Every great Claude response starts with context. minimi listens across your Mac - docs, calls, messages, tabs - and gives Claude the full picture. No prompting. All on-device and private.
I've been living inside Claude for most of my workday, and the one thing that always frustrated me was having to re-explain myself every single session. "Here's what I'm working on. Here's what happened in my last meeting. Here's the email thread you need to know about."
Minimi fixes that. It sits quietly on my Mac, reading what I read, hearing what I hear - and then feeds all of that to Claude as live context. So when I open a new chat and ask "what should I follow up on from this morning?", Claude already knows. No briefing. No copy-paste. Just the answer.
A few things I love about Minimi:
1. On-device memory - your context never leaves your Mac (the vector DB lives locally). We benchmark at 54% on BEAM vs the previous SOTA's 36%.
2. MCP-native - one link, paste it into Claude's custom connector, done. No new app to live in.
3. Granular control - you pick which apps it can see. Pause anytime.
If you use Claude and you work on a Mac, this is a no-brainer install. Three steps and it just works.
@jay_gadekar so excited to have built it alongside you and our team! <3
Report
Hunter
@jay_gadekar Many congratulations on the launch! :)
Really, really beautiful landing page, so cute, and I love the branding!
Minimi is your ambient memory for Claude, a Mac app that quietly captures everything you do on your computer (every tab, document, call, and Slack thread) and feeds it to Claude as live context.
Instead of manually briefing Claude or hunting through your history, you can just ask questions like "Who sent me the screenshot about the bug?" or "What did we decide in yesterday's meeting?" and Claude will know.
I endorse it because it's 50% more accurate than previous memory systems (54% vs 36% on the BEAM benchmark), keeps your memory on-device in a local vector database with nothing stored on the cloud, and lets you skip the extra prompting to get straight to answers.
This is exactly what AI assistants have been missing, true long-term memory that actually works while protecting your privacy.
@jay_gadekar Started using Minimi since yesterday. I am impressed. Now I am actually looking forward to use Shram from Monday. Great going and keep up the momentum! 👏
@jay_gadekar The on-device memory angle is really clever—keeping context local while integrating seamlessly into Claude's workflow removes a lot of friction. The MCP approach means you're not asking users to adopt yet another tool, which is smart. Curious how you're thinking about context relevance over time—does it get smarter about what's actually useful to surface, or does that depend on how Claude's context window handles it?
@saulfleischman - thank you! Partly both. Our memory layer benchmarks are 50% higher than the current state-of-the-art for BEAM and much better than SuperMemory and Mem0 for Long-Mem-Eval-S. We will keep improving retrieval as the context window of our users goes through the roof - that's the best we can do rn since the product is not even 2 weeks old. Also, the LLM in the loop being better definitely helps a ton!
Nice product! The on-device, you-pick-what-it-sees approach is the part that I think makes this actually look really usable. I spend my time in the Claude ecosystem too (building governance tooling around skills/access), so the granular per-app control especially caught my eye. Quick question: when you pause it or revoke an app, does the context it already captured from that app stay in the local store, or get dropped?
@tom_palmer_ux - thanks for writing back. When you pause - say for 5 or 10 min, your memory won't be created for that duration. Please feel free to ask more queries. Good day! :)
the context bottleneck is real. most bad AI output i see is a missing-context problem, not a model problem, so this direction makes a lot of sense. the part id be curious about is signal vs noise. passively capturing everything across docs/calls/tabs is powerful, but the risk is feeding Claude confidently-irrelevant context. how you decide what's actually worth surfacing feels like the real moat here. on-device + private is a smart trust call too. nice work.
In Minimi - updates, contradictions, and temporal order are handled as core behavior, not patched on.
It's why we measure ourselves on BEAM rather than the older recall-only benchmarks. BEAM runs at 1M and 10M token scale and can't be solved by a bigger context window, so it directly tests the staleness question.
We're at 54% vs the prior 36% SOTA, with most of the lead on the over-time tasks.
Short version: maintaining an accurate picture beats retrieving more, every time!
@ozandag Anyone can capture everything; the value is in what you choose to surface. We optimize for an accurate picture over raw recall, which is why we benchmark on BEAM and LongMemEval rather than recall-only tests — these run on very long conversations where the retrieval system has to surface only the relevant pieces. And keeping it on-device.
@ozandag You've named the actual hard problem. Capturing everything is easy. Knowing what's relevant to this conversation is the work.
The way we handle it: Minimi doesn't dump everything into Claude's context. It retrieves based on what you're currently doing - the app you're in, the conversation you're having, the doc you're editing. The surface area Claude sees is narrow and intentional, not a firehose.
The deeper moat is temporal + behavioural signal - what you've engaged with recently, what you keep returning to, what you've explicitly acted on. That's what separates useful context from noise. Still building on this, but it's the core of what makes the benchmark results hold up in practice.
Report
Neat idea. Can you tell Minimi to skip certain apps it shouldn't capture context from?
A lot of memory systems seem useful while a conversation is active, but the harder test is what happens after weeks of accumulated context.
How are you thinking about memory quality over time? Is the bigger challenge helping Claude retrieve more information, or helping it maintain an accurate picture of what's still true versus what's become outdated?
@zaid_mallik1 In Minimi - updates, contradictions, and temporal order are handled as core behavior, not patched on.
It's why we measure ourselves on BEAM rather than the older recall-only benchmarks. BEAM runs at 1M and 10M token scale and can't be solved by a bigger context window, so it directly tests the staleness question.
We're at 54% vs the prior 36% SOTA, with most of the lead on the over-time tasks.
Short version: maintaining an accurate picture beats retrieving more, every time!
@zaid_mallik1 Really the right question - and honestly the harder engineering problem. Retrieval is mostly solved. Accuracy over time will need more work.
The way we think about it: Minimi captures chronologically, so context has a timestamp. Claude can reason about recency - what you discussed last week vs last month - rather than treating everything as equally current. We're also working on explicit memory updates, where newer context can surface and deprecate older facts.
The bigger unsolved problem is knowing what you consider still true. That's more personal signal than technical - we're exploring ways to let users flag it directly.
@dynatrading Hi Andy - the infrastructure we rely on - Accessibility - is not currently reliable for Windows, thus we have not gotten around making a windows version.
However building ambient memory for Windows is something we are absolutely going to get on very soon!
@dynatrading We went Mac-first to get the capture quality right every app, zero integrations, completely passive. Replicating that on Windows takes time to do properly.
@dynatrading The re-explaining tax is real - most people just accept it as part of using AI. It shouldn't be.
On Windows - Mac-first was a deliberate call, not a limitation. The on-device architecture we've built runs close to the OS in ways that need platform-specific work. Windows is on the roadmap but we want to do it right. :)
Minimi is the most delightful part of my day. It has even made me a better, more thoughtful gifter haha 😛
SUPER stoked that others can now play around with it.
Here are some fun and work related things you can try doing!
Fun
"What should I get Jay for his birthday?" and it actually knows, because it remembers the offhand thing he wanted three weeks ago on a call.
"What was that restaurant someone raved about last month?" No idea who, no idea when. Minimi finds it.
"Recommend a movie for tonight" and the pick actually is awesome, because it knows what I've genuinely been into lately.
Work
"Draft a follow-up from my call with Niket" and it pulls exactly what we discussed.
"What did we decide about the UX copy?" answered in one line, across scattered Slack threads, docs, and calls.
"Catch me up on what I missed" after a long deep work session, so I walk back in already knowing where things stand.
Do try and let me know what you built <3
Report
@ojasvika_sahu the gift example — remembering an offhand thing someone said on a call weeks later — is exactly the magic tbh. but if its hearing everythiing, how does it know that one line mattered vs the 99% thats just background chatter? curious if thats tuned or you just store it all and let retrieval sort it out
@haotian_wang5 the retrieval is just the magic. We have built SOTA memory - it has beat the published benchmark of 0.36 (BEAM) by 50% - that is why the results are really on point!
Minimi catches literally everything. Claude basically now has my personal context and knows everything. Helps across the workday. Remembers weeks of conversations and makes work way more productive than earlier. I can ask it "What are the tasks I should handle urgently" and it knows. I can ask it "Who all did I talk to today" and it will tell me the names, platform the conversation happened on and the context. I also use it to remember followups. Really deep use-case.
Have been lucky to get early access to Minimi and my god it’s powerful! From getting random, small insights that I forgot from my meetings to tracking my work output to remembering things that I did 2 weeks ago. Minimi is like magic
Fr. Giving context to every LLM for the same thing I had it do yesterday and the day before is frustrating. About time someone built a plug-and-play memory layer and relieved me of the annoying ritual. Great work, team. Rooting for you.
Replies
minimi
minimi
@jay_gadekar so excited to have built it alongside you and our team! <3
@jay_gadekar Many congratulations on the launch! :)
Really, really beautiful landing page, so cute, and I love the branding!
Minimi is your ambient memory for Claude, a Mac app that quietly captures everything you do on your computer (every tab, document, call, and Slack thread) and feeds it to Claude as live context.
Instead of manually briefing Claude or hunting through your history, you can just ask questions like "Who sent me the screenshot about the bug?" or "What did we decide in yesterday's meeting?" and Claude will know.
I endorse it because it's 50% more accurate than previous memory systems (54% vs 36% on the BEAM benchmark), keeps your memory on-device in a local vector database with nothing stored on the cloud, and lets you skip the extra prompting to get straight to answers.
This is exactly what AI assistants have been missing, true long-term memory that actually works while protecting your privacy.
minimi
@rohanrecommends - thanks Rohan. :)
minimi
@jay_gadekar @rohanrecommends thanks for this Rohan! Do share your feedback with us :)
@jay_gadekar Congrats! Love the idea, especially that it's ambient (aka frictionless). Sadly, I'm PC - any chance you'll be doing a PC version soon?
minimi
@jay_gadekar @anna_ludwinowski Hi Anna! At this moment we are not on PC - but will be there soon :)
minimi
@anna_ludwinowski - thank you! Support for PC is certainly on the roadmap - will release soon. :)
AutoSend
@jay_gadekar Started using Minimi since yesterday. I am impressed. Now I am actually looking forward to use Shram from Monday. Great going and keep up the momentum! 👏
minimi
@designerdada - thanks Akash - means a lot coming from you. Will keep you posted! :))
RiteKit Company Logo API
@jay_gadekar The on-device memory angle is really clever—keeping context local while integrating seamlessly into Claude's workflow removes a lot of friction. The MCP approach means you're not asking users to adopt yet another tool, which is smart. Curious how you're thinking about context relevance over time—does it get smarter about what's actually useful to surface, or does that depend on how Claude's context window handles it?
minimi
@saulfleischman - thank you! Partly both. Our memory layer benchmarks are 50% higher than the current state-of-the-art for BEAM and much better than SuperMemory and Mem0 for Long-Mem-Eval-S. We will keep improving retrieval as the context window of our users goes through the roof - that's the best we can do rn since the product is not even 2 weeks old. Also, the LLM in the loop being better definitely helps a ton!
Cloudskill
Nice product! The on-device, you-pick-what-it-sees approach is the part that I think makes this actually look really usable. I spend my time in the Claude ecosystem too (building governance tooling around skills/access), so the granular per-app control especially caught my eye. Quick question: when you pause it or revoke an app, does the context it already captured from that app stay in the local store, or get dropped?
minimi
@tom_palmer_ux - thanks for writing back. When you pause - say for 5 or 10 min, your memory won't be created for that duration. Please feel free to ask more queries. Good day! :)
minimi
@tom_palmer_ux thank you for trying out Minimi! Please share your feedback with us soon :)
minimi
@tom_palmer_ux By pausing, we don't capture anything from that window from the moment you turn it on
minimi
@tom_palmer_ux Hi Tom, it stops from the moment you pause Minimi.
AISA AI Skills Test
the context bottleneck is real. most bad AI output i see is a missing-context problem, not a model problem, so this direction makes a lot of sense. the part id be curious about is signal vs noise. passively capturing everything across docs/calls/tabs is powerful, but the risk is feeding Claude confidently-irrelevant context. how you decide what's actually worth surfacing feels like the real moat here. on-device + private is a smart trust call too. nice work.
minimi
@ozandag Hi Ozan, even @zaid_mallik1 asked me the same question!
Here was my answer:
minimi
@ozandag Anyone can capture everything; the value is in what you choose to surface. We optimize for an accurate picture over raw recall, which is why we benchmark on BEAM and LongMemEval rather than recall-only tests — these run on very long conversations where the retrieval system has to surface only the relevant pieces. And keeping it on-device.
minimi
@ozandag We are super accurate with what to surface. The underlying tech of Minimi helps with the accuracy.
minimi
@ozandag Capturing everything is table stakes knowing what to show Claude, and when, is where it gets hard.
minimi
@ozandag You've named the actual hard problem. Capturing everything is easy. Knowing what's relevant to this conversation is the work.
The way we handle it: Minimi doesn't dump everything into Claude's context. It retrieves based on what you're currently doing - the app you're in, the conversation you're having, the doc you're editing. The surface area Claude sees is narrow and intentional, not a firehose.
The deeper moat is temporal + behavioural signal - what you've engaged with recently, what you keep returning to, what you've explicitly acted on. That's what separates useful context from noise. Still building on this, but it's the core of what makes the benchmark results hold up in practice.
Neat idea. Can you tell Minimi to skip certain apps it shouldn't capture context from?
minimi
@dhiraj_patel5 yes yes! You can exclude apps!
minimi
@dhiraj_patel5 Yes absolutely! You can block apps as well as websites.
minimi
@dhiraj_patel5 - yes, you can block apps to not make memory from on your Minimi home page :))
minimi
@dhiraj_patel5 yes
minimi
@zaid_mallik1 In Minimi - updates, contradictions, and temporal order are handled as core behavior, not patched on.
It's why we measure ourselves on BEAM rather than the older recall-only benchmarks. BEAM runs at 1M and 10M token scale and can't be solved by a bigger context window, so it directly tests the staleness question.
We're at 54% vs the prior 36% SOTA, with most of the lead on the over-time tasks.
Short version: maintaining an accurate picture beats retrieving more, every time!
minimi
@zaid_mallik1 Hope Ojasvika's answer has clarified your question. Feel free to ask if there's anything else, Zaid.
minimi
@zaid_mallik1 Really the right question - and honestly the harder engineering problem. Retrieval is mostly solved. Accuracy over time will need more work.
The way we think about it: Minimi captures chronologically, so context has a timestamp. Claude can reason about recency - what you discussed last week vs last month - rather than treating everything as equally current. We're also working on explicit memory updates, where newer context can surface and deprecate older facts.
The bigger unsolved problem is knowing what you consider still true. That's more personal signal than technical - we're exploring ways to let users flag it directly.
minimi
@zaid_mallik1 Both matter. The real test is memory that stays true weeks in that's exactly what we're building for.
The "no re-explaining yourself" pain point is so real — I spend a chunk of every session giving Claude context it had yesterday.
Love the on-device angle too. Privacy-first local storage is the right call when your context includes work meetings and personal projects.
One question: any Windows roadmap? That's my main blocker for trying it today.
minimi
@dynatrading Hi Andy - the infrastructure we rely on - Accessibility - is not currently reliable for Windows, thus we have not gotten around making a windows version.
However building ambient memory for Windows is something we are absolutely going to get on very soon!
minimi
@dynatrading We went Mac-first to get the capture quality right every app, zero integrations, completely passive. Replicating that on Windows takes time to do properly.
minimi
@dynatrading Hopefully soon, Andy!
minimi
@dynatrading The re-explaining tax is real - most people just accept it as part of using AI. It shouldn't be.
On Windows - Mac-first was a deliberate call, not a limitation. The on-device architecture we've built runs close to the OS in ways that need platform-specific work. Windows is on the roadmap but we want to do it right. :)
minimi
@dynatrading we went Mac first to get the capture quality right. Would love your feedback when it lands!
minimi
Minimi is the most delightful part of my day. It has even made me a better, more thoughtful gifter haha 😛
SUPER stoked that others can now play around with it.
Here are some fun and work related things you can try doing!
Fun
"What should I get Jay for his birthday?" and it actually knows, because it remembers the offhand thing he wanted three weeks ago on a call.
"What was that restaurant someone raved about last month?" No idea who, no idea when. Minimi finds it.
"Recommend a movie for tonight" and the pick actually is awesome, because it knows what I've genuinely been into lately.
Work
"Draft a follow-up from my call with Niket" and it pulls exactly what we discussed.
"What did we decide about the UX copy?" answered in one line, across scattered Slack threads, docs, and calls.
"Catch me up on what I missed" after a long deep work session, so I walk back in already knowing where things stand.
Do try and let me know what you built <3
@ojasvika_sahu the gift example — remembering an offhand thing someone said on a call weeks later — is exactly the magic tbh. but if its hearing everythiing, how does it know that one line mattered vs the 99% thats just background chatter? curious if thats tuned or you just store it all and let retrieval sort it out
minimi
@haotian_wang5 the retrieval is just the magic. We have built SOTA memory - it has beat the published benchmark of 0.36 (BEAM) by 50% - that is why the results are really on point!
minimi
@ojasvika_sahu @haotian_wang5 We capture everything — retrieval figures out what matters.
minimi
@ojasvika_sahu - yup! Proud to have built this together :))
minimi
@ojasvika_sahu The usecases are so insane!
minimi
Minimi catches literally everything. Claude basically now has my personal context and knows everything. Helps across the workday. Remembers weeks of conversations and makes work way more productive than earlier. I can ask it "What are the tasks I should handle urgently" and it knows. I can ask it "Who all did I talk to today" and it will tell me the names, platform the conversation happened on and the context. I also use it to remember followups. Really deep use-case.
minimi
@niketrajdwivedi - yup! Proud to have built this together :))
minimi
@niketrajdwivedi minimi started with such a small spark - to now see it become an awesome side project. Crazy stuff.
Writee AI
Have been lucky to get early access to Minimi and my god it’s powerful! From getting random, small insights that I forgot from my meetings to tracking my work output to remembering things that I did 2 weeks ago. Minimi is like magic
minimi
@prannay_kedia your initial feedback was critical for us to build ahead. Thank you for supporting us so early on!
minimi
@prannay_kedia - thanks Prannay for being amongst our earliest users!
minimi
@prannay_kedia Always glad to have your feedback, Prannay!
minimi
@prannay_kedia thank you
Fr. Giving context to every LLM for the same thing I had it do yesterday and the day before is frustrating. About time someone built a plug-and-play memory layer and relieved me of the annoying ritual. Great work, team. Rooting for you.
minimi
@kritarthmittal Thank you for trying out Minimi Kritarth! Really appreciate your support!
minimi
@kritarthmittal - thanks Kritarth - for always being a true believer in us. Hope you enjoy Minimi :)
minimi
@kritarthmittal Thanks, Kritarth!