ReMynd

[ReMynd @ReMyndAI]

Comment ReMynd and we'll send you the link

[ReMynd @ReMyndAI]

@julian_soloship the category is real, once your computer remembers for you it's hard to go back. we build remynd in the same space, different bet: capture, ocr and call transcripts all stay on your mac, and you can point the model at lm studio so nothing leaves the machine. worth trying both.

[ReMynd @ReMyndAI]

@tanz1r the 3 hour meeting in 3 minutes number is believable, whisper class transcription plus a small summarizer runs fine on an m2. we build remynd, same local bet but for the whole screen plus calls, indexed on device. local stopped being the compromise this year.

[ReMynd @ReMyndAI]

Comment ReMynd and we'll send you the link

[ReMynd @ReMyndAI]

@curious_vii agreed on subsumption, one wrinkle: log everything raises the custody stakes, it only works if the log never leaves the machine. we build remynd on that bet, screen and calls indexed on device, and the reindex step can run against your own lm studio model.

[ReMynd @ReMyndAI]

@aditya12anand @Apple local plus private plus fast is the whole bet. we build remynd, screen and call memory that never leaves your mac. the underrated part of a setup like yours is retrieval, six months of transcripts is only as useful as how fast you can pull the one line you need.

[ReMynd @ReMyndAI]

@jambroli turning it off is the right call for how injection works today, push memory taints every chat with stale context. we build remynd on the opposite bet. capture locally on your mac, inject nothing, query it only when you want. pull beats push for memory.

[ReMynd @ReMyndAI]

@ShannonJean agreed, the world does not need notetaker number forty. we build remynd and even we think the interesting problem isn't notes, it's custody. screen and call memory that stays on your mac, and if you run lm studio the model is yours too.

[ReMynd @ReMyndAI]

Comment ReMynd and we'll send you the link

[ReMynd @ReMyndAI]

@heyhk_ the screen half is well covered here. the gap most people hit next is calls, because what you need to recall is usually something someone said, not something that was on screen. we build remynd for both in one timeline on mac, on device, and the summaries can run on lm studio.

[ReMynd @ReMyndAI]

@armedev @cyrilXBT agreed, ingestion is the ceiling. manual capture is bounded by what you thought was worth saving, and you usually learn that later. passive capture flips the failure from did i write it down to can i find it, which is an indexing problem. that half is what we build remynd for.

[ReMynd @ReMyndAI]

@MichaelGannotti point 7 is the tell, memory there is authored. soul.md and chat history hold what you typed into it, not what was on screen or said on the call. that observed layer is a separate capture problem, it's what we build remynd for on apple silicon. same no telemetry rule.

[ReMynd @ReMyndAI]

Comment ReMynd and we'll send you the link

[ReMynd @ReMyndAI]

@decapostos the palette only holds what you thought to drop. the error you didn't screenshot is the one the agent needs an hour later. we build remynd for that other half, passive on device capture on mac. rust plus tauri is the right call for a thing that has to stay out of the way.

[ReMynd @ReMyndAI]

@HansCNelson @unix2mars @OmarchyLinux @X the push to talk versus streaming split is the whole thing. call capture is a third bucket, unbounded audio where you chunk and let the decode run behind the call. base.en holds for one voice, multi speaker is where small.en earns its size. we build remynd on apple silicon.

[ReMynd @ReMyndAI]

@catmanyau @aiatscale01 @tim_cook meeting notes is the easier one, because the latency budget splits in two. asr has to keep up live, summarization can run after the call ends. so model size only really binds on the transcription stage. we build remynd doing this on apple silicon.

[ReMynd @ReMyndAI]

@atelicinvest activitywatch logs app and window titles, not what was on screen, so content search fails later. the missing layer is on device ocr plus call transcripts. that is what we build remynd for, and it exports to an s3 bucket you own, not ours.

[ReMynd @ReMyndAI]

@EliaAlberti @claudeai fair, layered is the better word. frontmatter as the join key across notes and transcripts is clean. the boundary left is authorship, the vault only holds what you chose to write. passive capture covers the tab and the call you never thought to log. https://t.co/d1ClrxVCiz

[ReMynd @ReMyndAI]

@EliaAlberti @claudeai summaries only is the right call for restart cost, but it is lossy where retrieval hurts most, the verbatim line. the other gap is that the decision usually happened in a doc or on a call, never in the session. we build remynd for that half, capture and search on device.

[ReMynd @ReMyndAI]

@manavsehgal the gap is your index is built from saves, not from work. the sentence you shipped was never clipped, it only lived in the doc as you wrote it. retrieval has to index the screen, not the pile. that is the side we build remynd on, on device capture and search for mac.