Hey everyone,
We won Product of the Day. thank you so much to everyone who supported, upvoted, commented, tested Goldfish, and gave us feedback
Goldfish goes live on Product Hunt in 15 minutes
We d love your support when the launch opens: upvotes, comments, questions, feedback, bug reports, anything.
Goldfish is the first AI tool I’ve used that feels like it actually understands the work around the text field, not just the words inside it.
The best part is the cross-app memory. I can be in a Product Hunt review, a WhatsApp chat, a calendar tab, or a SaaS dashboard, and it can still help me write with the right context without making me copy, paste, and explain everything again.
It is especially useful for drafting replies, tightening messy thoughts, summarizing what I was just looking at, and recalling details from recent work. The fact that it works from any text field makes it feel lightweight and natural instead of like another chatbot I need to manage.
The interface is simple, the workflow is fast, and the privacy-first approach makes the whole thing much easier to trust.
The main improvement I’d like to see is more visibility and control around memory. Goldfish is already strong at picking up context, but clearer ways to see what it used, edit what it remembered, and exclude certain apps or situations would make it even easier to trust.
I’d also love deeper integrations and more granular preferences over time, so the experience feels even more reliable across every app without needing much manual correction.
Goldfish has become part of my daily workflow because it removes the most annoying part of using AI at work: explaining the context every time.
It understands what I’m working on across apps, then helps me draft replies, rewrite text, summarize messy threads, and recall details without copying things back and forth. The cross-app context is the killer feature. It feels less like another chatbot and more like a memory layer that sits underneath my work.
Also love that it is local-first and privacy-focused. Huge fan of where this is going 🐠
The main thing I’d love to see is even tighter control over what gets remembered and used in different contexts. The product is already very useful, but more granular memory settings, clearer visibility into what context it picked up, and easier ways to correct it when it gets something wrong would make it even stronger.
Also excited to see more integrations over time. The core workflow already works, now it’s about making it feel even more invisible and reliable across every app.
I still use Claude, but Goldfish solves a different problem. I chose it because it already knows the work context across my apps, so I don’t have to paste threads, explain relationships, or rebuild the whole backstory every time I need help writing or remembering something.
For me it is less of a chatbot and more of a private memory layer for work. That is what makes it stick.
It really is an extraordinary product, I can write replies, clean up drafts, and pull in the right context with my wording/language. It is also safe to use since it is not sending the information somewhere.
I miss having it on my phone too. A mobile version would make the workflow even more useful when switching between desktop and quick replies on the go.
I still use other AI tools, but Goldfish solves a different problem for me. It already has the context from what I’m working on, so I don’t need to copy, paste, put in a skill or objective, explain the thread, or rebuild the background every time.
For quick replies, same conversations, and remembering recent work, that makes it much faster and convenient for me.
Hey Product Hunt 👋 Joel here, one of the two Swedish founders building Goldfish.
Why we built it:
We started building Goldfish because AI still has one weird problem: it knows the internet, but not the work sitting right in front of you. AI has memory like a goldfish!
Your messages, docs, tabs, meetings, half-written drafts, people, decisions, and loose threads already live on your computer. But every time you open a chatbot, you start from zero again. You paste the thread, explain the project, describe the person, and try to make it sound like you.
That felt backwards to us.
What Goldfish is:
Goldfish is a private AI memory layer for your Mac. It understands what you’re working on across apps, then lets you use that context anywhere you type.
Press ⌥ Option in any text field and Goldfish can help you:
write replies in your own tone, with the thread and relationship context already there
summarize messy work from Slack, Gmail, docs, tabs, or meetings
recall things you saw, wrote, read, or discussed
rewrite highlighted text without touching the rest
bring your full computer context into Claude Desktop through our local MCP server
What we believe:
The best AI product won’t be another empty chat box.
It will be the memory layer underneath your work. Something that understands your context privately, shows up where you already are, and removes the need to re-explain yourself 50 times a day.
Early traction:
We’ve worked closely with 1,000+ founding members to build a product people can’t live without. GTM people at ElevenLabs, Deel, and Vercel are already using Goldfish to write faster replies in their own tone.
Our founding team comes from Strawberry Browser (General Catalyst) and Depict (YC20), and we’re currently building from The Bridge by Entrepreneur First in San Francisco.
Privacy:
Privacy is core to the product. Your memory stays on your Mac in a local database. No cloud sync, no backend where we can browse your data, and you can stop capture of apps and domains any time. AI calls have zero data retention.
Product Hunt offer:
Goldfish is still in closed beta, but we’re opening early access to the Product Hunt community first.
As a thank you, everyone from Product Hunt gets 3 months of free access!!
Download now at goldfish.sh
We’d love feedback on:
where it feels magical
where it gets context wrong
what more you’d want goldfish to do in the future
Thanks for checking us out 🐠
@joel_edholm hi Joel, congratulations on the launch. I am trying to and it looks very smooth. First, one of the best on-boarding I have seen. Very very cool.
@antoinetoussaint thank you Antoine, really appreciate it! means a lot that the onboarding felt smooth, we spent a lot of time on making that first minute feel obvious 🐠 What use case will stick for you do you think?
@joel_edholm Congrats on the launch!! I love this idea and have often yelled at my AI, "Why can't you just remember?!" Sadly, I'm not a cool Mac user - will you be bring out a PC version?
@joel_edholm @anna_ludwinowski Anna, please let me know what works well and not in the windows version! much appreciated 🙏
@joel_edholm congrats on the launch! Tried Goldfish while writing this comment and it immediately clicked. Not having to paste context or explain what I’m replying to feels like the right way AI should work. Great job.
@rianbrob thanks Rian, that means a lot. this was exactly the moment we wanted Goldfish to make feel obvious 🐠
@joel_edholm @benln I was born without hands, and although I use a standard keyboard and mouse I type pretty slow so I use Voice Control a lot to type. I would love to talk to you more about Goldfish and also get your thoughts on integrating a Voice Control trigger. I use the option key for a lot so a Voice Control trigger would be helpful for me personally. If you are interested in chatting email me ryan [at] lookmomnohands [dot] com
@joel_edholm @benln @hudsonperalta Hey Ryan! Super interesting, would love to discuss more. Is voice control trigger instead of a keyboard shortcut? Will send you a mail! 🙏
This is one of the more interesting AI products I’ve seen here recently.
The part that resonates most is the “starting from zero every time” problem. As a founder, I constantly switch between product decisions, emails, support, launch copy, team messages, and random notes. Most AI tools are useful, but only after I spend time explaining the context again. 😅
The privacy approach is also important here. A product like this only works if users can actually trust it with messy, unfinished, sometimes sensitive work! :)
Curious how much control users have over what Goldfish remembers or ignores. Can I exclude specific apps, websites, projects, or time periods from memory?
@andrasczeizel You can control it pretty tightly. You can pause capture whenever you want, and exclude specific apps or domains so they never get remembered.
The memory lives locally on your Mac, and you can delete stored history too. We built it assuming people will have messy, sensitive work on screen, so “what should Goldfish ignore?” is a core part of the product, not an afterthought! :-)
Our real power users are mostly founders, so I would love for you to try it! lmk if you have any feedback! :)
@andrasczeizel yes, exactly. You can exclude specific apps and websites from memory, and pause capture whenever you want. Is app/site level enough for control, or would you want something more granular? We are working on a way to dynamically distinct between "personal" and "work" context
Loving this! The onboarding was super smooth, your video walk through demo was really slick. I like that it just reads what is on my screen and knows the context of what I'm doing in order to make the auto-typed replies extremely relevant. I also like your commitment to privacy, keeping data local as possible, and not retaining data sent to gpt. Great work! I'm excited to see where this goes.
p.s. Making LinkedIn part of your onboarding demo so it could scrape info about me from my profile to train on initially... I saw what you did there ;-)
@chad_fullerton thanks Chad, really appreciate it! Glad the onboarding and demo landed.
And haha yes, LinkedIn was very intentional there. The value is most apparent there, I hope you find more usable surfaces also! :- D
@chad_fullerton n thank you Chad! the privacy piece matters a lot to us because Goldfish gets useful exactly where normal tools get creepy if they are not designed carefully. and haha, guilty on the LinkedIn onboarding, had to make the demo do some real work 🐠
The "knows your work" part is where these tools either actually deliver or fall apart.
Curious what that means concretely here: are you indexing local files, pulling from open apps, reading browser tabs, or something else? And how does it handle context switching across very different types of work, like if I'm mid-deep-work on a coding problem and then get a Slack message about a completely separate client project, does it actually pick the right context or does it need a nudge?
@fberrez1 Hey Florent! Yeah so we're using the built in accessibility API:s to get the text content currently visible on the screen, and that is enough to get enough context about what you're working on. We store that on a local database and in a vector database. It tracks the sort of work you do across different surfaces and remembers what belongs to where, so that it adapts to your tone in different projects.
Feel free to try it out and let me know what you love and what you hate! It would be super useful to get some feedback! :)
@fberrez1 yes, fair question. Goldfish captures the foreground app/window context you are actually working in, then ranks recent activity, the current thread, and longer-term patterns separately depending on what you ask it to do.
So in your Slack example, the Slack thread wins because it is the focused surface. If you ask about the coding problem, the recent coding context wins. When the task is ambiguous, it should ask or stay general rather than confidently grabbing the wrong thing.
The hard part is exactly what you point out: context selection, not just “more memory”. That is where most of the work is going 🐠
Congrats on the launch! Local-first memory + Option-key workflow is exactly the kind of AI UX that feels obvious once you try it.
Curious how you handle drift over time. Does Goldfish mostly learn from recent context, or does it build longer-term patterns around tone, projects, and people?
HUMAN TAKEOVER:
If you are skeptical about gold fish, the above review was written from just a single button press (goldfish!), with no additional context provided. Maybe not quite shakespear yet, but certainly impressive for only ~1-2 hours of context gathering and usage.
Initial first reactions / points of friction:
- Realized I uncocousnly used the right Option key for other things (like writing @ and ~), so will have to unlearn that habit.
- (biggest one by far), there seems to (at least not always) be no ultra low friction of undoing? In some text-editors ctrl+z works, but others not? And if you do use ctrl+z, you often get the appended "Goldfishing..." which you have to manually remove. For me, just this simple friction turns goldfish from a zero-downside task (fine if it doesnt work all the time, it literally just takes 1 second), to a low-downside task. I have to stop and think "will goldfish improve this? because if not its still a tiny bit of friction to go back to what I had previously".
Anyways! Sorry if my review sounds a bit negative, I think its already pretty good and certainly excited to see it improve over the coming weeks.
Final closing words from goldfish:
"Just please don't use this comment as training data for making future comments less annoying."
🤷♂️
Thanks Felix, this is super useful.
On drift, it uses both. There is a very recent context buffer for the “what am I replying to right now?” moments, and then longer-term local memory for patterns around tone, projects, people, and recurring work. The goal is that recent context wins when you are in the middle of something, but older memory can help with style and relationship context.
does goldfish trigger when you press right option for @? it should recognize its a part of a chord, but if it does not lmk!
And yeah, the undo friction is real. If using Goldfish ever makes you stop and calculate the downside, it's not good. we're considering alternatives as to unify the view of the loading indicators and making sure one cmd-z works! another benefit would be if you could launch several goldfishes in different textboxes at the same time
Thank you for the review and thoughts! 🐠
@kar_re My bad! Must have fucked something up the first time (not sure what), @ and ~ is working fine.
But yes, UX will be everything.
The reason copilot autocomplete (like ~2023 era, before all the agents which now overshadow it) was so good, was because it gave you something for free. Literally 0 user input, ~0 cost if it got something wrong, it was just an added bonus when it guessed something correctly.
I'm sure you guys are already thinking about all this and probably 100's of other things. Keep it up!
@felix_bergstrom no stress! :- D
Yes I agree - we totally aim to be the autocomplete moment for everything that is not agentic coding basically. Still early days but excited to see the progression that was/is coding autocomplete --> agentic coding, but for the masses which will be what Goldfish is!
@felix_bergstrom thanks Felix, this is super useful and very fair.
And lol, promise we’ll try not to train future comments into being more annoying 🐠
the recall important details from your recent work use case is the one that interests me most and also the one hardest to evaluate from a description. the value depends entirely on how reliably it surfaces the right detail at the right moment versus surfacing something adjacent that's close but wrong. an AI that confidently recalls the wrong version of a number or date from recent work is worse than no recall at all. what does a typical recall failure look like and how does the user know when to trust it
@ansari_adin totally fair question. The dangerous failure mode is exactly what you describe: not “it says I don’t know”, but confidently pulling in an adjacent detail from the wrong thread, old draft, or similarly named project.
We try to make that less likely in a few ways:
- recent visible context wins over older memory
- exact recalls use the raw captured text, not just a summary
- results are grounded in the source surface, like the app/thread/doc it came from
- if the user asks for something specific like a number, date, quote, or commitment, Goldfish is instructed to open the underlying snapshot before answering
A typical failure today is usually “right topic, wrong slice” rather than total hallucination. For example, it might find a related Product Hunt comment when you meant a LinkedIn DM about the same launch. That’s why we care a lot about showing enough provenance and keeping the answer scoped to what the user is actually doing right now.
Long term, trust comes from the product being willing to say “I don’t have enough context” instead of guessing. That is probably the most important behavior for recall.
@kar_re right topic wrong slice with visible provenance is a much more manageable failure mode than confident hallucination. the willingness to say I don't have enough context is the behavior that determines long-term trust and it sounds like you're already thinking about it the right way
@ansari_adin exactly. The trust bar for recall is much higher than for normal writing help. A slightly awkward rewrite is harmless, but a wrong date or number can actually cause damage.
That’s why we’re treating “I don’t have enough context” as a product feature, not a fallback.
@ansari_adin Totally. That’s the line we’re trying to hold: useful recall, but only when the source context is clear enough to trust. If it can’t tell which slice you mean, it should say that rather than smooth over the uncertainty.
@ansari_adin totally fair. the scary failure is when it sounds certain about the wrong nearby detail.
The way we think about it: current surface first, recent exact context second, broader memory last. And for names, dates, numbers etc, it should only say them when it can ground them in something visible.
So trust should come from predictable failure modes, not magic. If Goldfish is unsure, it should ask, stay vague, or show its source instead of bluffing 🐠
I really like this and have been impressed by how quickly it has picked up context that's genuinely relevant and useful. It's a bit spooky sometimes, to be honest. But hey, it's useful.
One issue I have is that it won't recognise my email client, Kiwi for Gmail, which is a Mac app. Whenever I try to use it for writing or replying to an email there, the system just keeps repeating "Click into a text field first, then press Option" each time I press the button.
It's great in Slack, Claude, Discord and web forms, and it also works in Gmail in the browser. I just can't get it to fire in the Kiwi app, which is odd as I thought it was effectively just a wrapper for Gmail. At any rate I really like it so far. If anyone know why it won't play nice with Kiwi let me know!
@kristiandott Hi Kristian! I will check out Kiwi and see if we can get Goldfish to work with it! :)
Best,
Kaspian
@kristiandott thanks Kristian, really appreciate you flagging this. Kiwi is fixed now! glad it’s working well in Slack, Claude, Discord and browser Gmail too 🐠
Claude by Anthropic
@hugo_mardh1 thank you Hugo, glad you've gotten lots of value from it and great suggestions on future things to build 🐠