolution engine that checks incoming memories against existing data and evaluates them based on date, origin, author, and other factors. Most of the time, it handles it well, but if it gets unsure, then it will leave a note next to the change, explaining why and when it happened.

@albattran thanks a lot!

@asiyel42174 we are glad you like it!

@huisong_li Thanks a lot!

@michael_yagudaev thanks!

@therealced thanks! With unabyss this should never happen again ;)

@celalgmubusiuw awesome!

@aima Thanks, so the initial setup takes up to one and a half minutes. Then, depending on how many items are pulled from connected apps, the first sync will take minutes; larger syncs take 10-30 minutes. Either way, it's a one-time setup, and after a few minutes you can start getting value from Unabyss, and that value only grows the longer you use it.

@matthieu_poitrimolt thanks!

@ashishkingdom Thanks! Every time you connect a new agent to Unabyss, you can select what this agent will get. You can give it full context access or filter out confidential information, personal or business, but you can also filter out context that was fetched from the chosen app.

@natella_nuralieva thanks! we aim to polish not only the context but the designs too ^^

@dima_antoniuk I like that Pebbles AI brings the entire GTM workflow into one place instead of forcing teams to jump between multiple tools. Combining strategy, lead generation, outreach, sales, and shared company knowledge in a single AI-powered workspace could make execution much more consistent. The use of neurosymbolic AI trained on a company's own business is an interesting differentiator, especially for creating more relevant campaigns and personalized outreach. Best of luck with the launch!

@dima_antoniuk On the buyer side, most "AI sales platform" demos fall apart the moment they hit a messy CRM and a sales cycle that doesn't match the template. How much manual mapping does it take before the AI is actually useful, and does it augment the reps' judgment or try to own the first touch?

@dima_antoniuk The "augment, never replace" framing is one I want to believe, but I've seen it used as a soft landing for tools that quietly just replace people. The stat that stands out to me is one marketer delivering the output of three, genuinely impressive if true, but it raises a real question about what happens to the other two.

I'm not being cynical. I think there's a version of this that's genuinely positive for the people using it. But I'd love to hear from anyone on the team or in the community: does the tool expand what you're able to do, or does it mostly compress headcount? And how does the team think about that tension internally?

For investors (VCs/Angels) we are incidentally also raising our first institutional round (Seed). Get in touch!

@oleg_tsizdyn Let's ask Pebbles itself. I tried both prompts, just a single feature or tailored by domain

@oleg_tsizdyn On your question, the neurosymbolic cores (brains) underneath. Not the flashiest thing to demo, admittedly, but it is the reason why everything works with insane efficacy.

Most tools bolt an LLM on top and hope. That is fine until the model invents a confident wrong strategy and you only notice three campaigns later. Ours reasons, then checks its own work before it reaches you.

The neural side handles language, the symbolic side handles logic and verification. Think of it as an assistant with a fact checker built in, rather than one that just talks fast and means well.

Everything you actually touch, the Strategy Assistant, Smartbox, Fresh Leads, sits on that foundation. So the strategy it hands you is one it can defend, not one it dreamed up between tabs.

Give it a spin and push it hard. And when you are in, tell me which part of your stack you would happily retire first.

@oleksandr_knyga You were here before there was a "here". No team, no safety net, just a hard idea and one person willing to build an architecture that could actually carry neurosymbolic logic.

We only had 10k in the bank. And an ambitious vision.

Most people would have called that impossible and gone back to a normal job with a normal salary. You stayed, and you built the foundation the whole thing now stands on.

Then you backed it with your own money, more than once, when nothing obliged you to. Skin in the game, long before the results were indisputable with our early customers.

Thank you for all of it. This launch is yours as much as anyone's.

@priyankamandal Thank you Priyanka. Our journey has been indeed tough.

We built a lot of this through blackouts and air raid sirens in Lviv (Ukraine), and the team shipped anyway. I mention it because it shaped how we build.

When your office has an actual siren, you stop shipping things that only work in the demo, or don't actually solve anything.

Most GTM tools bolt an LLM on top and hope for the best, which is fine right up until the model invents a confident wrong strategy and you only notice three campaigns later.

Ours reasons first and checks its own work. Neurosymbolic under the hood, boring to demo, but the whole reason we can run an entire OS.

So please do not be gentle with it. Break it, argue with it, tell me where it falls short. Blunt feedback in the comments is worth more to us than a polite nod, and I will answer every one myself today.

🙏🏻

@dima_antoniuk  @priyankamandal  @istiakahmad :

🙏 This means a lot, thank you for hunting us.

You've articulated the problem better than we usually do. GTM becoming a patchwork where managing the tools costs more than doing the work, that's the exact frustration we built Pebbles to remove.

And you nailed the why behind the neurosymbolic bit. It reasons from your company's knowledge instead of spraying generic output. That's the whole difference.

To throw your question back to the room, because it's a good one:

What's the single biggest bottleneck in your own GTM right now?

We built the demand through our own platform, but had no resources to keep up with it. Too much inbound.

We ran a great campaign on our own platform and the positive replies flooded in. We hired and let go 5 people in a year attempting to keep up. Alas, to no avail.

The SDR function is broken, and even we couldn't fix it. The real gap: business netiquette, critical thinking, creative problem solving, consistency, and execution speed. Though it is not their fault. Hardly anyone trains SDRs. The industry turned ruthless. Universities do not prepare them. They are set up to fail.

So we found another way to save our own sales pipeline, and that of our customers.

What actually went wrong:

❌ One SDR quietly did not touch the inboxes for 2.5 months, around five LinkedIn and 10 outbound email inboxes. Campaigns and inside sales looked great, but nothing trickled down. We only found out when we went looking for why

❌ The SDR is the first human contact in the company. We watched roughly 20 interested, ready-to-pay prospects get so put off by the replies that they walked. Essentially a first-impression problem.

Here's the before and after building

🐖 Total costs
• Before: ~$120k+/yr | 3-4 SDRs / Jr. Sales Managers plus a manager
• With Auto SDR: $0 + $350/mo Team subscription (~$4,200/yr, zero SDRs hired)

💸 Pipeline value
• Before: -$32k/yr lost | ~90% of quota missed, warm replies rotted
• With Auto SDR: Confidential, ~7x more gained | demand caught, founders on the calls

🧠 Sales Efficacy
• Before: ~10 SQLs/mo | ~30 MQLs at ~32% MQL→SQL
• With Auto SDR: ~22 SQLs/mo | same demand at ~70%

⚡ Reply time
• Before: Hours, or never | the 2.5-month blackout
• With Auto SDR: 3-7 min | 24/7, every inbox

📈 Net margin
• Before: Confidential | baseline, if we had staffed it
• With Auto SDR: Confidential, ~1.9x | running Auto SDR

@julia_shtogren Great question. And a designer's eye would land on exactly this, because bad personalisation is obviously cringe.

The short version: our personalisation goes both wide and deep.

Wide means we look at every angle of your prospect before writing a word:

• The marketing angle: what message actually resonates with them?

• The sales angle: where's the commercial fit, the pain we can solve?

• The human angle: who is this person, really, beyond the job title?

Deep means we apply proper sales methodology, mapping the impact of your solution at three levels:

• The organisation: what does this mean for their company?

• The team: what does it change for their unit's goals?

• The individual: what's in it for this specific person?

But here's the important bit: none of that sits on its own.

We fuse it with true hyper-personalisation. Our systems work out what's actually top of mind for your prospect right now, what they've been posting about, what they care about, who they are.

Then we use science-based methods to find genuine common ground, the shared interest or the icebreaker that actually starts a real conversation.

I hope that make sense.

@busmark_w_nika Thanks, we are happy to have our hunter on board!

@andrasczeizelGREAT questions!!! You've named the exact thing that caused me anxiety as a founder. You wake up one day and realise you're paying for 10+ different tools related to GTM.

That's insane even for an established small business of 100 people, let alone a startup finding its feet with 2 co-founders.

Each tool is another subscription, another login, another line item, another learning curve, and that creeping OpEx is the silent killer of your runway.

Let me break it down for you:

🧱 At the base sits your organisational intelligence, the source of truth about the company, the approach, and the products/services

📚 Above it, the GTM knowledge base: best practices, business netiquette, and the neurosymbolic logic for what to do in each case, and IFTTT logic for complex requests

🔬 On top, the sciences: communication science, applied persuasion sciences, hyper-personalisation methods, brand voice rules, persona-centric writing, and cultural nuances

On how much comes from your data versus fixed playbooks, think of it as an 80/20 split:

• The 80% is us: the GTM sciences, B2B heuristics, neurosymbolic workflows we've distilled from how the top 1% of management, marketing and sales actually operate, judge, and executes. We never leave it to the base AI models, we use our battle-tested reasoning system that adapts to each case

• The 20% is you: your organisational data, your brand voice, and your company history to date; that's the only data we need, onboarding takes only 3 minutes

On how clearly you can inspect the why why a lead, segment, or campaign , which is the part I care about most:

• Every output has an audit trail. Because the reasoning runs over explicit workflows, sciences and market intelligence, you can follow the whole chain, even as a spider spider web of neurons, and see exactly how we arrived at a strategic recommendation, an ICP analysis, or a campaign

• That means you can verify the system isn't making random calls: the logic is traceable end to end, so a lead, a segment or a campaign is always backed by a reason you can inspect. We built this specifically for enterprise as they deem this VERY important, but provided access to companies of all sizes.

• For research it draws on roughly 10x more sources than a typical base model (Strategy Assistant draws at least 60 sources per inquiry), and we've categorised every source by tier (Tier 0 to Tier 4): data providers like Statista, market intelligence companies like Gartner, down through Reddit threads. Each Tier is used only when it's actually appropriate

• In fact, the Marketing Assistant takes a bottom-up approach: it runs semantic analysis across sources like Reddit and X to surface what the public actually thinks and feels about a given topic (for example, sentiment analysis on a product category or a competitor)

• The Strategy Assistant works the other way, top-down: it runs market intelligence analysis over open data sources like the World Bank Open Data and Eurostat to forecast how markets are likely to move (e.g. a DIKW approach, turning raw data into information, knowledge and finally insight) so you can plan against where the market is heading, not just where it is today

• Put together, it's like having a senior analyst from McKinsey, a senior copywriter from Saatchi & Saatchi, and a senior enterprise closer from Big Tech, all working in one place. No more stitching together expensive consultancies and senior hires you can barely afford, and often can't justify before you've even found product-market fit. You get that calibre of thinking from day one, at a fraction of the cost of a single one of those salaries

And so you know this isn't a weekend project?

This wasn't built over a weekend. It was built on 10+ years of first-hand GTM experience, 18 months of PhD-grade research, and 3 years of development, roughly 500 weekends, but who's counting. ;)

@alex_tomilinThank you. This is my favourite question by a mile. Most people admire the Ferrari (the GTMOS). Don't really care about the engineering that went into the engine (neurosymbolic AI). Let me break it down in 3 levels.

Macro Level | What is it?

A base LLM is inherently a chatbot system. It predicts the most probable next word, brilliantly, but that is not enough for complex domains such as B2B GTM, Corporate Law, and Human Resources. Nothing in it stops to ask "is this true, and does it obey the domain rules".

Neurosymbolic AI is 3-step systems working together, with a check between them:

• The neural half reads language and context, the way any strong LLM does

• The symbolic half applies explicit logic, rules and a structured knowledge base

• A verification step sits between that and the output, catching anything that breaks

So you get the fluency of an LLM with a reasoning and fact checking layer bolted underneath. Outputs are accurate, reproducible and explainable (even auditable) rather than a confident black box.

It is also rare: academic interest went from 112 papers in 2015 and 2016 to over 9,000 in 2025 and 2026 [Google Scholar], yet real production systems are almost nowhere, because building one needs machine learning, formal logic, knowledge engineering and domain science in the same room at once.

Meso Level | How did we build it?

The engine is not a weekend project. The numbers behind it:

• 18+ months of research before the first line of production code

• 70,000+ engineering hours in the proprietary neurosymbolic architecture

• 800tn parameter permutations in business communication alone

• Over 3,000 rule-based IFTTT rules across all B2B use cases and workflows

• 7 specialised neurosymbolic cores under GTMOS™, each mimicking the top 1% domain experts

Every output reasons up through a layered stack. Seven layers are proprietary Pebbles IP, two are heuristics tuned to you: your organisational truth at the base, then the GTM knowledge base, neurosymbolic logic, applied persuasion sciences and brand voice, with persona and cultural nuance on top.

That knowledge layer encodes six GTM disciplines, persuasion science, neuromarketing, behavioural economics, competitive strategy, segmentation theory and sales methodology, codified from closely-guarded secrets.

Break a rule or assert something not in the data, and it gets caught and corrected before it is ever sent. It also doesn't agree with you. It cares more about your success, than your ego. It will not allow you to make mistakes.

Micro level | Why it matters?

Let's look at some numbers. First lets look Claude Opus on the Max tier with a single instruction versus the full Pebbles pipeline:

• Accuracy: 33% vs 87%

• Precision: 57% vs 91%

• Sales Efficacy, MQL to SQL: 15% vs 85%

And at the architecture level:

• 82% lower error rate than LoRA fine tuning, because the architecture is structurally accurate rather than nudged

• 3x better gross margin than wrappers, because the reasoning is not rederived from scratch on every call

• ~2% hallucination on rule bound queries, versus 31.4% across real world use HalluScore benchmark

But don't take my word for it: Claude Opus sits around 33% factual hallucination on the public HalluScore benchmark, while neurosymbolic methods approach ~100% accuracy on rule based tasks [arXiv 2502.01657]. We are closing up at 98%.

Cost is where it gets almost silly. To rebuild one reply with a raw model:

• Around 12 prompts per reply, each re sending 25,000 to 35,000 tokens of context

• Roughly $5 to $7 per usable output, and that's not even top 1% reasoning.

• About $5,000 to $7,000 a month at a outputs, before 500 hours of human prompting

Pebbles does the same job inside a subscription near $450 a month, all in. Same output, a fraction of the cost, none of the babysitting.

The neursoymbolic reasoning is what lets it carry complex, multi-faceted, and cross-functional B2B work all the way through. These are examples you can build and execute on, which is impossible with base-models or tools with wrappers:

• A full go to market strategy, grounded in your ICP, positioning and live market signals, then turned into the campaigns that run it

• A beachhead strategy, picking the wedge segment worth attacking first, sizing it, and sequencing the entry instead of guessing

• Industry trend analysis read across macro, meso and micro signals, so you see the shift before it hits your pipeline

• Investor decks and enterprise sales assets, two pagers, RFPs and proposals that hold up when a sharp reader pushes on them

• An omnichannel outbound engine, from fresh leads to reasoned email and LinkedIn sequences to replies captured and qualified in Smartbox, built and run end to end

And this is only the current stage. We are makingthe first steps toward a true Jarvis for go to market: a system that can safely, securely and reliably run the work fully autonomously, with no human in the loop.

Try it yourself, break it, and see where it holds.

@raihanshezan Brand voice here isn't a static style guide you set once and hope it sticks. It's actively built through a neurosymbolic workflow inside Pebbles that structures how your voice gets defined – not assumed or inferred passively. Once constructed, it's saved to the centralised Library alongside your other company context, making it a living asset you can reuse and refine over time, not a passive setting buried in a panel.

From there, the Library acts as persistent memory. It holds your outputs and team knowledge so the AI draws on your real content, not a blank slate. To be precise about the mechanics: this is context-driven memory retention, not model fine-tuning in the ML sense. No weights are updated. The system gets sharper as usage grows because it's working from richer, more specific context.

On follow-ups, the logic is adaptive – not a fixed drip sequence. Assistants use stored context and reply signals to decide what comes next. It's not being retrained on the fly, but it's also not running a rigid script.

Your core question is the right one to ask: does this reduce downstream editing, or just move it somewhere else? Persistent context – including the brand voice you've explicitly constructed – is exactly how we're trying to solve that. One more thing worth knowing: tenant isolation is in place, so your data never crosses over with other clients. What you build stays yours.

Happy to go deeper on any part of this.

@raihanshezan You clearly know this space at a veteran level, the kind of read that only comes from running real outbound and watching exactly where it breaks. Your sharp questions. One at a time. Here we go:

Trained on your own content, or a style guide you configure upfront (Brand Voice Creation Feature)?

Your own, and the way you build it is the fun part. It runs as a guided Q&A, about 2 hours, closer to a sharp interview than a setup form.

It pulls your voice out of your answers and your best existing writing, then hands you a full spec: a word arsenal you actually use, a forbidden list you never touch, your signature phrases, and the mechanical fingerprints like sentence rhythm and punctuation.

What comes out is a brand voice that is distinctly yours and, more to the point, one that actually performs.

Our Brand Voice Creation Feature was built on principles drawn from McKinsey strategy practice and Saatchi and Saatchi creative heuristics, then layered with persuasion science, communication sciences, and a full library of anti patterns.

So it does many things at once. It captures how you sound, sets you apart, and makes the voice ACTUALLY effective, not just a gimmick.

This is the first half of the puzzle.

How does it learn and hold the voice over time?

Two parts, depth and enforcement.

The depth is a layered stack sitting under every message in every feature (from Marketing Assistant, Auto SDR to Smartbox). It uses 7 layers as enforcement. All proprietary Pebbles IP, only 2 are general heuristics tuned to you.

From your foundation up to the surface:

• Organisational intelligence, your company's source of truth

• GTM knowledge base, proprietary best practices and business netiquette

• Neurosymbolic logic, what to do in every case, even the edge cases

• Applied persuasion sciences

• Brand voice framework

• Hyper-personalisation

• Persona-centric writing

• Cultural nuances

Every draft runs a final check against your spec before it leaves, so the voice stays put instead of drifting the way a fine tuned model does. This is around (a) precision, (b) accuracy, and (c) efficacy.

Think of the baby of a senior Saatchi and Saatchi copywriter and a marketing scientist. It has your style guide memorised, knows every persuasion principle, and never has an off day. As you approve and edit, the spec sharpens toward your style too (the cherry on top).

This enforcement is the other half of the puzzle.

Do follow-ups adapt to reply signals, or is it a fixed sequence?

They adapt. The logic reads the intent behind each reply, then acts on it:

• An objection gets answered on its merits

• A "not now" gets a gentle nurture

• A no gets turned into a maybe

• Not me gets the colleague in

• Silence gets the auto follow up

The sequence bends to the symbolic signal instead of marching on like chatbot.

There is more IFTTT neurosymbolic logic built in, but I'll spare you the novel 😂

@ethan_lee8 Hi, thanks! We offer free trial and then you can pick subscription that matches you most. Professional and team tiers are available right now with option to buy bundles for AI and leads.

@ethan_lee8 Free first, because a paywall on the first hello is a poor way to make friends. 😂

You can run the full platform for nothing, no card, cancel whenever. For the Product Hunt launch the free entry comes with a glass of champagne, some snacks, and a shout out.

• The champagne: 3,000 fresh leads to put it through its paces

• The A generous AI compute allowance (approx. £250 GBP)

• And you can invite 2 more team members for free (3 seats in total)

Professional, £49 a month
For solo operators doing the work of an entire team.

• 1 seat, 400 fresh leads a month, 3 neurosymbolic cores

• General Assistant your GTM pilot, Smartbox, your adaptive Personal Library, Google Workspace, CASA Level II security

• Retires the solo stack: a data tool, a sequencer, Sales Navigator and a couple of AI subscriptions. That pile runs north of £200 a month. This is £49. One login, roughly a quarter of the cost.

Team, £349 a month (down from £499)
For startups and small businesses growing smarter.

• 3 seats, 2,000 fresh leads a month, 6 cores

• Adds Strategy Assistant, email and LinkedIn outreach, the Brand Voice core, cross-feature memory, the collaboration hub, Company Library, Microsoft 365

• Retires the above across three people, plus a team CRM. A comparable three person stack lands around £900 to £1,500 a month. This is £349, so most of that spend goes back in your pocket.

Organisation, £1,499 a month
For mid-size businesses uniting every team on one plan.

• 10 seats, 6,000 fresh leads a month, 7 cores

• Adds team productivity analytics, centralised memory, a dedicated CSM, expert onboarding, and the SCALE 4 week GTM programme

• Retires the full mid-market stack: data, outreach, CRM, analytics and enablement. That typically runs £3,000 to £6,000 a month. This is £1,499.

Enterprise, £3,499 a month
For multinationals aligning GTM across regions and teams.

• 20 seats, 15,000 fresh leads a month, 8 cores

• Adds custom core and AI model tuning, SSO and SAML, persistent memory, and a dedicated account manager

• Retires an enterprise GTM stack that, with autonomous SDR tools alone, runs £8,000 to £15,000 a month. This is £3,499.

💛 Priced like a tool, performs like a department 💛

@varun1jan Exactly, that context switching tax is a huge hidden cost for GTM teams.

@varun1jan Exactly, Varun. The hidden tax in most stacks is not the software, it is the human hours lost carrying context between tools that refuse to talk to each other. Even though, that is also very costly for a big company.

A better way is one operating system, one shared context, a neurosymbolic brain underneath. Your strategy, leads, and outreach all read from the same brain, so the rep spends the time on the prospect instead of playing courier between 10 tabs.

Assistants think. LeadGen hunts. Auto SDR warms. SmartBox closes. Every other feature on the OS amplifies.

@jasnoor_singh_oberoi There's good news. The £1,499 isn't the entry point. That's the Organisation tier, built for a 10-seat commercial team, often for lower-mid to mid-market organisations.

The way in is Pro, at £49 a month. Even better, the Team subscription at £349 a month has the most value. The most bang for your buck.

We wanted to make the value stupidly high. Like a NO-BRAINER. Something that would absolutely flabbergast new users. How is it possible to add so much value at such a low price.

(Spoiler alert: it's possible because our system is built on neurosymbolic AI)

So the exact person you're describing, the solo founder buried in fragmented tools with no sales team, isn't an afterthought for us. They're arguably who we built this for first. Don't forget, we were also in that position.

In our experience, solving for smaller teams has proven to be much more difficult than for larger teams, ironically. So we did it the hard way first. Naturally haha.

On the positioning itself, you're right. When we say "enterprise-grade" or "McKinsey-level," we mean the quality of the reasoning and judgment, not the size of the customer or the price of the plan. We should be more clear.

Our mission is David versus Goliath. Give a two-person startup the same GTM firepower a big company pays top-tier consultants and 30 premium tools for, at a price a founder can actually afford.

In short, you only scale up when your team does.

Best way to judge it is to run it yourself against the mechanic and overwhelming tool fragmentation problem you described.

Start free, no credit card needed. I'd recommend trialing the Team subscription.

You won't regret it. 😊

@tarqiya_forgah Great question! Let me break it down for you.

ChatGPT: replaced completely.

For GTM specifically, this is the cleanest swap. Generic AI gives you fluent text with zero memory of your company, your ICP, or your positioning, so you spend more time briefing it than it spends adding tangible value back to you.

Pebbles' Marketing Assistant and Strategy Assistant cover it, and actually make an impact on your pipeline, revenue, and margins. Not to mention it gives you energy back, because you don't need to learn so many tools or keep so many tabs open that never inform each other.

Clay and Apollo: largely absorbed.

This is where the patchwork usually collapses into one flow. Clay's real value is its enrichment waterfall and AI research, and Apollo's is its contact database plus sequencing.

Both run natively inside Pebbles, on our own multi-provider enrichment waterfall, feeding straight into reasoned email and LinkedIn sequences.

So the sourcing, the enrichment, and the outreach stop being 3 tools you stitch together by hand. It's all connected now.

On top of that, we use specialised profiling that reads each prospect from both a marketing angle and a sales angle. Pebbles doesn't just hand you leads. It tells you who they actually are, and recommends how to build a relationship with them.

HubSpot: keep it.

Your CRM is your system of record, and it should stay that way. It's wired into your reporting, your deal tracking, and your support.

We are not asking you to rip out the ledger. Pebbles sits on top as the reasoning and execution layer, and the CRM stays the source of truth. We'll be integrating with HubSpot, Notion, Salesforce, and Pipedrive soon.

Cost

There's a cost angle too. Those four as a combined stack, across seats, is not cheap. We priced Pebbles to come in under Clay alone at every tier, and that's before you count the three or four other subscriptions it folds in.

So the short version: keep the CRM as your record, retire the patchwork on top of it. We'll have integrations soon with Notion, HubSpot, Salesforce, and Pipedrive.

@nhanife56159 This made my day, thank you! The tab-switching was the exact thing that drove me mad as a founder, so hearing that the strategy-to-outreach flow cut it down for you is the best feedback we could ask for.

That "one workspace" feeling is the whole reason we built it, so it means a lot that it is received well. If you hit anything confusing or have a wishlist as you go deeper, I'm all ears. We're shaping it around real users and we ship fast.

@nikitaeverywhere The main interface today is the web app. We also have desktop apps that live in your dock on iOS (mac) and Windows, so it sits right alongside your other daily tools rather than buried in a tab.

No installation required. Super light.

On Claude specifically: we don't integrate with Claude, and that's on purpose. Pebbles runs its own reasoning models, our cores, which are purpose-built for GTM rather than general-purpose.

MCPing Claude would mean inheriting a general model's guesswork for rule-bound, domain-specific complex work, which is exactly what we set out to avoid.

Instead, we make switching effortless. Think of it like changing mobile providers and keeping your number. The system does the heavy lifting, and you're moved over in about 3 minutes without losing your most up-to-date company info.

You don't rebuild your progress that you've carefully built within Claude (or ChatGPT, Gemini, etc.). You just move it over to Pebbles Ai within 3 minutes, and run it on much smarter models.

Lastly, every customer gets a custom API that allows to connect the Pebbles Ai neurosymbolic brain trained exclusively on your company to your existing techstack.

Think of it as the primary brain of your existing techstack. You can wire it into your website to answer inbound inquiries, power the chatbot on your site, and plug into the other automation tools you already run.

In short, you get both a sophisticated multi-purpose platform + a connectable brain that makes your other tools smarter too.

Hope that make sense.

@omri_ben_shoham1 You're absolutely right. Most of these tools overlap, and the category is crowded with point solutions that do one or two things very well.

But none of them focus on reasoning. They only focus on execution.

People think outbound is a sending problem. It's usually a reasoning problem, and it breaks into two formulas most tools never touch.

The first is the sequencing itself. A message that actually lands isn't a template with a {first_name} slot. It's at least 5 disciplines working together:

• Persuasion science: the established principles of influence and buyer psychology (reciprocity, social proof, commitment) and knowing which to use at which moment

• Communication science: how the message is structured and paced to be understood. Clarity, framing, timing, and the netiquette that makes it read like a human wrote it

• Hyper-personalisation: real profiling of who this person is and what they care about, so you connect on a human level despite being total strangers. Not a dumb variable field like job title

• Heuristics: the judgment rules for each situation. When to push, when to nurture, when to stay quiet, distilled from the combined wisdom of top operators

• Lexical semantics: precise word choice and sentence construction, because the same point lands or dies on phrasing and tone

Miss one and the sequence underperforms. Which is why so much outbound gets 1% reply rates and blames the list.

But here's the part almost nobody does. All of that only works if the upstream formula fired first. Before a single message goes out, you need:

1. Geolocation (e.g. Greater London)

2. Industry (e.g. financial services)

3. Persona cluster (ICP and archetypes)

4. Common denominators (historical data)

5. Market trends (the ones that actually move you)

You can do all of it on the platform. Great sequencing on bad upstream work is just a beautifully written message, sent to the wrong person, with the wrong value proposition.

That's the whole reason we built Pebbles as one neurosymbolic engine, not another point tool. It reasons over both dimensions: strategy upstream, execution downstream.

And this is the real problem with the point-solution pile. Those tools were built for a pre-AI world. They aren't AI-native, nor domain-specific, so they physically can't reason at this level.

There's a deeper issue too. Dig into the research and you'll find most GTM point solutions are built exclusively by engineers with no background in strategy, marketing, or sales.

So ask yourself: how do you solve something you've never lived? It's like our tream trying to fix a problem in corporate law. We don't know how paralegals work with associates, how associates work with solicitors, how solicitors work with partners, how partners work with the board, and how each of them handles clients through completely separate workflows. Hand on heart, personally I wouldn't know where to start.

Don't get me wrong, these engineers are brilliant. Seriously smart, and they've built genuinely great products. But each one covers one or two stages of a funnel that has at least ten.

Because go-to-market is no different from building software. Teamwork is everything. Back-end works with front-end, front-end with the product designer, everyone with QA and DevOps, and a solution architect sits across all of it. Engineering has Jira for exactly this.

Nobody had built the Jira for commercial teams. We're the first. It's called the Go-To-Market Operating System, and it could only exist with extraordinary intelligence beneath it.

Think of it like iOS. Each app (feature) has its own tech stack, its own intelligence, its own knowledge base, its own mapped-out GTM science. It's like having multiple startups in one place. And every app talks to the others.

Point solutions can only bolt on an "AI feature," which is usually just a free ChatGPT or Gemini model in a sexy trench coat (UX). Execution works, so you gain speed.

But it doesn't move the needle on pipeline, revenue, or margin. You and your team just do busywork faster. Three months later you look back, and nothing material has changed on the financials.

The meme says it perfectly. You think the fix is one more point solution. It isn't. It's more busywork, plus a fresh learning curve for yet another tool that most people never fully learn.

That's exactly the problem Pebbles solves, with a completely new approach.

Great question Patrick!!! Love when technologists go deep.

For small B2B teams, full autonomy is mostly a pitch-deck fantasy today, and the reason is arithmetic, not opinion.

Agent errors compound multiplicatively across steps. At 95% reliability per step, a 10-step task lands around 60% overall (0.95^10). Even at an optimistic 99% per step, a 20-step chain only succeeds ~82% of the time.

Empirically it's worse: in Carnegie Mellon's TheAgentCompany benchmark, even the strongest model completed only ~30% of realistic, multi-step office tasks end-to-end (earlier runs landed at 24%).

And the market is pricing it in. Gartner expects over 40% of agentic AI projects to be canceled by the end of 2027, citing unclear value and inadequate risk controls, and notes today's models can't reliably follow nuanced instructions over long horizons.

People are frustrated, and frankly, when I talk to customers, angry!

So the winning pattern isn't more autonomy. It's selective autonomy: aggressive automation on low-stakes steps, hard human gates on high-stakes ones, with escalation paths that cap the blast radius. Which is precisely the handoff quality you're describing.

You framed these four around the sales agent, but they're really the design spec for the whole platform, strategy, marketing and sales alike. So here's how each one works at that level:

What changed?
Whatever the surface, a market shift the strategy core detects, a competitor move, a new buying signal, an inbound reply, the platform isolates the delta and leads with it. You get the one thing that moved and why it matters to YOU, not a raw feed you have to reverse-engineer. Same behaviour whether it's your positioning or your pipeline.

What evidence supports it?
Every recommendation carries its reasoning, a strategic call, a beachhead segment, a line of outbound copy. This is where neurosymbolic earns its place over a raw LLM: the symbolic layer keeps each inference traceable back to the rule or data point that produced it, so a conclusion arrives with its "why" attached rather than as an unexplained assertion from a black box. Inspectable reasoning like that is what the research points to as the precondition for trustworthy human-in-the-loop oversight.

What needs human judgment?
Whether it's a strategy decision, a targeting call, or an outbound message, when confidence is low (we have "judges") or the call is genuinely ambiguous the platform surfaces that explicitly and routes it to you, instead of papering over uncertainty with a confident-sounding guess. Overconfidence in failure is one of the documented ways agents break, so we treat "I'm not sure, your call" as a feature, not a bug.

What should never be sent without approval?
Anything irreversible or externally facing, a published campaign, an outbound send, an investor or sales asset going out the door, is gated by default. The platform can research, reason, write, sequence and prepare the whole thing, but the irreversible action stays with a human until you explicitly choose to loosen it.

Notice the common thread: every one of those four depends on reasoning you can actually trust, not a confident guess. That trust is the exact thing the arithmetic up top says nobody has yet, it's why 40% of these projects get canceled. So the real question is whether our engine is any different. Here are the numbers.

How does our neurosymbolic actually perform?

A base LLM is inherently a chatbot system. It predicts the most probable next word, brilliantly, but that is not enough for complex domains such as B2B GTM, Corporate Law, and Human Resources. Nothing in it stops to ask "is this true, and does it obey the domain rules".

Neurosymbolic AI is 3-step systems working together, with a check between them:

• The neural half reads language and context, the way any strong LLM does

• The symbolic half applies explicit logic, rules and a structured knowledge base

• A verification step sits between that and the output, catching anything that breaks

So you get the fluency of an LLM with a reasoning and fact checking layer bolted underneath. Outputs are accurate, reproducible and explainable (even auditable) rather than a confident black box.

It is also rare: academic interest went from 112 papers in 2015 and 2016 to over 9,000 in 2025 and 2026 [Google Scholar], yet real production systems are almost nowhere, because building one needs machine learning, formal logic, knowledge engineering and domain science in the same room at once.

Let's look at some numbers. First, let's compare Claude Opus on the Max tier with a single instruction against the full Pebbles pipeline:

• Accuracy: 33% vs 87%

• Precision: 57% vs 91%

• Sales Efficacy, MQL to SQL: 15% vs 85%

And at the architecture level:

• 82% lower error rate than LoRA fine tuning, because the architecture is structurally accurate rather than nudged

• 3x better gross margin than wrappers, because the reasoning is not rederived from scratch on every call

• ~2% hallucination on rule bound queries, versus 31.4% across real world use HalluScore benchmark

But don't take my word for it: Claude Opus sits around 33% factual hallucination on the public HalluScore benchmark, while generally neurosymbolic methods approach ~100% accuracy on rule based tasks [arXiv 2502.01657]. We are closing up at 85-98%.

Cost is where it gets almost silly. To rebuild one reply with a raw model:

• Around 12 prompts per reply, each re-sending 25,000 to 35,000 tokens of context

• Roughly $5 to $7 per usable output, and that's not even top 1% reasoning

• About $5,000 to $7,000 a month at that volume, before 500 hours of human prompting

THIS MEANS WE CAN GIVE MUCH MORE AI ALLOWANCE TO OUR USERS!!! 🫶🏻❤️📈

Finally, the neurosymbolic reasoning is what lets it carry complex, multi-faceted, and cross-functional B2B work all the way through. These are examples you can build and execute on, which is impossible with base-models or tools with wrappers:

• A full go to market strategy, grounded in your ICP, positioning and live market signals, then turned into the campaigns that run it

• A beachhead strategy, picking the wedge segment worth attacking first, sizing it, and sequencing the entry instead of guessing

• Industry trend analysis read across macro, meso and micro signals, so you see the shift before it hits your pipeline

• Investor decks and enterprise sales assets, two pagers, RFPs and proposals that hold up when a sharp reader pushes on them

• An omnichannel outbound engine, from fresh leads to reasoned email and LinkedIn sequences to replies captured and qualified in Smartbox, built and run end to end

And this is only the current stage. We are making the first steps toward a true Jarvis for go to market: a system that can safely, securely and reliably run the work fully autonomously, with no human in the loop.

I hope that makes sense.

@spiri7 Thank you, that really means a lot. And congrats right back for being an innovator-minded person that likes to try novel things.

Hearing it looks like exactly what you need, especially after a thinner tool let you down before, is the best thing I could read this afternoon.

Now, your P.S. That's the most valuable part of your whole comment, so thank you for flagging it. Honest UX feedback like this is how the product actually gets sharper.

You're right that it can feel like doubling up. There is a reason behind it: the ICP step is about sharpening who you target, which is subtly different from the raw company data you gave us. But if it read as "didn't you just ask me this," that's a gap on our side to close, not yours to decode.

So here's what I'd love to do. Let me give you a 30 min walkthrough. I'll show you the tips and tricks, why a few steps exist, and how to get the system really flying for your company.

Would you like that?

@canice_neary this genuinely means a lot, thank you!!!

Coming from a partner at an agency doing outreach in a specialist space is high praise.

If there's ever an angle where the system could serve the Gray Market even better, tell us. Feedback from our customers using it in the wild is how the GTMOS becomes more powerful.

Really grateful to have you on board.

@imtiaj_ahmad That gap is the whole reason why Pebbles Ai exists. Hearing it from someone who ran growth at 20 people is very interesting. You had the talent. You just did not have the war chest for the institutional playbook.

That's what we built. Enterprise firepower without the price tag.

Also, your skepticism is the correct default. "Thinks before it writes" is easy to print on a landing page and still be a templated-wrapper prompt chain.

A standard LLM predicts the most probable next words. That is the whole thing. It has no separate step that asks "is this true, and does it follow the rules." If the sentence looks right, it goes for it, even when it is confidently wrong.

With Pebbles Ai, the neurosymbolic layer adds a second system that reasons with explicit rules and a structured knowledge base.

In other words, Pebbles Ai is a neurosymbolic reasoning system, not a wrapper chatbot.

The neural half drafts. The symbolic half checks that draft against the rules of GTM and against grounded facts before it provides you the output.

If the draft breaks a rule or asserts something that is not in the data, it gets caught and corrected instead of sent. One half writes fluently, the other half checks the writing against logic and evidence. This massively simplified btw, the truth is much more complex.

A concrete example. Ask a normal LLM to personalise a cold opener using the prospect's recent funding. If it does not actually have that data, it will often invent a plausible one, "congrats on the recent Series B," because a confident guess reads better to the model than admitting it does not know.

That is how people end up congratulating a company on a round that never happened, which is a fast way to torch the first impression.

Our symbolic layer only uses a signal that exists in the verified lead data. No real funding event, no funding line. It reaches for a different, true angle instead. The model reaches for a nice sounding sentence. The symbolic layer reaches for a correct one.

The same logic covers strategy. If it drafts a plan that contradicts a constraint you set, or pitches an enterprise motion to a 30-person startup, the neurosymbolic rules catch the mismatch rather than letting a fluent paragraph paper over it.

In our own testing this cut errors to roughly a third of naive prompting, measured on HalluScore. Not zero, we would never claim that. A system that checks its work beats one that only sounds sure of itself.

Happy to run a live one. Give me a prospect and a claim you would want in the opener, and I will show you where it refuses to make something up.

Here are some actual stats:

Claude Opus (MAX) vs Pebbles Ai

• Accuracy: 33% vs 87%

• Precision: 57% vs 91%

• Sales Efficacy, MQL to SQL: 15% vs 85%

• Cost per usable reply: ~$5 to $7 vs $0.012

• Hallucination on rule-bound queries: 31.4% vs under 2%

@tihomiropacic Totally fair point. We don’t expect teams to rip out everything at once. The better path is often to start with a single use case, get value fast, and then expand from there. Pebbles Ai is built to support that kind of step-by-step adoption.

@tihomiropacic Appreciate it. This is exactly the concern we built around. Moving from N tools to a bundle is a real risk, so the sensible route is a foot in the door.

You can start with one module, Fresh Leads for pipeline or Smartbox for outreach, and run it beside your current stack.

No rip and replace, and no burning down the old setup before you have judged the new one. We don't want hostages as customers, we want our customers to be in love with us.

Customer that have Stockholm syndrome 🤣. Joke.

All kidding aside, the payoff grows as you expand. Because it is one operating system with a neurosymbolic brain underneath, each feature you switch on makes the others smarter.

The features within the OS share the same context but have different intelligence. That's why I personally call it your digital marketing-sales department.

Different people, different domain experts, different hard-skills, different expertise, different judgement, and different knowledge.

We built the platform in that image.

For the next 24 hours you can start free, no card, so the foot in the door costs you 3 minutes onboarding.

@benjouss Yeah, open rates are basically noise at this point. Bots, preview panes, and Apple MPP have made the metric too unreliable to act on, so we made a deliberate call to drop it entirely. What Pebbles does instead is classify reply intent: positive, neutral, or negative – so a "thanks, not right now" gets tagged differently from a reply that actually moves the conversation forward. The system is specifically built to tell politeness from a real buying signal. Rather than opens, we track reply rate, positive reply rate, meeting booked rate, a message quality score the AI runs before anything sends, and deliverability and inbox placement.

The "decent opens, zero replies" pattern you described is exactly the problem we set out to fix – because opens are vanity and replies are the only signal worth optimising for.

@benjouss Ha, "decent opens, zero replies" is something I hear often. An open tells you the subject line worked. It says nothing about whether anyone actually wants what is inside.

The silence tells you nothing worked once they did.

When a whole campaign gets opens and no positive replies, it is not bad luck. It is an equation you have not solved yet.

Picture outbound as a formula, not a sum:

Y (outbound success) = Strategy × Positioning × Targeting × Value Proposition × Offer × Netiquette × Communication Sciences × Persuasion Tactics × Lexical Semantics x Persona-centric Writing

One incorrect part of the formula (variable) drags the whole result toward zero, however strong the others are. Bad luck is just the name people give the equation when they have not solved for the variables.

So Pebbles treats a reply as a strategy signal, not just a lead status. A campaign of polite nothings gets read as a diagnostic. Wrong audience, weak hook, or a value prop that does not answer "so what".

The Strategy Assistant is built to pull the formula apart, so instead of running the same dud again next month with new subject lines, you fix the term that was actually near zero.

Or my personal favourite response to a dud campaign: MOREEE!!! More emails, more LinkedIn messages.

Doubling down on a formula that is still unsolved, which is a bit like flooring the accelerator when the handbrake is on.

On telling a real reply from mere manners, it reads for intent over tone. "Sounds interesting, will keep you in mind" is the corporate cousin of "we should grab coffee sometime". Warm words, empty calendar. A reply that asks about price or fit, raises a real objection, or floats a next step is the one that counts. The neural side reads the tone, the symbolic side checks whether anything was actually asked or committed to.

Does that make sense?

@tehreem_fatima5 tool sprawl is genuinely one of the most exhausting parts of running a small team, so I'm really glad this resonates.

What we do make easy is bringing your contacts in via CSV, Google Contacts, or Microsoft Contacts, so you're not starting from scratch.

Deep integrations with products like HubSpot, Attio, Monday.com are already on our roadmap for Q3. And we are not going to stop there and gonna bring our own agent-first CRM solution to support the vision of replacing 10+ tools

@crystal_j As a loyal user of Kimi in China, I feel I have a good say on this. Personally, I think its processing capabilities are truly impressive. The coding experience is great, and its reasoning is always on point. I still vividly remember the massive speed boost from the previous 2.7 update. Will definitely keep supporting Kimi!

@crystal_j so incredibly excited for this! Congratulations on the launch. We can't wait to use it with PromptQL!

@crystal_j AM GOING TO TRY TO BUILD A GAME IF IT PASSES FABLE 5 THAN IF FABLE 5 CAN MAKE A GREAT GAME THIS WILL MAKE IT EVEN BETTER

@zaczuo This is quite astonishing. I'd give it a try for frontend tasks.

@masaki_iino Congrats on the launch! From a product perspective, what surprised you most during early testing? Did users get the biggest value from faster experimentation, AI-generated insights, or something completely unexpected? Those early adoption patterns often reveal where the strongest product-market fit is emerging.

@masaki_iino I'm curious, how do you balance AI-generated recommendations with real user signals? Is there a mechanism that continuously prioritizes insights based on actual customer behavior rather than model confidence alone?

@gavin_porter1 Thanks for asking, Gavin!

I’m currently working with a single-display setup, so I haven’t been able to fully test every multi-monitor configuration yet....

Pocket Screen has a free version, so please feel free to try it in your environment.

If you notice any issues, I’d be happy to look into them for a future update.

@hi_i_am_mimo Thanks, Valeria!

Once a window is pinned, Pocket Screen stays attached to that specific window rather than continuously following whichever window is frontmost.

If the source window is simply covered by another app, the floating view generally stays live—I’ve tested browser video continuing to update while the source was covered. If the source is minimized, macOS or the source app may stop rendering new frames, so the mirror can temporarily pause, but it resumes automatically when the window is restored without needing to be pinned again.

The floating window is not automatically hidden from screenshots or screen recordings, so it will appear when capturing the full display.

@artstavenka1 Thanks, Art!

The floating view is currently a live, passive mirror rather than an interactive remote view, so scrolling, clicking, and typing are done in the original window. The click-through option lets pointer input pass through the floating window to whatever is behind it.

Keeping the experience lightweight and predictable was the priority for this version.

@mason_clark2 Exactly!

It’s essentially picture-in-picture for almost any app window on your Mac.

@galdayan Thanks, Gal!

In the current version, the floating window can be dragged across displays like a normal macOS window, and Pocket Screen saves its last position. With the same monitor arrangement, it should generally reopen where you left it.

However, the current release wasn’t specifically designed for changes such as disconnecting or rearranging displays, so I can’t guarantee reliable restoration in every multi-monitor setup yet. I’m currently adding more display-aware position restoration and recovery for the next update, which I’m planning to submit to the App Store early next week.

@dipankar_sarkar Thanks, Dipankar!

That was one of the key cases I tested carefully. In my testing, a source window can be fully covered by another app and the floating view continues updating, including browser video playback.

Minimized windows are a little different: depending on macOS and the source app, new frames may temporarily stop. Pocket Screen keeps the window association and resumes automatically when it is restored, so it does not need to be pinned again.

I’d like to keep improving Pocket Screen based on feedback like this and make it more practical and reliable for everyday use.

@lennoxbeflying Thanks, Ziang!

The floating window is designed to stay above full-screen apps. One limitation is macOS Spaces: if the pinned source window is on a different inactive Space, Pocket Screen temporarily hides and automatically resumes when that source Space becomes active again.

I’ll continue refining this behavior to make the experience as seamless and practical as possible.

@andrea_j Thanks, Andrea!

You can download Pocket Screen directly from the Mac App Store here: https://apps.apple.com/app/id6788211562

Sorry the download link wasn’t clear!

@aymnart Thanks, Aymen!

The floating window can be resized freely, while the source content keeps its original aspect ratio and scales to fit rather than stretching.

If the source window itself changes size or aspect ratio, Pocket Screen updates the capture and reframes it automatically.

@ramazan725022 Thanks, Ramazan!

That video-and-notes workflow is exactly the kind of everyday use case Pocket Screen was made for. I’m glad it stayed smooth and well framed for you.

@ilko_kacharov Thanks, Ilko! Exactly—Pocket Screen processes the window capture locally on your Mac and does not upload your screen contents to the cloud. Keeping it private and practical for real work was an important part of the design.

@melodie_sh Thanks so much, Mélodie!

That single-laptop-screen workflow is exactly what Pocket Screen was made for. I’m really glad it resonates with you.

@omri_ben_shoham1 Thanks, Omri!

You don’t need to re-pin it each time. Pocket Screen keeps the window attached when you switch Spaces. If the source window is on an inactive Space, the floating view temporarily hides to avoid showing a black or stale image, then automatically resumes when the source becomes available again.

@alex_watson2110 Thanks, Alex!

That’s exactly the kind of everyday workflow Pocket Screen was designed for—keeping a document, reference, or video call visible while staying focused on the main window.

Glad to hear the use case resonates with you!

@andrasczeizel Thanks, Andras!

Pocket Screen is designed to stay lightweight, but resource use naturally depends on what you keep visible. A mostly static document or reference page should have a much smaller impact than video, live dashboards, or fast-moving chat content.

In my testing, a single floating window has remained smooth across Safari, QuickTime, and Chrome, including video playback. Battery and CPU usage will increase more with frequently changing content, so for longer sessions, keeping the floating window smaller and using it mainly for reference material is the most efficient setup.

@paige_lauren2 Yes!

It’s designed to be a simple pin-and-forget tool rather than a full window manager.

@reid_anderson4 Thanks for asking, Reid!

Pocket Screen currently supports one floating window at a time. Multiple floating windows are clearly an important use case, so I’ll prioritize adding support in an upcoming update.

@zcan1758563 Thank you, Özcan!

Keeping the processing local and making the floating window feel smooth and lightweight were both important goals, so I’m really glad you noticed that.

@selinmd2r Thank you, Selin!

Keeping Pocket Screen simple and native to macOS was one of my main goals, so I’m really glad that came through.

@bayraktuta26773 Thanks, Nehir!

Pocket Screen currently supports Option + Command + P to pin the frontmost window, but I agree that customizable shortcuts and a quicker way to switch the target window would make the workflow even smoother. I’ve added this to my improvement list.

@james_ketteringham I'm curious, how does Timely handle users with multiple calendars and dynamic availability? Does it intelligently combine personal and work schedules while still giving users granular control over what gets shared?

@james_ketteringham its amazing Gmail hasn't solved for this. in, fact I couldn't even tell you where the icon is to add my cal. this is a great idea. I like the / aspect. Maybe think from from a sentence aware aspect for future improvements.

@james_ketteringham Congrats on the launch. Small typo "pressing its number" should probably be "their number" (referring to the exec).

Great concept otherwise!

@tansuutlu5uei That's a great idea - perhaps even a / shortcut as you type...

@dipankar_sarkar Thanks! Yes - it's fully client-side, no proxy. Both Outlook and Google refresh tokens are stored encrypted in the OS keychain and refreshed silently in the background (MSAL's silent flow for Microsoft; refresh-token exchange for Google), so you don't re-auth periodically. You'd only need to reconnect if you change your password, revoke access, or your org's admin policy forces a re-login.

@omri_ben_shoham1 Thank you! Yes it'll handle back-to-back meetings and blocked slots. You can set buffers too, to protect time in between meetings already on the calendar.

@brahimiperkej1 You got it!

@utkufqkz Thank you! This is already handled - we have working hours and safety limits to ensure you're never suggesting an unreasonable time!

@ridhwikvinod Absolutely - if you head into the settings tab, you can set safety limits to ensure you're never offering a time that would be inappropriate for the recipient. Image below.

@maria_wall_ball Thank you! Don't forget to use the code for 3 free months! Code: PH3MONTHS

@vollos Thank you! When adding someone else's calendar, the exec will either have to share their calendar with your account - or you'll need to log in to your executives account. Both options work, depending on how the exec is currently setup.

@andrei_rebrov1 Thank you! It's customisable too - depending on which timezone you want to send availability in. Inferring from domain is super smart - we'll look into how we can achieve this.

@tarcan_hak99601 That's a great suggestion. Thank you!

@masal1468301 Thank you!

@ege1g2 You got it - we're keeping it simple (and old school).

@rojintahanjgqn Thank you! I initially built this for Executive Assistants (given I'm well connected in this space) - and for them, it's super important that they're protecting their executive/principals privacy, given many of them are dealing with highly sensitive engagements.

@emmanuel_costa5 Exactly the reason I built this! It was such a faff.

@darly_selby Great question! At the moment, if the connection fails, the affected calendar won't appear as an option when selecting which calendar you'd like to pull availability from. You've made a great suggestion though, rather than it being hidden, we should show an indicator encouraging them to reconnect their calendar. This would drastically improve UX.

@varun1jan Completely agree - before this tool I've made a couple of errors in timezones and it always results in me looking a little silly, and spending extra time having to go back to the person and correct my availability.

@johnnvho Thank you! Download the app (mac or windows) - then connect your calendars. Hit CMD + Shift + A (you can change this in the settings) and go through the steps on the terminal style window!

@maxmusing My worry with AI analysts is confident-sounding numbers quietly built on the wrong join or filter. How does it show its work so a non-technical person can sanity-check a suggestion before acting, and will it flag when it's unsure instead of just answering?

@maxmusing Congrats on the launch. Proactive BI is a useful shift. How do you evaluate whether suggested analyses are genuinely useful versus just plausible-looking questions based on past dashboards and chats?

@maxmusing how does the suggestion engine adapt over time? Does it learn from which recommendations users engage with, dismiss, or turn into dashboards so future suggestions become increasingly tailored to each team's decision-making patterns?

@michael_shollenberger yes, accuracy is the most important part. We surface the full SQL query for review, and we also have a semantic layer built in that lets you define your metrics once (e.g. MRR, activation rate) and have the AI deterministically reference them in charts and reports.

@krekeltronics for sure. We built our core agents with all of those important qualifiers built in, so our suggestion system now just needs to surface good ideas and pass them along to our core agents.

@dhiraj_patel5 right now just surfaces the suggestions, but that's a cool idea!

@abhineetarora good question. We intentionally designed the suggestion system to be personalized per-user, not across the whole organization, so suggestions are always based on areas of the data that matter for each user. There was a lot of work tuning the system to make sure it always had good signal and didn't resort to noisy slop.

@grselkalpcztqo agreed on no account needed to use something. It should be the standard

@siyan_chen1 I've hit the wall with browser agents on the 5% of edge cases that break a recorded flow. When a layout changes or a step fails, does Aye stop and ask or improvise and risk the wrong action? And how much correction before a task actually sticks?

@siyan_chen1 when users teach Aye a workflow once, how resilient is that Skill as websites change over time? Can Aye adapt to UI updates and minor layout changes on its own, or does it require users to retrain the workflow? It feels like that adaptability could become one of Aye's biggest long-term advantages.

@siyan_chen1 As users build dozens of reusable browser skills, how do you plan to organize, discover, and recommend the right one at the right moment?

@omri_ben_shoham1 That’s an important security question.I couldn’t find anything public that explains exactly where Aye stores credentials or session data, so I don’t want to make something up. What I could confirm is that Profile windows keep tabs, cookies, sign-ins, and site data separate, and Aye hands sign-in back to the user.

It also reads the current page and checks results instead of simply replaying recorded clicks. It’s probably best to confirm that directly with the maker: https://okaapps.com/support

@dhiraj_patel5 I checked Aye’s own FAQ on this. Sign-in and CAPTCHA are handed back to the user, while form submission, payment, and other high-risk actions require explicit approval.

So it shouldn’t try to push through multi-step authentication or a sensitive confirmation on its own. The user stays involved at those points.

@duyguerayaa60h yep, that stood out to me too, and it’s one reason I was excited to share Aye. Letting it handle routine steps while keeping sensitive actions under the user’s approval feels like the right balance.

@andrewzakonov Artem raised a similar point above, so I asked Aye about it and I also tried it. As I mentioned in the reply, a correction sticks for the current task and its checkpoint, but it doesn’t automatically change other workflows or future sessions. If you want it to become part of the taught flow, you save it back into the skill.

I actually like that boundary. One correction shouldn’t quietly change everything else.

@shivangit26 That’s a fair question. Aye works from the page in front of it and checks the result instead of only replaying a fixed path. If a task is interrupted, its checkpoint keeps the current progress, so it can resume instead of starting from zero.

Sign-ins and CAPTCHAs are handed back to the user. That doesn’t mean every page change can be handled automatically, but it avoids blindly pushing ahead when the flow no longer matches.

@advin_jadis I read “mistakes” as a saved skill doing the wrong thing during a run.If a saved skill does the wrong thing during a run, your correction applies to that task and is kept in its checkpoint. It doesn’t automatically rewrite the skill for future runs, though. You need to save the change back into the skill.

Work Records already show whether each run finished, stopped, or failed. A simple before-and-after view of skill edits would make those changes much easier to follow. Aye also has a creator incentive program for useful new or improved skills: contact@okaapps.com

@fahri1585951 Good news, you shouldn’t need to rebuild it. Aye’s Profile windows keep tabs, cookies, sign-ins, and site data separate, but your skills are shared across profiles. So you can open a work profile and use the same skill you created in your personal one.

If it’s not showing up for any reason, you can reach the maker at https://okaapps.com/support.

@pixycad Good luck!

@pixycad Since your focus is rapid iteration, what product metrics tell you a user reached their "first successful model" faster than they would in existing CAD tools?

@pixycad What's your experience with 3d printing, and are you gonna add parametric modeling?

@sutton_willow A freinnd of mine is always looking for active people to follow in the startup space. I think this would make it much easier to discover new voices instead of seeing the same accounts every day.

@sutton_willow one question I have is whether users can filter the leaderboard by category, like AI, developer tools, or SaaS. That would make it easier to find people working in specific areas.

@sutton_willow The name definelty caught my attention. It's playful, memorable, and matches the personality of the product really well.

@huriyeqknk Great to hear! That's exactly the kind of gap we built this to catch. Thanks for trying it and sharing the result.

@ali9slr Noted, good one. Thanks!

@merve1519747 Great to hear! Thanks!

@ahmetkok Thanks! I really appreciate the detailed feedback. we took notes.

@teomandemirhan Thanks Teoman!

@srazgencilswsc Thanks! That’s exactly what I built it for. If it helped you discover products you would’ve otherwise missed, then it’s doing its job. Really appreciate the feedback! 🚀

@demet46p6 Thanks, I really appreciate it! ❤️
That’s a great idea. A personalized feed based on favorite categories, makers, or keywords would make the daily recap even more valuable while keeping the main Top 50 intact. Definitely adding this to my roadmap. Thanks for taking the time to share such a thoughtful suggestion!

@habibevxxp Thanks a lot! Really glad to hear that 😊

I actually love this idea. A persistent “Try Later” list that survives the weekly reset sounds like a natural fit and would make Daily Top 50 much more useful beyond just browsing. I’ll definitely add it to my roadmap. Thanks for the suggestion!

@major_grooves congrats Steven and team!

@vikramp7470 yes, concurrent data access is not an issue for most of the tools the server provides. There are a few tools though that wouldn't make sense to use concurrently: namely everthing that drives the UI state (switch pages, open diagram, etc.). But everything data related is safe.

@mlobo09 Rabbit hole fits very well for this kind of challenge. It's always interesting to see how people completely misjudge how complex it can get. Well, I guess there is a reason that despite being an 80 year old problem, there is no perfect solution yet for ER. Or should I say "was". ;)

@peadar_coyle3 Glad to hear that you like it. As a matter of fact, we have quite a sophisticated and well working approach for company name matching (see Exiger case study on our website). Let us know if you're struggling next time, we might be able to help.

@cschagen primarily our production deployments we run on AWS, but we can also do on-prem (which would mean we can deploy on any cloud) via a container. Just need to be able to handle the Devops side of things. Let's have a chat after the launch to discuss your use case.

@janoberhauser Thanks Jan! Appreciate the feedback.

@varun1jan True. We see a huge benefit for LLMs to not judge on the data themself, but to first unify it and then make it accessible. This provides accurate context for LLMs in various use cases.

@shivangit26 Currently the Tilores Studio uses optimized fixed rule sets for the two built-in use cases (person and company matching). While they are a good start to test the general process on customer side with the least efford, we typically fine tune the matching process together with the customers needs during onboarding or evaluation periods to get the best results. The result of that is a solution that can be either hosted by the customer themself or by us and can easily be implemented in existing pipeline. I'd recommend to look at a few of the use case studies on our website that provide further details.
We're planning to also offer customization options in the Tilores Studio directly. Expect further improvements comming soon. :)

@semihkodaruge If you like the speed of the Studio, you would love the speed of the SaaS solution. We can easily scale into tens of thousands of records concurrently.

@charlotteschmitt Thank you very much. Personally I love the MCP integration a lot since it easily lets me automate the standard tasks when using the tool for customer onboarding. Basically I can automate the whole process from data ingestion to automated reports for customers. The wide variety of visualization options also helps a lot and the MCP server makes it easy to export them and include them directly in the report.

@semaslnhatdugf Thank you for your feedback! We've put a lot of actual work into the docs.

@semaslnhatdugf nice. Not often people pick out the API docs, but I am told they are indeed quite good.

@habib_daigency pretty much any potato should be able to handle 100k records - challenge usually starts at 1M or more records for badly optimized algorithms. Our SaaS/on-prem versions can easily handle hundreds of million records - by default they are built on a serverless stack and as such users wouldn't have to worry about machine requirements.

@kuzey487176 Thank you! Entity resolution alone is complex enough. Let's not make it harder by using too technical terms.

@eymenyabasoqjc For our SaaS solution we offer exactly that. A visual editor for customizing all rules. We expect to have the same functionality in the studio comming soon.

@makimum_dev I like that Vetta X is focused on helping people engage consistently instead of just generating posts. Reply alerts, AI-generated drafts that match your own writing style, and one-tap publishing could make it much easier to stay active on X while still keeping your authentic voice. The fact that you remain in control of what gets published is a nice balance between automation and personalization. Best of luck with the launch!

@nurmukhamed Best result is 165k views 45000 engagements +100 new registered users.

@selahattinkwt9 Thanks for honest feedback! That sounds like a truly useful addition to have.

@ethan_lee8 It is completely safe, we are using X API for what we are posting on your account. Your account wont be shadowbanned and there is no restrictions from X. We don't have any cases of X restricting content that our users post.

@kayrahyua Thanks, Kayra! Glad the headers held up, that's usually where things break on long reports. Tip: multi-page exports merge into one file automatically, so no manual re-stitching needed.

@andrei_rebrov1 Thanks, Andrei! Right now it stacks: each PDF page becomes one sheet, and everything on that page (including multiple tables) lands in one grid based on x/y position, so it follows the original layout instead of getting jumbled. For scanned PDFs the AI mode actually detects individual tables and can split them into separate sheets, that same option for text PDFs is on my list. Appreciate the kind words on the launch!

@andrei_rebrov1 Thanks, Andrei! Depends on the mode. In AI mode (used for scanned or messy PDFs), each detected table gets its own worksheet by default, so two tables on one page end up on separate sheets. In the standard local mode for text PDFs there's no per-table detection yet, everything on a page lands in one grid, so multiple tables currently get merged onto one sheet. Splitting them there too is on my list. Appreciate the kind words on the launch!

@derinbornoksyj Glad it worked well, Derin! Quick clarification for anyone reading this thread: native text PDFs are parsed 100% locally in the browser. Scanned docs, like your bank statement, go through the AI mode over HTTPS since OCR needs that, but files are processed and discarded immediately, nothing gets stored. Thanks for stress-testing it!

@brandon_chase Thanks for the suggestion! 🙌 We completely agree—a before/after slider works really well for editing tools like background removal or photo restoration. For some of our generation features, though, the output can be very different from the original, so side-by-side comparisons tend to tell the story better. We're always looking for the best way to showcase results, and your feedback is really helpful!

@ella_cooper Thanks for giving us a try! 🙌 Hair and fine edges are definitely some of the toughest cases for any background remover. We've put a lot of effort into improving edge detection to preserve details as naturally as possible. We'd love to hear what you think after you've tested it!

@aurora_parker Thanks for the great question! Consistency depends on the model and the prompt, but we've optimized DreamPixel AI to keep styles as consistent as possible across batches. We're also exploring even better workflow features—like reusable style presets—to make consistent results even easier.

@sedanurazaaa70 Thanks so much for the suggestion! 🙌 You can already reuse prompts from your My Images history, so it's easy to recreate or build on previous generations. That said, we agree that dedicated style presets—where you can save an entire setup (prompt, model, style, settings, etc.) and apply it with one click—would be much more convenient. It's a great idea, and we'll definitely consider adding it!

@ashton_blake Thanks for bringing this up! 🙌 We currently support 4K image upscaling, which is available on our higher-tier plans for users who need maximum quality. You're absolutely right that we should make the output resolution and print suitability much clearer on the site—we'll improve that. Thanks for the helpful feedback!

@bradley_simon Great question! 🙌 We believe the best AI creative experience is a balance between speed and quality. DreamPixel AI focuses on delivering high-quality, detailed results while keeping the generation process fast and smooth. We continue to optimize our models and workflows so creators can spend less time waiting and more time creating.

@nora_mitchell Thanks for asking! 🙌 DreamPixel AI isn't built around a single proprietary model. We integrate multiple leading AI image models and focus on delivering the best results through an intuitive workflow, carefully designed prompts, and optimized image processing. Our goal is to give users the right model for each creative task without the complexity.

@oliver_hayes1 Thanks for the question! Batch uploads aren't available yet, but they're definitely on our radar. We know they're especially valuable for users working with large numbers of images, and we'll keep this in mind as we continue improving DreamPixel AI.

@nurcanzalp1zl8 Thanks so much! 🙌 That's exactly what we were aiming for. We believe powerful AI tools shouldn't feel overwhelming, so we focused on keeping the workflow clean, intuitive, and fast from the very first upload. Really appreciate you noticing the details—and thanks for the support!

@exopaul Thanks for pointing this out! 🙏 You're right that DreamPixel AI offers 10 free trial credits rather than unlimited free usage. We really appreciate the feedback, and we'll make this much clearer in our messaging so expectations are set correctly from the start. Thanks for helping us improve!

@nurayg61521 Thanks so much for trying DreamPixel AI and sharing your experience! 🙌 We’re really happy to hear that the portrait lighting and painterly variations surprised you. Creating images that feel natural and believable is something we care a lot about, especially when transforming personal photos. Thanks for the support—we’d love to see what else you create!

@berkayotip Thanks for the detailed feedback! 🙌 You’re absolutely right—working through a whole photo set one by one can become very repetitive. A bulk upload and generation queue workflow would be a huge time saver for creators who need to process many images. We’ll definitely keep this in mind as we continue improving DreamPixel AI. Really appreciate the suggestion!

@masal1428947 Thanks! That’s exactly what we were going for. Really glad the in-block editing feels more natural.

@xfei Exactly. Meeting notes are much more useful when you can actually ask them questions later instead of digging through old transcripts. Really glad this clicked for you.

@parsons_wu_real Exactly. We wanted Phrase to be a real place to think and write, not just somewhere your transcripts pile up. Really glad that came through.

@serapzpek3qih Thanks! We’re constantly refining the in-note experience, and this kind of side-by-side diff is already in progress. The goal is to make AI rewrites easier to review and accept without interrupting your flow.

@n_lok72335 You’re describing exactly where WizeMe’s Weekly Pattern Scan is headed. The partners already identify recurring blind spots, postponed decisions, and behavioral patterns, and WizeMe’s opt-in notification layer supports email, Telegram, and push. I’m tightening those pieces into one user-controlled weekly digest, with source boundaries and no silent inbox access or unapproved sending.

@omri_ben_shoham1 Fair pressure, and you're right that storage was never the hard part.

Here's the actual mechanism. It's not a background job quietly deciding to interrupt you — it's a structural loop you run, not one WizeMe runs at you:

The night-to-morning bridge. Nightly De-brief asks one question — "how did today land?" — and structures whatever you say into open loops, decisions, people, and one tomorrow-intention. Whatever you leave open comes back in the next Morning Brief, by name. Not searched for. Not buried in a list you have to open. It's re-surfaced in the one place you're already looking at the start of the day.

That's the trigger: you closing yesterday is what queues what shows up tomorrow. The loop is user-paced, not push-based — there's no silent notification deciding for you what's important. If you skip a night, it bridges the gap without guilt or streak-shaming; nothing punishes you for not feeding it.

Above that sits Weekly Throughline — it looks for a repeated loop, a decision you keep drifting on, a pattern across several nightly closes — and it only surfaces when the signal is actually strong enough. That's the deliberate answer to your exact worry: it's built to not resurface noise. If nothing repeats clearly, it says nothing, rather than manufacturing a pattern to justify existing.

So the honest shape of it: this is not omniscient proactive surfacing — it's not going to interrupt you mid-afternoon with "hey, remember that thing." It's a daily structural rhythm — close the loop at night, get it back by name in the morning — plus a weekly pattern check that stays quiet unless the evidence is real.

The fair name for the risk you're naming: it only works if you're in the daily rhythm. Skip the nightly close for two weeks and there's less for it to bridge — same as any partner that needs you talking to it to know what matters. That's a real tradeoff, not a solved problem, and I'd rather say that than pretend it isn't.

@gullu60488 That instinct — audit it like a statement — is exactly the right one for anything holding memory about you, and I want to answer precisely rather than round up to "yes, we have that."

What's actually there today: memory is source-labeled at ingestion — notes, recordings, threads, documents, uploads all carry where they came from, and if something wasn't loaded, the partner says so rather than pretending. You can say forget this and a specific item is gone, not soft-hidden. And you can ask in-conversation to audit the last three turns for a blind spot — a live check on what it's actually using, in the moment.

What's not there yet, honestly: a dedicated weekly timeline view — remembered/edited/forgotten, laid out the way a statement lays out transactions. Right now the audit trail exists as labels on the memory itself and point-in-time checks you ask for, not as one rolled-up view you'd open and scan.

That's a real, specific gap, not a small one — a statement view is a different thing than "ask and it'll tell you." The primitives to build it are already there (everything's source-labeled, everything's forgettable), it's a surfacing problem, not a data problem. Worth putting on the list with your name on it.

@gullu60488 Update for accuracy: Weekly Pattern Scan is live — it's the feature I was describing, correctly named this time. (I called it "Weekly Throughline" in a reply further down this thread; same underlying mechanism, I was sloppy with the name, not the mechanism.) It looks across recent nightly closes and surfaces a repeated loop, blind spot, or decision drift only when the signal is strong enough — not a forced weekly report regardless of whether anything's there.

The digest-delivery layer on top of it — one user-controlled weekly email/Telegram/push summary — is what I'm building now.

@jlaramie congrats! a novel idea in browser extension land!

@joe_parrish2 - we're looking closely at the others in the top 10; from Ahold to Aldi to Wakefern and then some key regional players like a Wegmans and Meijer.

In terms of the data, we're starting to engage with several inbound b2b inquiries looking to understand and track pricing, competitor products, etc.

@glhanbinatwoef you got it - we'll look into it. thank you!

@sefa3n8y woohoo - love it! thank you!

@ozdenoglu41658 Love it - that's super straightforward to implement. Happy to do it. Thank you!

@neriman91088467 totally. it's planned in two releases...we'll automatically coordinate across the retailers the available pickup / delivery windows, select them in the best group, and consolidate into the shortest window!

@neriman91088467 how many stores do you usually shop at each week? and what would be your top 3?

@reillybrennan Thanks Reilly! It was a big technical task to do SKU matching, validate pricing and availability and agentically move it to the next best retailer. 350M prices a day we're tracking so there's a lot of moving parts!