PH热榜 | 2026-06-24

Hey Product Hunt 👋 Kitty here, product lead for Tencent EdgeOne Makers.

Over the past year, I've watched more and more people—including myself—start building their own AI Agents.Today, building an Agent has never been easier. A solid idea and a few hours gets you a working demo. But the real work starts after the demo ships.

Suddenly, you're hit with a wall of production questions: How do you manage memory? How do you run tools securely in sandboxed environments? How do you trace and debug execution paths? How do you scale when a hundred users hit it at once? And how do you deploy it globally so it's actually fast?

Most builders end up choosing between two painful paths: spend weeks building all of this boilerplate infrastructure from scratch, or lock themselves into a restrictive platform that dictates which framework, language, or model they have to use.

We wanted a third option. That's why we built Tencent EdgeOne Makers.

Tencent EdgeOne Makers is an edge platform for modern web apps and AI Agents. It fits into the workflows developers already know, with familiar CLI, Git, and CI/CD support. You get Agent runtime, sandboxed tools, memory, observability, model gateway support, serverless functions, and storage built in, without having to stitch together complex infrastructure yourself. In other words, you can deploy AI Agents the same way you deploy web apps.

We kept the platform completely open. No vendor lock-in, no framework constraints:

• Framework agnostic: Works out of the box with Claude SDK, OpenAI SDK, LangGraph, CrewAI, and more.

• Polyglot: Full support for both JavaScript and Python.

• Flexibility: Use whatever model or tech stack makes sense for your application.

Whether you’re looking to plug an Agent into an existing SaaS, website, or e-commerce flow, or you're building a brand-new AI application from scratch (like an AI recruiter, sales rep, data analyst, or fitness coach)—Tencent EdgeOne Makers is designed to let you spend your time on your product, not the plumbing.

We're excited to share this with the Product Hunt community today. Give it a spin, ask us any tough questions, and let us know what you think! Feel free to join our Discord to chat with us.

Thanks so much for the support!

Congrats on the launch!
For framework-agnostic support, does the sandboxed tool execution behave the same across Claude SDK, LangGraph, and CrewAI, or do some frameworks get more native support than others?

Congrats on the great product! The edge angle is the interesting bet. Edge runtimes are usually tuned for short requests, but agents that actually do work run long, with memory and retries between steps. How do you guys handle an agent that runs for minutes or picks up a scheduled task later?

Since this is polyglot (JS + Python), can a single agent mix both, or is it one language per deployment?

congrats on the launch!
the idea of treating AI Agent deployment like web app deployment is really interesting.how does Makers handle long running agent workflows that may span hours or even days?

Congrats on the launch! Interesting direction. What does debugging look like when multi-step agent workflows fail across tools and runtimes in production?

Web and agents in one project with unified deploy is a really smart move. Less context switching, less glue code.

Congrats on shipping this. Agent demos are easy now, but production is still messy. Memory, tracing, scaling, storage, all the boring stuff is exactly where most projects slow down.

Finally, agent hosting that doesn't make me wire up five services first. Nice work!

Nice launch! The deployment experience looks really clean.

One question: how do you handle long-running agent workflows? For example, if an agent needs to wait for user approval or perform multiple asynchronous steps, is there built-in state management or checkpoint support?

The honest truth is most of us under-budget the "make it production-ready" phase by like 3x. A platform that absorbs that is genuinely valuable.

Running Web and agents from the same project is a clever architectural choice. Less to manage, fewer moving parts.

This feels like a good fit for teams moving from AI experiments into production. That transition is still more painful than it should be.

Congrats on the launch! The no lock-in part is important. A lot of teams don’t want to bet their whole AI stack on one model or one framework.

curious how edgeone handles the growing wave of AI agent traffic specifically. traditional CDN caching works great for human browsing patterns but agent requests tend to be API-heavy, bursty, and less cacheable. is there anything in the stack tuned for that kind of workload or is the focus still primarily on conventional web delivery?

congratulations to the team! i appreciate the focus on reducing infrastructure complexity.how does pricing scale for teams that suddenly experience rapid user growth?

congratulations on the launch! supporting both Python and JavaScript is a huge plus.are there plans to expand support for additional languages in the future?

The fact that I can start from a working example instead of a blank repo is underrated for actually shipping.

Congrats Kitty and team. The sandboxed tools part is the most interesting to me. Do you have examples of what kind of tool permissions or isolation developers can control?

It's interesting how much of user experience depends on things people never see directly. Faster load times and reliability rarely make headlines, but users definitely notice when they're missing. Nice work.

The fact that I can start from a working example instead of a blank repo is underrated for actually shipping.

Congrats @kitty_lee1 and team. The sandboxed tools part is the most interesting to me. Do you have examples of what kind of tool permissions or isolation developers can control?

Love the "push and it runs" approach — shipping agents has been way too painful. Upvoted!

Edge deployment for agents = low latency for users worldwide without me thinking about regions. That's a strong angle.

As someone who ships side projects on weekends, "live in minutes" is the dream. Can't wait to throw an agent at this. Congrats on the launch! 🙌

Congrats on the launch!

nice product, nice team

Coming from stitching together a bunch of cloud services, having this as one coherent thing is refreshing.

Congrats on the launch!

Is "sandbox" referring to the browser's sandbox environment?

Hey Product Hunt 👋

Our founding team has been building products for 10+ years at scalable SaaS and deep learning AI companies. We wanted to fix our own problems. The way product teams work has not changed much in a long time, and we think it is time for that to change.

Over the last several months we went very deep with hundreds of product teams. Two camps: stuck in legacy tooling or building everything from scratch. What surprised us is that both are doing 90% of the same work. Collecting customer feedback. Shaping and evaluating ideas. Handing off to coding agents. Every team. Every time.

The same shift Cursor brought to engineering is coming for product.

How you understand your customers at scale, how you operate and strategize, how you collect and make sense of everything. All of that should just be provided for you. Automatically, with your context, your intent, and your personalization built in.No one should be rebuilding that infrastructure individually, at every company, every time.

We are building Propane for the people who want to focus on the primary work. To think, strategize, and shape the future of software. To have time to talk to customers, look at the market, and build products that everyone loves ❤️

How Propane works

• Collect: we pull everything into one shareable context. Your customer and market brain, for your entire team and agents.

• Collaborate: you and your agents work together in one shared workspace. No copy-pasting, no context switching.

• Commit: from that same context, hand over to your coding and design agents with a canonical data set built for meaningful outcomes.

Everyone has access to the same context. No more sharing documents across other systems. Everything compounds and stays in one place where humans and agents can work together.

We think pricing should be different.

We want to make this more accessible and more valuable. That is why we are introducing one price, as many users and tools as you need. You only pay for the new context we find and index each month. Capped. No surprises.

Our offer to the Product Hunt community: use code PH001 for three months free on the base plan.
That is $150 in value, on us.

If you want to know more:

• Pricing: https://www.usepropane.ai/pricing

• Changelog: https://www.usepropane.ai/changelog

• Try it: https://app.usepropane.ai/auth/signup

Sign up, try it, and help us shape the future. We are reading everything.

Best,
The Propane team ❤️ 🚀

The ''both camps doing 90% the same work'' point really lands. We're the built-from-scratch type and honestly half our internal tooling exists just to move context around.

Let’s goo!!🚀 most excited to be part of building signals, my favourite part of propane, that gathers the most useful product context where I already work, and then makes that simple and accessible.

We really built this for ourselves as builders to actually build what matters - would love to speak to anyone building in the ai-era that feels limited by context, that was me!

Amazing!

Everyone on a product team can build now. That makes one thing the real bottleneck: signal. Not what to build, but why it’s worth building.

On our team every engineer ships end to end and picks their own problems to solve. Getting them (and me as a PM) to the “why” faster, with confidence, is exactly the gap I'm feeling right now. Been waiting for this one. 👀

At @GameAnalytics, our product team has been an early alfa → beta tester of Propane, and we’ve had a very exciting experience experimenting with it. The direction is interesting: bringing customer context, product thinking, and agent handoff closer together in one workflow.

Great to see it launching publicly 🚀

Flame on! 🔥 So excited to finally launch this here on PH! 😽

We really put our blood, sweat and tokens into this one. the part i'm most excited about: for any PM, the days of context scattered across folders on your system are over — it now lives in one place you can actually work and collaborate in, a shared doc or prototype.

We built this for ourselves first, because we where drowning in scattered context, re-explaining the same stuff every time — so if that's you too, we'd genuinely love to hear your feedback! 💭

I've burned through a lot of tools trying to solve the same problem, keeping design decisions grounded in actual customer reality instead of internal assumptions.

What I like about Propane is that I went in thinking "this is for PMs." It's not. As a design lead, the context layer is where I spend a disproportionate amount of time, pulling from research, support threads, sales calls, Slack, just to make a case for a direction. Propane collapses that. When you're committing context directly to a design agent, you're starting from signal instead of noise.

This is the right problem to solve. ❤️💪

What does that handoff look like, generating a spec or directly creating tasks in a tracker?

Looks very polished! What vendor did you use for document processing/multiplayer text editors?

Looks very interesting - is there a minimum of data (maturity of company / product) that is needed in order for these insights to be valuable / actionable?

Great work and congrats on the launch. Am I correct to assume the always-current, auto-collected part is the hardest element? Wondering what happens when two tools disagree about the same customer (ex. - sales notes say one thing, support tickets another) - what do you treat as truth before committing that to an agent?

The day has finally come 🔥

Super excited to get this out and see what problems you'll solve with it.

Being on a team where everyone believes so deeply in what we're building tells me this was made with real heart - and I can't wait for you to experience it. Go check out what we made for you.

I've been watching what you've been building for a while now and is genuinely excited for this.

With the teams you've been testing with, where do teams typically see the biggest lift first: support speed, product discovery, or better handoffs between teams?

And how do you see your tool stack up against, for instance, harvestr?

And here we go!!!!

I watched product teams spend years deciding what to build based on whoever talked the loudest that week, while the real evidence of what customers needed sat scattered across support chats, sales calls, and notes nobody had time to read together.

With Propane, the reasoning behind a decision finally sits next to the feedback that shaped it, and an AI agent can use that same context to actually help instead of guessing.

It's been a pleasure building this for you all!

@greenlieber This feels like a real product team problem. Collecting feedback is not the hard part anymore; the hard part is turning scattered customer context into decisions without losing the reasoning behind them. I like the idea of giving both humans and agents the same shared context instead of forcing everything through separate docs and handoffs. Congratulations.

Really like this. Keeping customer context current automatically is the piece that actually matters, since every other tool in this space just hands you another doc to maintain. The shared canvas and committing straight to an agent are great touches too. Will be trialing it. Congrats on the launch and the #1.

Great idea. what part of the product do users usually understand the fastest?

@greenlieber very cool, congrats on the launch!! Would you mind going a bit deeper into to the type of information / data that is fed into the product? You touched on it in the video but would love to hear a little bit more about that. Thanks!

Looks promising, but what's the biggest reason teams decide not to move forward with Propane?

@greenlieber Congratulations. And happy product launch.

the idea of giving agents and product teams the same customer context is smart. right now most teams have the PM reading one set of signals and the agent working off something completely different. does propane handle cases where the data conflicts though? like support tickets pointing one direction but usage data saying the opposite?

Nice! So intrigued - will have to test

The canonical context idea is the right problem. The release risk is provenance: if an agent is going to commit from customer context, the team needs to see which source changed the recommendation and how stale that source is. Otherwise the shared brain quietly turns into another untrusted doc pile.

When a customer's context changes after a coding agent has already committed work based on the older context, does Propane flag that drift, or is keeping the agent's snapshot up to date on the team?

Congrats on the launch 🚀 Propane feels like a tool I’d actually want to try, connecting workflows with real intelligence is exactly what I’ve been looking for

Congratulations on your launch! To gain maximum leverage in an AI era business context is the secret sauce - both for building the right products & for winning the distribution game. Building out the winning USPs is all about understanding the space, and foundational to this is context.

How do you collect and synthesize context across sources, and how do weigh the importance of each piece of information? A classic example is a freemium user saying; I'd buy your product if the UI was french, and a red account or $1m lead says; if you had a plug and play integration for SAP, I'd sign tomorrow.

Synthesizing and actively reasoning across the feedback to surface the true value - how do you think about and solve that?

Congrats on the launch!!

I had the pleasure of giving Propane a spin with @atherkildsen and I have to say that I believe they're really onto solving "the real" headache for product managers and anyone else working with qualitative (plus quantitative) data collection. Distilling insights from phone calls, emails, usage patterns and shower-thoughts used to be an art form. Propane helps make the process transparent and solid from insight to deployment. Junior PMs get a nudge towards quality and senior PMs get speed and collaboration on the artefacts that matter.

Excited to see where the team will take this!

What are you looking at next @greenlieber ? I suppose with launch on product hunt there will be a lot of insights for you to capture now, and distill? ;)

As a designer, the context layer is where I lose the most time. It takes loads of time pulling from research, support threads, and sales calls just to justify a direction. Love how you guys are solving this problem with Propane! Congrats on the launch.

Hey Product Hunt, I'm Vincent, Co-founder of Crewdle!

A simple question

If you could hire an assistant, an automation engineer, a content creator, and a web designer for your business tomorrow, you'd do it in a heartbeat. The problem? You can't afford four salaries, and most small businesses can't either.

So owners settle for the next best thing: a pile of software. And that's where it falls apart.

❌ The tools don't talk to each other — Your chat tool doesn't know what your automation tool is doing. You become the integration.

❌ You pay whether you use it or not — Five subscriptions sit on your card every month, draining cash even in slow weeks.

We built Crewdle so you don't have to choose between an expensive team and a frustrating pile of apps.

Think of it as your AI crew

Crewdle is a single workspace where each "crew member" handles a different job, and they all share the same context:

🔹 Chat answers questions and drafts on demand.
🔹 Connect runs autonomous agents around the clock.
🔹 Forge turns plain-English requests into working automations.
🔹 Create produces your content.
🔹 Build ships your website.
🔹 Admin keeps every dollar visible and under control.

One login. One bill. No integrations to babysit.

The part owners love most

You pay for what you actually use, not a flat subscription. Quiet month? You spend less. Busy month? Your crew scales with you. Most small businesses land between $20 and $40 a month.

Who is this for?

Bakeries, accountants, contractors, studios, solo founders, anyone running a real business who wants AI to do the work, not add more work.

Get started today

Hire your crew free at crewdle.com. No credit card required.

@vincent_lamanna Strong direction, unifying chat, automation , and creation in one workspace can genuinely remove a lot of tool fatigue if the experience stays seamless in practice.

@Crewdle Good idea bringing everything into one place , if it stays smooth in real use, it could save a lot of switching between tools.

@vincent_lamanna Love the pay as you go idea. What made you choose that over subscriptions?

@vincent_lamanna How steep is the learning curve for someone non-technical?

@haylli_weintraub What's the first feature most people end up using?

@vincent_lamanna Does everything really work together without extra setup?

Build spinning up real working web apps from text is a bold promise. Huge congrats for shipping , qq does it spit out a completely modular stack like react and tailwind that we can export via a CLI, or are the generated sites locked entirely into the crewdle runtime container?

I can see local business and service based companies getting value from this. Having content, automation and a website builder together removes a lot of setup work.

The subscription prob is real. I've seen businesses pay for tools they only touch a few times a month, so the usage based approach definitely stands out.

For someone completely new to AI tools where would you recommend they start inside Crewdle?

Most all in one platforms end up doing a lot of things okay but nothing great. Which part of Crewdle do users end up relying on the most?

The automation side caught my attention. Can Forge handle multi step workflows or is it mainly for simpler tasks right now?

Love the pricing angle. What's been the biggest surprise after switching customers from multiple subscriptions to usage based billing.

"Stop juggling five AI subscriptions" hits hard. I'm running 4 right

now and the monthly bill compounds quickly. Pay-as-you-go is the right

pricing model for this category.

Curious how you handle the model-switching cost, when I move from

Claude to GPT in a workflow, does Crewdle abstract that or do I

manage it?

The pay-as-you-go model is what actually sets this apart. I think Thomas's point about a flat subscription pressuring owners to justify the cost every month names a tax most small businesses never articulate. And $20–40 is a believable number, which lands better than a flashy one. One question on the shared context: each component keeps its own memory but can share knowledge. When Forge builds an automation that depends on what Create produced, does the user have to pass that context between crew members manually, or do they find each other's outputs on their own?

How do you handle context between models? Is there a source of truth where I can store all instructions and reference documents/knowledge base?

The "stop juggling five AI subscriptions" framing is relatable. most small teams are paying for separate chat, agent, content, and automation tools that barely talk to each other. honest question though, when you bundle everything into one platform, how do you keep each piece from being a watered down version of the standalone tools? that's usually where all-in-one platforms struggle. the breadth is nice but depth is what keeps people from churning back to point solutions.

Consolidating AI tool access into a single billing and auth layer is genuinely hard. The context isolation problem alone gets messy when users switch tools. We've dealt with similar multiplexing challenges routing between different model providers. How do you handle session state when a user's workflow spans multiple tools with different context windows?

Solid concept. How's the billing handled when one month suddenly spikes compared to others?

The subscription consolidation angle makes sense on paper, but I'd push back on the "one platform for everything" promise a bit. Every all-in-one tool I've used covers ~80% of each individual tool's functionality - and the 20% gap is usually the exact feature the team was relying on. The risk isn't the price, it's the capability ceiling.

What's the honest answer when a user needs something Crewdle's automation layer can't do that Make or Zapier could? Do they export and move on, or are they stuck? The lock-in risk of an all-in-one tends to show up 6 months in, not at signup.

With usage-based billing, is there a way to set a hard monthly spend cap or alerts? For a small business, an autonomous agent running around the clock could get unpredictable, so I'd want guardrails.

Most founders I know are not looking for more AI features. They want fewer tabs open and fewer invoices to track. The positioning here seems much stronger than simply calling it an AI platform.

Interesting! Will test

Interesting launch 🚀

The biggest challenge I've seen with all-in-one AI platforms is context sharing between different tools and agents.

I'm curious: if one agent creates content, another builds an automation, and another updates a website, how do you maintain shared context and prevent conflicting actions across the workspace?

Congrats on the launch!

Congrats! I wonder how does Crewdle orchestrate the handoff between Connect, Forge, and Create?

the usage-based angle is the right call for small businesses. the subscription fatigue is real — most teams pay for 4-5 AI tools and use maybe 20% of each. curious how you handle quality routing though, does the user pick which model runs each task or does Crewdle decide automatically?

Very interesting for a new product, some onboarding and UI work could help make this a gem.

I love the idea. This sounds similar to invideo Agent One for video production, correct?

The MPP angle makes sense - agentic commerce needs a directory layer to work. But I'm skeptical about the chicken-and-egg problem here: the directory's value depends on how many quality services are indexed, and MPP is still very early. Is this pulling exclusively from Stripe Apps + MPP-registered services, or is there a broader listing process? Also curious how it handles the majority of Stripe-integrated tools that exist but haven't opted into MPP yet - those are often the most useful ones an agent would actually need.

Cool idea. How's it handled when a business name is too generic and matches dozens of results?

This sounds very good! Can I integrate my subscription CTA button to a Stripe recurring subscription payment plan in a static HTML page? Can an AI agent do that?

The "agents discover, evaluate AND integrate autonomously" part is what makes this interesting and a little scary at once. Once an agent can act on a listing without a human in the loop, the directory becomes a trust gateway: what stops a malicious or typosquatting service from getting indexed and an agent integrating or paying it before anyone notices? Is there a verification layer on who gets listed, or is the index open and trust is left to the agent?

Stripe Directory — the single place to discover businesses and services across the Stripe network.

Developers and AI agents struggle to find and integrate Stripe services (apps, projects, machine payments APIs). Stripe Directory solves this by letting you search by keyword and get structured, actionable results you or your agent can use.

What’s Different: It’s the only Stripe index combining Stripe Apps, Stripe Projects providers, Machine Payments endpoints, and the broader Stripe business network—all searchable with structured output for agents.

Features:

- CLI-powered search (`stripe directory search`) with compact or JSON output

- Structured results with provider slugs, MPP endpoints, and app listings

- Built-in agent skills to let AI agents discover, evaluate, and integrate services autonomously

Benefits: Instant discovery, faster integration, and autonomous agent workflows without extra instruction.

Who It’s For & Use Cases: Developers building Stripe-integrated apps (e.g., finding a database provider like Neon and integrating in one flow) and AI agents that need to pay for services (e.g., PostalForm for sending mail via machine payments).

P.S. I hunt the latest and greatest launches in tech, SaaS and AI, follow to be notified → @rohanrecommends

I can see this being useful for reducing integration friction. Finding services is rarely the hard part, but understanding whether they fit a workflow usually is. How are quality and trust signals surfaced inside the directory?

Hey Product Hunt!

When we started Clarify, we promised you an autonomous CRM. Today, we make good on that promise with our take on Agents. We’re pumped to launch it today!

They trigger on a schedule or a signal, run across your stack, and work directly with your CRM data.

You describe what you want in plain English. Clarify builds it. You control what each agent can read, write, and do.

We also built a template library so you don't have to start from scratch — pipeline digests, lead enrichment, data hygiene, call coaching. Pick one, customize it, ship it.

This is the feature I've wanted to put in front of users since we started the company.

Curious to see what you build with it!

— Patrick

Love this! Having been using it ever since I heard about it. By far the best CRM in the market today. It's not the biggest. But its by far the smartest and the one that is not bloated with a bunch of graphs just because. It shows you the essential information and works flawless!

Interesting

Congrats on this launch - seems like such a no-brainer to take the M out of CRM and ask AI to do the job!

What was the hardest part of building the agents?

The CRM maintenance tax is something I hear about constantly from smaller B2B operations, where the person closest to the deal is also the one manually updating fields at 10pm on Friday. The idea of agents that run on a schedule or a signal rather than waiting for someone to remember to log something feels like the right direction.

I'm curious how the agents handle situations where the underlying data is patchy or incomplete. Like if a contact hasn't had any activity in 60 days and there's no clear next step on record, does the agent surface that gap and flag it, or does it mostly work with what's already there?

This feels very useful. CRM admin is one of those things that takes way more energy than it should.
What’s the first agent most teams usually set up?

A bit biased because I watched the team build this, but can't emphasize enough how cool these agents are. I used to cry myself to sleep trying to build automations in zapier or workflow builders, and now I can wire those up with a sentence and a couple of minutes. Nice that I don't have to go outside of the crm to do it, too.

The 'relationship agents' framing is interesting — the hard part with any AI layer on a CRM is trusting the underlying data, since reps under-log and records go stale fast. Does the agent actively enrich/verify (pulling from email, calendar, web) to fill the gaps, or does it reason over whatever's already in the CRM? The first is far more useful but also where accuracy gets tricky.

The "set them fully autonomous or require approval before acting" toggle is the part that stands out to me — that's usually where automation tools force an all-or-nothing choice. Is the approval gate configurable per action type (e.g. let enrichment run free but always hold writes that overwrite existing CRM fields), or is it set at the agent level?

Cool product! would like to know the guardrails for agents not completing tasks, or we can control their usage.

The idea of a CRM that updates itself is appealing because manual data entry is everyone's least favorite part of sales. I'm curious, what's the biggest habit founders stop doing once they switch to Clarify?

The autonomous part is exciting, but the failure mode I'd worry about in a CRM specifically is a confidently-wrong write — merging the wrong duplicate, enriching a lead from a stale source, updating a field based on a misread signal. A bad autonomous write is worse than no write because it silently pollutes the source of truth and everyone downstream trusts it. Question for you Patrick: does Clarify ever say "I'm not sure" and hold instead of acting, or does every triggered agent always write something?

I like the idea that the CRM work finally moves into the background.

A lot of the painful part is not managing relationships, but keeping the system clean enough so those relationships don’t get lost. Pipeline digests, lead enrichment, and data hygiene feel like exactly the right things for agents to handle.

Personally, I’d probably start with follow-ups and pipeline summaries first. Curious what teams automate most often after setup: cleanup, enrichment, follow-ups, or reporting?

The interesting bit is not that the CRM can update fields. It is the authority model around each update: which signal triggered it, what data it touched, whether approval was required, and what receipt proves it happened.

How are you exposing that when an agent runs fully autonomously?

We've been looking for something like this at Yoodli for ages. So excited to finally see actual CRM agents be a reality!

Congrats on the launch!!! Do you guys make your agents primarily for support or sales?

Hey Product Hunt, I’m Melissanthi, Senior Product Engineer at Mindstone.

We’re launching Rebel as a Fair Source AI desktop app for agentic work.

The reason we’re doing this now is simple: AI work shouldn’t be trapped inside one closed platform or one model provider. Rebel is built as a desktop workspace that people can download, run locally, inspect, customise, and connect to their own tools through MCP.

Fair Source means the code is available, but with practical restrictions that let us keep building the product sustainably. Small teams and individuals can use and adapt Rebel freely, while larger organisations need a commercial licence.

This release is also part of a broader push toward more portable AI workflows: model choice, local-first files, MCP connectors, and workflows that teams can actually understand and adapt.

We’d love feedback from makers, developers, operators, and AI teams on:
• what you’d want from a Fair Source AI workspace
• how you think about model choice and local-first AI tools
• what would make Rebel useful in your own workflow

where exactly do you draw the line between what a small team can adapt freely vs what needs a commercial licence? Asking because that clarity is usually what makes or breaks adoption for us.

My favourite thing about Rebel is the way the memory works - learns as you go, and shares its memories with other Rebels (e.g. throughout your company).

The interesting part is how it works to judge which memories should be shared and with whom, and which should be kept private. And it asks if it's unsure.

A bit of personal context on why I care about this so much.

I don't use Rebel as a demo environment. I run a large part of my work through it.

Most days, that means speaking rather than typing: meeting prep before important calls, catching the follow-ups I would otherwise delay, pulling together context from email, calendar, Slack, docs and old conversations, drafting things in my voice, and challenging assumptions before something goes to a customer, investor or the team.

The shift for me has not just been "the same work faster", but more that I can now do work at the level I always wanted, but often did not have the time or attention for. Better prep. Better follow-through. More context. Less living out of my head.

This is also why fair source matters to me.

If an agent is going to sit this close to my actual work (memory, tools, approvals, the messy context of their day) trust cannot just be a brand promise. People should be able to inspect it, run it themselves, and understand the boundaries.

Rebel started as my own operating system for work, and I wouldn't be able to do my job without it today. Opening it up feels like the right next step.

The 'asks first' framing is smart — most AI agents just barge ahead and you find out later it did the wrong thing. Is this confirmation step configurable, or always-on by default?

@melissanthi_papacharalampous1 The “ask first” layer is the strongest part for me. Agents become much more useful when they can work freely on low-risk tasks, but pause before actions that touch email, shared spaces, or anything sensitive. That middle ground between locked-down and fully autonomous feels like where real trust starts.

What kinds of actions can Rebel actually execute today, like email and calendar, or internal app automations?

The ask-first default is strong. The piece I’d want durable is the rule change after an approval: original action, future scope, who accepted it, and when it should expire or be reviewed. Otherwise learned permissions can quietly become permanent permissions. Are those rule updates inspectable later?

The approval-before-action model works well for individual workflows but gets complicated fast in team settings. Imagine two people on the same team have given Rebel conflicting learned rules for the same action type. One has approved it, one has flagged it. When a shared automation runs, whose learned policy governs? This seems like it could silently diverge across team members over time, especially as each person's rule set adapts independently. Have you thought about a policy reconciliation layer for shared workflows, or is the current model intentionally keeping rules personal and non-shared?

If I’m understanding this right, the local-first/customisable thing makes sense, especially if the point is not getting stuck inside one AI vendor’s way of working.

One thing I’m wondering though is what happens once this moves from one person using it to a small team using it. If people start adapting their own workflows, connectors and approval rules, do you risk everyone ending up with slightly different ways of working?

Is there a way to share the setups that work well without making it feel like one central locked-down workspace again? Curious how you’re thinking about that.

The planner/worker/background-safety-classifier split is a genuinely interesting routing decision. When the safety classifier and planner disagree on risk - does planner intent override a conservative flag, or does the classifier always win?

"Ask first" is probably the right default. The challenge is that the more context an agent has, the more confident people become in its actions. Have you found a point where users start approving things without really reviewing them? Curious how you're thinking about trust calibration over time.

"For all of it, not one task" is the right framing, most AI tools

optimize for single workflows and break the moment your work spans

multiple systems.

How does Rebel decide what needs approval vs what can run autonomously?

The friction between trust and speed is where most agentic tools either

slow you down or scare you.

Congrats to the whole Mindstone team, this is a serious piece of work! Really like the approval layer being impact-aware, writing to a shared space treated as riskier than a private one is the kind of detail most agent tools never bother with. We've been building an agent that takes real actions and know the gate is the hardest thing to get right, too loose and its dangerous, too tight and nobody uses it. Rebel learns rules from approvals instead of re-asking forever and you clearly ran this on a lot of real work - very impressive. Another great thing is tool-level permissions like an email MCP that can draft but not send - IMO exactly the granularity people actually need. Nicely done!

Tried out Rebel across our internal ops and the approval gating is the part that actually changed how we work — most agent setups force you to choose between "full autonomy and pray" or "so locked down it's useless," and the ask-first middle ground is what finally let us point agents at real workflows instead of toy demos. The fact that it's aware of the original query when deciding what's risky (sending an email is fine if I asked for it, sketchy if I asked for research) is a small detail that removes a ton of approval fatigue in practice. Pairing that with model choice + local-first means we're not re-architecting every time a provider changes pricing or limits. Genuinely useful! thanks

Strong angle, and the maker replies show you've thought about the hard part. One thing I'd add from doing this on customer-facing work: the risk with ask-first isn't only where you draw the line, it's approval fatigue. If the agent interrupts too often, people start rubber-stamping every prompt and the gate quietly stops protecting anything. Two things seem to matter most: making each approval information-rich enough to decide in a couple of seconds (what it's about to do, why, and the source it's acting on), and the auto-updating rules you mentioned so it stops re-asking about things someone has already approved a dozen times. Get those right and ask-first scales; get them wrong and it becomes click-through noise. Curious whether Rebel surfaces the "why" and the source inline at the approval moment. Congrats on the launch.

the approval checks for sensitive actions is something more agent tools should be doing. most platforms either give the agent full access or keep it so restricted it can't do anything useful. that middle ground where the agent works freely on low risk stuff but pauses for human approval on anything sensitive is how you actually get teams to trust agents with real workflows. curious how granular the approval rules are, can you set different thresholds per agent or per action type?

The ask-first posture is important. For small teams, the useful split is usually reversible vs irreversible actions: summarize, draft, search, and organize can move fast; writes to customers, CRM, payments, deploys, or files that other systems trust need a visible checkpoint and a trace.

The 'ask first' principle is something we've had real debates about internally. Autonomous agents that act without confirmation can compound errors in ways that are hard to recover from, especially in stateful workflows. What's your approach to calibrating when the agent should interrupt vs. proceed? Does it use confidence thresholds, or is it more rule-based?

Nice approach. How's conflicting context handled when agents pull from multiple different projects?

Love the Fair Source approach. Curious why did you choose fair source over fully open source for Rebel, and what feedback have you received so far from developers?

Since it's local-first with model choice, can the approval checks for sensitive actions run entirely on a local model, or does the gating logic still require a cloud model call?

Starting with DoorDash is a smart wedge, a bounded low-stakes purchase to prove the buy flow before agents touch anything bigger. For someone running several agents, can I issue a separate card per agent or per workflow so limits and spend are scoped independently, and does each transaction carry back enough metadata to reconcile which agent/run made the purchase? Trying to picture how the accounting maps back when it's not just one Claude ordering dinner.

Where do you get a $15 dollar lunch? The real product is a time machine to get lunch back in the 2000s.

can we get this for grubub?

If agents are going to do real tasks for us, payments become unavoidable. Ordering food is a very understandable first use case because it is concrete, low-friction, and easy to test.

The big question for me is control. How do you handle limits, approvals, refunds, wrong orders, or cases where the agent misunderstands what the user wanted?

As someone who lives in these AI tools all day, ordering food straight from Claude is wild. Congrats on the launch.

My fav agentic payments company and it's not even close!

great product guys! Hope this becomes the default way to shop in the agent economy era.

What are the ways to fund your card? do you guys support x402?

Genuinely curious how you're handling the fraud surface here — when an agent can place an order, what stops a prompt-injection or a spoofed merchant from turning "order dinner" into a drained card? Is there a spend cap or a human-confirm step, or does the user set the guardrails up front?

Single-use cards are a strong first boundary. The next thing I’d want is the order-intent receipt: merchant, amount, delivery address, items, who/what approved it, and the cancel path.

For agent payments, the proof after the purchase matters almost as much as the card cap.

This is super cool! I'd love to get onboarded - how so?

Are there any features coming in the future or product expansions to UberEats etc? Also curious to know how you connect to DoorDash, api/mcp?

Very cool! Any plans for cards provisionable programmatically by agent companies like us (Solid) so that our agents can provision cards on behalf of our users to do things like sign up for Twilio & DigitalOcean?

How does the agent actually make the payment? It uses chrometools to select the product and navigate to checkout?

Ordering food is convienient, yet I would hesitate before letting an agent make higher-value purchases. Curious how you think about building confidence over time.

Cool product! Curious how guardrails will work

Does it detect when a PR is blocked on you or waiting on someone else’s review?

The traffic-light menu bar icon is such a clean way to surface this — green/yellow/red is instantly readable. With up to 5 independent monitors running, how does the single menu bar dot decide its color? Does it reflect the most urgent state across all monitors, or is there a way to prioritize one?

Hey, very useful product. Is this also available for Windows as well ? Are you planning to make this for some of the most popular OS as well ? Thanks

Simple, sharp idea — the kind of tool that solves a real daily annoyance without overengineering it. Built this solo, or with a small team? Always curious how indie devs scope these micro-tools.

the no-server constraint is the real tax — sandboxed means polling, and 'red when stale' needs review state per pr, not just the list. 5 monitors on busy repos and the 5000/hr github ceiling gets close.

Congrats on the launch. Love a tool that makes you feel the problem instead of just listing it in a dashboard. It's so easy to let that stuff slide to the bottom of the to-do list, and a red menu bar is a smart nudge.

Love how this flips the usual "open GitHub to check" workflow into a passive ambient signal you can't ignore -- your menu bar becomes a guilt-free health check for your team's review hygiene.

green→yellow→red in the menu bar is the right kind of ambient signal, glanceable beats another notification you'll dismiss. does the color ease in or just snap?

The menu bar warning is a simple but useful idea. PRs often go stale quietly, so making review debt visible before it piles up could help teams keep momentum.

Nice launch 🚀

This solves a small but very real developer pain. Missed PR reviews can quietly block a team for days.

I'm curious: can StaleMate prioritize PRs by age or urgency, for example PRs waiting on my review vs PRs where I'm the author and need to respond?

That kind of separation would be super useful for developers juggling multiple projects.

This actually solves a real annoying gap with PRs slipping through. Would be good to know how quickly it reflects updates across different providers. Congrats on the launch

The invisible pill UI is the right call, the moment it becomes visible to others it turns into a crutch and kills authenticity. What I'm curious about: does Ruby learn from the calls that went well? Like, if I closed a deal after asking a specific sequence of questions, does it start surfacing similar patterns in future calls?

One thing worth flagging for anyone getting an error on first launch 👇

You'll need Claude Code installed in your terminal for Ruby to work, not just the Claude desktop app. They're two separate things, and the app alone won't connect.

If you've hit an error during your first prep chat, that's almost always why. Install Claude Code in your terminal and you'll be set.

Happy to help if anyone's still stuck 🙌

Ruby looks like the kind of tool that’s useful because it stays simple. I like products that solve one clear problem without making you learn a whole new workflow first.

Does Ruby run its real-time transcription and suggestions locally, or is call audio sent to the cloud? Curious how that works for calls under NDA or with privacy-sensitive interviewees.

Cool idea to be more productive, love the ui. Great Reimagination of Ruby programming language branding :D congrats on the launch guys

Running customer discovery calls as a first-time CTO, I know that post-call feeling of "I should have dug deeper on that." The in-call nudge approach is smart - it's the difference between coaching before the game vs. yelling from the sideline after. Curious how Ruby handles long tangents - does she still queue the question or re-adapt context? Either way, trying this on my next user interview.

The interesting part here is the narrowness of the wedge. There are already plenty of tools trying to capture everything after the call, though staying present while still noticing the thread you should probably pull on is sometimes the hardest part.

Curious how Ruby handles that in real time. If the conversation goes off-script but starts revealing something useful, does it follow that thread and suggest deeper questions, or does it mostly anchor back to the objective you set before the call?

This is such a good idea. As a neurodivergent person, I always remember the important question after the call is over. A small real-time nudge would help a lot.
How subtle are the prompts during the conversation?

Would love a tutorial on how to get started!

Hi everyone!

Swipe typing is one of those features that seems simple, but is actually incredibly hard to get right without relying on closed keyboard systems.

FUTO built a three-model system to solve this: a 635K-parameter layout-agnostic encoder, a 300K-parameter QWERTY decoder, and a 1.5M context language model. Because the footprint is tiny, the whole thing can run locally in milliseconds.

Releasing the weights and the C++ beam search library means people can now build proper swipe typing for VR, laptops, or alternative mobile platforms without relying on closed-source blobs.

The best way to experience it today is FUTO Keyboard.

Releasing both the models and the 1M-swipe dataset is what stands out here. Beyond keyboards, have you seen developers experimenting with FUTO Swipe for non-traditional input systems like VR, AR, smart TVs, or accessibility-focused interfaces? Curious which unexpected use cases have emerged so far.

privacy-first ux only works when the core interaction feels genuinely good, and this is a rare case where it does. keeping the typing path local while still shipping something polished is the right product bet, especially from a team that keeps backing this space.

The split between a layout-agnostic encoder and a layout-specific decoder is the interesting call here, since it implies adding something like AZERTY or a custom split layout only needs a new ~300K decoder rather than retraining the whole stack. Is that actually the case, can I train just the decoder against the released 1M swipe dataset while reusing the encoder, or does the context LM need retuning per layout too? Also curious what the license allows for embedding the C++ runtime into a third-party keyboard.

I've been using swiftkey after trying out a bunch of different swipe-y keyboards a couple years ago. Any plans to launch this on the app store? I'd love to try it out!

shipping open on-device models AND releasing the 1M swipe dataset is the rare open-source move that actually moves the field. respect.

Well done, this is a really worthwhile project, thanks for sharing. There seems to me to be a lot more scope for open-standard solutions being picked up and baked into devices, and maybe even built upon as a store, like Chrome's plugins. That would be a big unlock across devices.

Interesting launch 🚀

I like that this runs fully on-device instead of depending on closed keyboard systems or cloud inference.

Curious how well the model handles different typing styles over time. Is there any personalization layer planned, where the keyboard can adapt locally to a user's common words, slang, or swipe patterns without sending data off-device?

Open models for keyboard input feels like an important direction.

The 'do the work, not another dashboard' framing is the right wedge, since most GEO tools stop at showing you how invisible you are. Living in Slack, how much actually executes end to end: does Nimt ship the content and page changes itself (CMS writes, PRs) or hand back drafts for a human to push? And where does the state of what's been tried per brand live, inside Slack threads or a separate workspace you set up on first run?

Living in Slack instead of a separate dashboard removes the biggest barrier to actually using an SEO tool daily, most teams open the dashboard once a week and forget about it. The multi-model tracking is what caught my eye though: are you aggregating ranking data across Google, Bing and AI search like Perplexity in the same dashboard, or is it primarily Google-focused? That distinction matters a lot in 2026.

Congrats on the launch. I like the move from passive AI-search dashboards to actually doing the work. Curious how you measure whether suggested content changes improve visibility without hurting content quality or brand trust over time?

Hey 👋 Product Hunt!

Pourav here, PM at CometChat.

We kept hearing the same thing from React teams: the chat backend is the easy part. Then they hit the UI. Message lists that scroll the right way, typing indicators, read receipts, threads, retries, real-time listeners that don't leak on unmount — that's weeks of frontend work that has nothing to do with their actual product. And every team rebuilds it from scratch.
So we rebuilt our React UI Kit V7 around one idea: every component handles its own data, its own real-time listeners, and its own state. You drop it into your layout. It works.
A few highlights:

• Compound components — compose only what you need (Root, View, EmptyState, and friends). No monolithic widget to fight.

• Plugin system — custom message rendering through one interface, so your polls, payments, or whatever render the way you want.

• Theming with CSS variables — light and dark out of the box, every token overridable.

• SSR-safe — no browser APIs at module scope, so the Next.js App Router doesn't fight you.

• 19 languages built in, plus optional voice and video when you need them.

There's a live CodeSandbox in the docs if you want to fork it and poke at it.
Try it out for yourselves and let us know what you think. I'm in the comments all day.

Pourav

Chat UI is one of those things that looks simple until you start dealing with threads, retries, read receipts, and all the little edge cases. I like that this gives teams a cleaner starting point instead of making everyone rebuild the same pieces again.

How do you handle presence and reconnection, like does it speedily recover when someone’s network drops?

V7's compound component model sounds cleaner - each component owning its own state and real-time listeners is a genuine improvement over the monolithic widget approach. The honest question: when every component manages "its own data," that data is wired to CometChat's backend. What does migration look like if a team later wants to swap to Stream Chat, Sendbird, or a self-hosted solution? "None of the complexity" is only true if the CometChat lock-in is a tradeoff teams understand clearly upfront.

The Digital Crown navigation without killing Water Lock is the detail that tells me you actually swim! Every watch swim app I've tried makes you fight the wet-screen lock mid-set and it wrecks the session:) Well done! And one question. On the AI side: a pool only hands you splits and times, so how does the model know I blew up on a set versus cruised it?

Swimming is the most underserved sport in fitness tech - every running app has AI coaching now but swimmers get basic lap counting. The Water Lock detail stands out: the mid-set scramble to unlock your watch kills the flow, and the Digital Crown navigation fix shows you actually swim. Curious whether the AI workout generation adapts to pool length - 25m vs 50m changes pace targets and rest intervals meaningfully. Nice work filling a real gap here.

As someone who swims 3-4x a week the water lock thing is exactly the pain - every other app i've used i end up jabbing at a wet screen mid-set and losing count, so being able to scroll metrics with the crown is honestly the feature that'd make me switch. and tracking swolf + stroke efficiency instead of just laps is what actually tells you if you're improving or just tired. clearly built by someone who's spent real time in the pool. nice one.

I was curios if you think eventually this could work with Garmin as well?

Congrats on the launch! Swimio looks really polished. I like that it goes beyond basic lap counting and brings workout planning.

Curious if Garmin support is on the roadmap as well?

Congratulations on the launch! 🚀

I like that Swimio focuses on actually helping swimmers improve rather than just tracking laps and distance.

I'm curious: have you considered using trends in pace consistency, heart rate, and recovery between sets to automatically adjust future workouts? That seems like it could make the coaching experience even more personalized over time.

The Apple Watch experience looks especially well thought out.

The Apple Watch detail matters. In a pool, the interface has to respect attention and water lock instead of pretending the user can babysit a screen. I would keep testing the handoff between generated plan, live watch guidance, and post-session correction; that is where coaching apps tend to become either useful or noisy.

The Digital Crown to move between metrics without unlocking Water Lock is such a smart detail — anyone who's actually tried to tap a wet touchscreen mid-set knows how much that matters. Tracking SWOLF and stroke efficiency rather than just laps shows this was built by a swimmer. Out of curiosity, does the Apple Watch workout support set-based structures (e.g. 8x100 with rest intervals), or is it continuous-distance focused?

This feels like one of those products that was clearly built by someone who actually swims. The Apple Watch experience sounds especially thoughtful because small usability details matter a lot when you're in the water.

Hi! Looks interesting!

Could you explain more about how AI adapts or creates a workout? Based on what parameters? and using any specific model?

The "pull people who engaged with this LinkedIn post and dedupe against my contacts" use case is the one I'd actually build a workflow around. Most of our outbound coordination already runs through Slack, so keeping the agent there removes the context-switch tax completely.

Two questions: how does ICP filtering work before building the list - do you configure that upfront or is it prompt-level? And does it connect to existing CRMs like HubSpot so the dedup is against real pipeline data, not just a local contact list?

👋 Hey Product Hunt!

Quick question: how much time did your last presentation take to create?

For many people, the workflow usually looks like this:

Before:

• Open a blank presentation
• Figure out the structure
• Write content slide by slide
• Search for visuals
• Rearrange everything several times
• Spend 2–4 hours getting it presentation-ready

After, with Premast AI:

• Paste a topic, document, or URL
• Get a complete presentation draft in minutes
• Refine and customize instead of starting from scratch

That’s the problem we wanted to solve.

We kept seeing people, including ourselves, spend more time building slides than communicating their ideas. The content already existed, but turning it into a well-structured presentation was still a surprisingly manual process.

So we built Premast AI Presentation Generator.

Give it:

✅ A prompt
✅ A document
✅ A URL

And it generates:

• A structured presentation outline
• Slide-by-slide content
• Visual recommendations
• A draft that’s ready to edit and present

Our goal isn’t to replace creativity. It’s to eliminate the repetitive work that happens before the real polishing begins.

I’d love your feedback on one specific thing:

👉 If you could automate just one part of creating presentations today, what would it be?

Structuring the deck, writing slide content, finding visuals, or something else?

Looking forward to hearing your thoughts and answering questions throughout the day 🚀

This is a really great idea and I think it will make creating presentations much easier and faster for a lot of people.

I have one question: if I'm working for a company and already have established brand guidelines, such as specific colors, fonts, and overall visual style, can I upload or define those guidelines so that the generated slides automatically follow my brand identity?

We built AI Presentation to turn ideas into decks in minutes.

What's your biggest presentation pain point today? 🤔

Love the experience working with you @momen_elshamy @reem_kamal Good luck!

Wow Momen! It's impressive. Presentation made easy than ever is super cool and I'm sure many of us will take the most of it. Wish you all the best!

Congratulations on the launch 🤩

As a designer, platforms that simplify the creative process especially for presentations are always appreciated.

Wishing the team all the best for this exciting new chapter and looking forward to seeing the impact you’ll make. 🚀❤️

How is this different from Gamma?

Congratulations on the launch! 🎉

As someone who works with presentations regularly, I know how much time can be spent searching for the right templates, icons, and visual assets. Premast does a great job of bringing everything together in one place and making the presentation design process much more efficient.

Wishing the team a successful launch and lots of growth ahead! 🚀

For presentations with lots of data can it pull in live charts or tables from something like Google Sheets or it purely text and static visuals for now.

Hey PH 👋 Maxime here, co-founder of Well.

You know how Obsidian turns scattered notes into one connected graph you can actually navigate? We built that for your whole business.

It's not only about mapping the data altogether. We created an ontology that stores in a unified data model your payments, invoices, customers and 55+ entities, kept in one living model your agents can actually reach.

We went full into MCPs, connecting the tools that we use on the daily basis, but never gets connected under one roof: productivity tools like Notion, Slack or Gmail; 11K banks feed connectable like Mercury, Ramp or Qonto, Accounting tools, CRM, Ecommerce... and many more.
When a MCP is available, it's automatically part of our collection (thanks for that!). If not, we leverage public API documentation and generate a MCP out of it, so that we never miss a tool.

We build what we called a "Business context graph".
-> The more you connect tools to your second brain, the smarter it is.
Today you reach it over MCP and CLI.

The smart part? When you install a MCP with Well, you don't need anymore to maintain it in any of your Claude or Codex session. We maintain it, and things get simpler for you, 1 MCP -> All MCPs, no broken auth. One MCP to rule them all.

More upcoming... This context unlocks real work: close your books in under 10 min, or just ask a question and we draw the chart that shows where you stand. CRM, tax and payments are next, brace yourself. But shh...🤫

How to test it?
-> For the builders: one MCP endpoint, api.wellapp.ai/v1/mcp, point Claude, Codex, or Cursor at it and your agents reach your whole business and every connected tool. One MCP to rule them all, that's the motto, you got it now!

🎁 PH offer:
We made several launches prior to this one, we want to reward the community that helped us in the early days.
3 months free for lower plan, or unlimited tokens on a paid plan. Both time-limited, code in the link:
-> https://wellapp.ai/?utm_source=producthunt&utm_medium=referral&utm_campaign=context_graph_launch

One question for you: which tool would you connect first, and what's the first thing you'd ask your business?

Make your business legible. Build your second brain right now!

Massive congratulations on the launch, @maxime_champoux1keeping all mcp tool authentication alive across different claude and cursor sessions is an absolute nightmare, so having one single unified endpoint completely changes the game. let's gooooo👏

Really enjoyed working on this one 🙌 .
Worked on the layer under the chat: reconciliation, enrichment and the data model that builds the graph.

Reconciliation matches what's actually what across your tools; enrichment pulls in everything we can find on each entity and how they connect. That's what turns scattered records into one connected business: the chat can actually reason over them, a lot of LLM work to make it hold at scale.

Curious which integrations you'd connect first and which would be most useful to you. 👀

Looking at it from an Ops/Growth perspective, I'd probably connect my bank and CRM first, just so I could quickly see which closed-won deals actually turned into cash, and which "paying" customers still owe us.

First thing I'd ask Well: show me this month's pipeline against what actually hit the account.

That used to mean a bunch of tools wired together, messy exports, and too much time spent building reports.

Just opened my new company entity and it looks like just the right tool to handle all my scattered invoices, accounting tools and company expenses, from a centralized plateform with just a chat.
That's honestly a game breaker to spend more time building and loose less time on what doesn't provide value to my business

Not their first launch on PH but the shipping velocity is something else. I love the Obsidian analogy applied to finance data that connected graph mental model finally makes sense for payments, invoices, and customers instead of dev notes. One MCP to rule them all is a bold promise but if it holds, that's a genuine time saver for finance teams. Congrats @maxime_champoux1 and the Well team.

Congrats on the launch!

You used to be the glue between a dozen apps that don't talk. Every tool is now addressable and agents reach across all of it for you. Same data, none of the manual stitching, the second brain we've all been asking for.
Really proud of what the team pulled off here.

Looks incredible! Can't wait to try it out

This team keeps dropping bombs! Love the attention to detail!

Genuinely think this solves a real problem: every new tool you add is one more silo. Getting bank, accounting, CRM and email into one context graph the agents can reach on demand is a serious time saver for finance/ops teams.
Proud of what the team shipped !

Honestly the thing that got me is that it's one graph instead of another integration to keep alive. our finance data is spread across qonto, pennylane, stripe and the crm and pulling it was never the hard part - it's getting it into something an agent can actually reason over without making stuff up. having all the entities in one model the AI can reach is what makes "just ask" give you a real answer instead of a guess. keen to see the crm and tax pieces drop but the base is right.

Hi Well team, thanks a lot !! Every other connector setup I've run ends the same way: an auth token silently expires mid-week and half my Claude/Codex sessions start failing without telling me why. Offloading that maintenance so the context layer stays alive across sessions is the actual unlock here, not just "another connector." The Business Context Graph framing makes sense too,  a unified model beats stitching together fragments from Stripe, HubSpot and Notion that each only know a slice. Solid execution

The signal/noise framing is the right call — most agentic-coding content right now is either token-selling hype or AI-generated filler dressed up as a tutorial. How do you actually filter for 'human-written' each week: is it you manually reading through the firehose, or do you have heuristics/tooling to pre-screen candidates? And is there a public archive or RSS so I can pull past issues into my own reading workflow, or is it email-only for now?

Happy to find this project.

The amount of irrelevant and destructive noise is annoying. I am quite tired of reading "you're cooked", "you're replaceable", "one prompt will change everything," blah blah.

Being a software engineer, this is a fascinating tool that we all need to learn meticulously, see problems, disadvantages, weak points and learn how to trick them and make our solutions better.

Another problem I have is that it's annoying to read an AI-generated article. I've unsubscribed from a few tech bloggers I've been reading so far because of the AI-slop smell in their text.

I like that this project combines both - top-picked relevant articles for agentic coding and double-verified that those articles have a human touch and human deep thinking, not like "make an article about agentic coding with 1000 words, make no mistakes".

Agree with the problem, it's almost as depressing as the sheer amount of AI-written comments on PH 😂
Will be one of your readers!

Hey all.

I don't think I've ever seen this amount of noise to signal content ratio for any other topic as I see for agentic coding. The amount of slop (sponsored slop, hype/doom rage-baiting, ai-generated slop) is unbelievable.

My bet is that figuring out agentic coding workflows will be a very long, unsexy process done by devs who actually use agents in real-world scenarios and document their findings, not by corporations whose incentive is to sell more tokens to companies.

I spend some time each week sifting through mountains of garbage to find original human-written articles about real-world applications of agents. I decided to make a newsletter out of it. Hope you find it helpful.

The amount of agentic coding content has gotten noisy fast, so a human-curated filter actually sounds useful. I’d rather read a few real workflow notes than another recycled “AI will change coding” post.

The noise problem in agentic coding content is real - half of what circulates is vendor marketing dressed as tutorials or pure hype. We spend actual team hours triaging what's worth reading vs what's just token farming. The human-written constraint is exactly the right filter. Subscribed.

I've seen agencies absolutely pull their teeth while working with shopify stores to fix product images. You've actually pulled up a core issue most of these stores face - photoshoots take a lot of planning, time and of course, money.

A friend runs a clothing brand and wanted to repurpose a photoshoot - she already had images of models wearing her clothes and she just needed more variety. I messed with google flow by adding different backgrounds and get the model in different poses but it took a lot of time to get something usable and believable.

In cases like this, can we upload our own model images and generate new backgrounds and poses for them? If yes - I need to send this to her!

All the best for the launch @rahul_bhargava1 & team!

This'll be extremely helpful for my Shopify store. Quick question, can i create my own configurations and use it in the future?

Congrats on the launch! 🎉 The focus on maintaining catalog consistency at scale is really interesting.
How does Crop.photo handle products with very different aspect ratios while still keeping the catalog visually aligned?

Just checked your website and tried the demo using that "Try a Shopify Recipe in Seconds" block. And honestly, I'm impressed.
This has everything that I was looking for.
Image resizing, On model images, video generation for my products.. everything at a single place.

Congrats Rahul and team! We recently worked with multiple Shopify brand on abandoned cart recovery use cases and got to the ops heavy nature of maintaining product photos. This seems super useful.

This solves a real cost problem for small brands honestly, reshoots every season add up fast. Being able to generate on-model and lifestyle shots from a single photo is a smart move. How different does the output look if you try multiple poses/scenes from the same source photo, does it start looking repetitive after a point?

Congratulations on the launch! 🎉

The Shopify use case really resonates, image preparation is one of those tasks that's incredibly important but rarely gets the attention it deserves until it becomes a bottleneck.

I'm curious about AI Match: when a brand has a very specific visual style (for example, luxury products with precise spacing, shadows, and composition rules), how does the system learn and maintain that consistency across thousands of images?

I've tried a couple of background removal tools before and the results were always a bit "off" around hair and fabric edges. What makes the output here cleaner than what's already out there? Genuinely asking because if it actually works on complex product shots I'm signing up today.

Interesting product. We've reached a point where generating images is becoming easy, but keeping an entire Shopify catalog consistent still feels like a full-time job.

Have you seen any common numbers around how much time merchants save once they start using AI recipes instead of managing product images manually?

Looks especially useful for Shopify stores with large catalogs. Bulk cropping and resizing could save a lot of time when preparing PDPs, ads, and social creatives. Does crop.photo also handle different marketplace image ratios automatically?

I have a lot of bag images that needs to be resized for marketplace specs with proper margins. Can that be handled with Crop.photo?

Been a power user of crop photo, totally content with the value for money, it does what it promises, which is still a rare occurrence with today's softwares suites

Bulk editing product photos is one of those tasks that sounds simple until you're doing it for 500 SKUs at once. The AI Recipes angle is smart, reusable presets mean you build the workflow once and apply it forever. Does it handle transparent backgrounds for marketplace listings like Amazon or Etsy, or is it primarily Shopify-focused?

Congrats on the launch! @rahul_bhargava1
Quick question: does the model support multiple input images for fashion products?

For example, can I upload several photos from the same shoot and use them as references to generate new poses, angles, or backgrounds while keeping the garment and styling consistent?

Congrats! Looks good

👋 Hey Product Hunt fam!

I’m Jennifer, and I’m excited to share what we’ve been building.

🌐 What’s TaskFord?

TaskFord is an integrated work delivery platform that helps teams plan, manage, and actually deliver work, all in one place.

Instead of juggling multiple tools for tasks, resources, time, cost, updates, and collaboration, TaskFord brings everything into a single, connected workflow so teams can stay aligned and move faster.

💡 Why we built it?

Most tools today are great at managing tasks, but not at delivering outcomes. Work gets split across different tools, context gets lost between updates, and teams end up spending more time coordinating than actually shipping.

The problem isn’t a lack of tools; it’s the lack of connection between them.

We built TaskFord to change that. By bringing planning, execution, and delivery into one integrated flow, teams don’t just track work; they actually move it forward and get it done.

🚀 Why try TaskFord?

⚡ Integrated workflow: no more tool-hopping to get work done

🔍 Real visibility: always know what’s happening without chasing updates

🤝 Connected work:  tasks, projects, resources, time, and collaboration stay in sync

🧠 Built for delivery: not just managing tasks, but helping teams actually finish and ship work

🙌 Join us
We’re still early, and we’re building this with the community.

I’d love to hear:

• What’s the hardest part of delivering work in your team?

• Where does your current workflow usually break?

Start free on the site, or reach out anytime: 📨 support@taskford.com

Feel free to try it out and share your thoughts, I’ll be here all day to chat and learn from you 💙

The gap you describe — between tracking tasks and actually delivering work — is real. Most tools solve the visibility problem but not the momentum problem: you can see everything clearly and still feel stuck. Curious how TaskFord handles tasks that drift between projects, the ones that don’t clearly belong anywhere but keep moving?

Massive congratulations on the launch @jennifer147 and team being the manual human integration between three disconnected task and resourcing apps is an absolute nightmare. definitely signing up for a free sandbox run this afternoon.🙌

Really nice work. I'd love to see some AI-driven scheduling capabilities in the future. Is that something you're exploring?

TaskFord team member here 👋

Lately, I’ve been feeling this more in daily work: finishing tasks isn’t the hard part — getting work all the way to done is.

There’s always something in between. A missing update, a quick discussion somewhere, or just not being fully sure who’s owning what.

Working on TaskFord made me think a lot about that gap. Not just how to manage tasks, but how to make it easier for work to actually move forward without getting stuck in the middle.

Curious if others feel the same in their team.

Exactly what I need right now. I am using Slack, Notion, Google Drive, iMessage, well as my company's internal tools.

Sometimes it's really hard to direct teammates, and the workflow is a bit chaotic. centralization works so much better than disconnection.

Congrats on the launch! Any easy migration steps from tools like Linear?

Hey everyone 👋 Annie here

From a marketing side, I’ve always felt the hardest part isn’t planning, it’s what happens after things start moving.

Plans change, priorities shift, small updates keep coming in… and you end up spending a lot of time just trying to stay in sync.

Working on TaskFord made me think more about how teams deal with that day-to-day reality, not just the ideal workflow.

Curious how others handle this in their teams.

Hey PH 👋

Feels like most teams already have enough tools — the harder part is just keeping everything in sync.

That’s something we’ve been thinking a lot about.

Hey PH 👋 Teresa here from the team

One thing I’ve noticed is that workflows rarely “break” suddenly, they just slowly get messy.

At first everything feels fine. Then over time, tasks, discussions, and decisions start living in different places, and it becomes harder to keep track of what’s actually going on.

Being part of building TaskFord made me pay more attention to how to keep things connected as work grows and changes.

Would love to hear if this is something others run into, too.

I’ve been searching for something like this for so long!!

Big congrats 🙌 Taskford seems refreshingly straightforward, I’m curious to see how it fits into my routine.

Looks like a simpler way to manage tasks and balance workload. Gave it a quick try and it feels pretty smooth. Congrats on the launch!!!

I like how simple this looks.

A lot of tools feel heavy right from the start

The "tool sprawl" problem is real, but I'm genuinely curious what makes TaskFord stickier than the 10+ tools already fighting this same battle - Linear, ClickUp, Notion, Monday, Asana. Every one of them launched with the same "one workspace to rule them all" pitch. What's the thing you've seen users actually switch for that they couldn't get elsewhere?

Hey Product Hunt 👋

I’m Stellar from Liner. Today we’re launching Liner Developer Platform, search agents for AI products, plus lower-level tools for teams that want to build their own workflows.

Most AI apps need more than model memory. They need to search fresh sources, cite evidence, research complex questions, and sometimes explain the answer visually.

Liner gives developers two ways to build that:

Agents
🔎 Search Agent — searched answers with citations and source evidence
⚡ Quick Answer Agent — fast, lightweight cited answers
📄 Deep Research Agent — multi-step research reports
📊 Visual Answer Agent — answers with interactive visuals

Tools
🌐 Web Search — structured web results
📚 Academic Search — scholar results and paper metadata
🧩 Visualization — turn queries into embeddable HTML visuals

We’d love feedback from people building AI search apps, copilots, agents, RAG workflows, and research tools.

Question for builders: would you rather start with a ready-made search agent, or use lower-level search tools and build the answer flow yourself?

The fastest & cheapest search API I've ever used. Amazing

Been playing with it for a bit. Solid value for the price 👍

This api is amazing. I’m using this for my personal project, and I really appreciate for this great product! Thank you liner team!

Search costs are a real line item when you're running agents at scale - the OpenAI web search pricing adds up fast once you get past toy usage. $1/1K requests with cited answers is the kind of pricing that actually makes grounded AI features worth shipping. Curious how latency compares on the academic search vs the web search endpoints.