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.
Replies
Best
The BEAM benchmark result is interesting 54% vs 36% SOTA is a meaningful jump. But I'm curious about the retrieval side in practice: when I open a Claude chat about a specific project, how does Minimi know which slice of my captured context is relevant to surface? Is it purely semantic similarity against my current query, or are you also factoring in recency and which app I'm currently in? Asking because the failure mode I'd worry about is Claude getting confidently wrong context not missing context especially after weeks of accumulated memory where older stuff might contradict newer decisions.
@veerhunt_agai - I understand. The benchmarks measure the efficacy of the memory layer we have built. It is pretty good! And as context grows, it continues performing just well enough. So you need not worry about it I would say :)
Report
Not totally sure from the page, but the useful part seems to be reducing context switching rather than just adding AI. How are you measuring that in real usage?
@xiaosong001 - hey, there are benchmarks called Long-Mem-Eval-S and BEAM that measure the efficacy of memory layers like Minimi. We have a 50% higher score than the current state-of-the-art benchmark.
In real usage, I guess you will just experience the AHA moment within 30 minutes of querying Claude about something personal that none of you connected MCPs can capture :)
I think this is a must-have for a Claude! Will give it a try. We're launching today as well, congrats to all of us!
Report
Minimi sounds powerful, but context quality is often more important than context quantity. How does Minimi decide what information is actually relevant to send to Claude, and how do you prevent important details from getting lost in large amounts of captured context?
Report
Great idea, but how can it be confidential if everything is then passed to Claude? Is it an isolated cloud-based Claude instance, or am I misunderstanding something?
Replies
minimi
@veerhunt_agai Great question this is exactly what we're heads down on.
minimi
@veerhunt_agai - I understand. The benchmarks measure the efficacy of the memory layer we have built. It is pretty good! And as context grows, it continues performing just well enough. So you need not worry about it I would say :)
Not totally sure from the page, but the useful part seems to be reducing context switching rather than just adding AI. How are you measuring that in real usage?
minimi
@xiaosong001 - hey, there are benchmarks called Long-Mem-Eval-S and BEAM that measure the efficacy of memory layers like Minimi. We have a 50% higher score than the current state-of-the-art benchmark.
In real usage, I guess you will just experience the AHA moment within 30 minutes of querying Claude about something personal that none of you connected MCPs can capture :)
Scoutflo
I started using Minimi for the last 12 hours and I could instantly see a drastic change in my Claude usage.
Really need this!
Good to know it's completely on my device memory.
minimi
@kalpesh_bhalekar1 Glad that you are enjoying it!
minimi
@kalpesh_bhalekar1 - thanks a ton for always supporting us. Hope you enjoy using Minimi. Just remember Cotton is now on your laptop always :))
Loved seeing the product in action. Can't wait to use it- I'm tired of pasting messages from one tab to another to give "context"
minimi
@sarvagya_k Glad to hear - I hope you enjoy using Minimi. Thank you! :)
Tried downloading and trying it but says not compatible with this Mac.
minimi
@new_user_2455f400b4 - hi, what Mac are you on? Minimi only supports Apple Silicon :))
minimi
@new_user_2455f400b4 Sorry about that! Which Mac are you on? It should work on M1 and above DM us and we'll get it sorted.
Is it free to use???
minimi
@sosna yes please try
minimi
@sosna Yup:)
Veltrix AI
Great idea, will give it a try! We're launching a product today as well, so good luck to all of us!
Veltrix AI
I think this is a must-have for a Claude! Will give it a try. We're launching today as well, congrats to all of us!
Minimi sounds powerful, but context quality is often more important than context quantity. How does Minimi decide what information is actually relevant to send to Claude, and how do you prevent important details from getting lost in large amounts of captured context?
Great idea, but how can it be confidential if everything is then passed to Claude? Is it an isolated cloud-based Claude instance, or am I misunderstanding something?