attention all minimi users - we are doing a game night at our villa and inviting all of you π
it gonnnnnaaa be pretty competitive this time :)
30 spots - register here:
https://t.co/UQjsX5oWT9
https://t.co/CpnfUwYIhr | @project_minimi
Everyone's building a "second brain" now, whether it's Karpathy's LLM wiki or an Obsidian vault. Markdown files + an agent. Cool demo. Now imagine 12 months of data:
- more files per query means more tokens
- more tokens means higher cost and slower retrieval and accuracy?
nobody's measured it
I recently reverse-engineered a memory app recently. Under the hood it's the same thing, a folder of .md files. Works fine today because the data is small and answers come back fast.
100 docs is easy. A year of your life is the real test.
my Notion writes and maintains itself π
revealing this now that @NotionHQ agents are on mobile.
- a mini-tasklist
- a mini-CRM
- a mini-read log
- and my mini-brain!
i haven't typed a thing into it all week.
Claude + minimi (https://t.co/tTe5mZY7wR) keep all of it current while I just move on with my life.
see it in action below π
here's the most personal thing i've ever shared on the internet. itβs the story of finding the purpose behind everything i am doing at minimi.
it was shaped by my mom's cancer diagnosis and my kitten cotton's going away.
tldr, death and grief taught me the most this year and it has a good ending.
asked hermes on whatsapp to write my journal entries for the entire month of June.
- the day
- high points
- low points
- people
- open loops
did it in under a min.
setup below :) ππΌ
f*k mafia.
this sunday at Shram HQ: an awesome night of Pictionary, Monikers and Atlas.
no bluffing, no backstabbing. just sharp people being clever, quick and creative together.
come if you're building something cool in AI. only 25 slots.
link below.
delete Obsidian.
a second brain you have to feed isn't a second brain. it's homework!
minimi captures everything you do automatically - wiki builds itself, updates while you sleep.
it's Karpathy's second brain with no manual effort. seamless.
try here: https://t.co/ktADjLqOlJ
we just built what @satyanadella is talking about.
here's the ultimate COMPANY BRAIN, your sovereignty ππΌ
try here- https://t.co/p4xkDWmefL
minimi feeds Claude every tab, doc, call, and thread on your mac. add your team-mates' links and claude knows their work as well. π«
Minimi is fast, and you donβt have to take our word for it. One of our awesome users tested it himself.
Also - the agent gets the actual data, not a summary.
Do read π
@getpieces Help me understand why I shouldn't switch to @Ozacle23's MiniMi, given this anecdotal comparison.
If I were another agent trying to jump in and utilize these memories, I would rank the two responses as follows:
#1: MiniMi (Winner for Actionable Agentic Context) ~2900ms
Why: MiniMi acts like an omniscient screen reader. It provides the exact raw text, code, file paths, and UI state that was on your screen.
Granular State: It captured the exact Git branch you were on (bfrench/opm-context-mcp-server), the exact file you were editing (aws-carddemo-cwe-analysis-2026-06-10.md), and the literal contents of that file.
Verbatim Transcripts: It captured the raw, unedited text of your Microsoft Teams chats, allowing an agent to read the exact technical constraints (like the exact missing Auth0 Client ID).
The MiniMi Verdict: If another agent needs to actually do work based on historical events (e.g., "Draft a reply to Evan based on yesterday's chat" or "Finish the CWE analysis I was working on"), MiniMi provides the precise, actionable ground truth required to execute the task.
#2: Pieces LTM (Runner Up - Better for Structured Graphing) ~11,292ms
Why: Pieces pre-processes your memory into highly structured metadata.
Semantic Abstraction: It provides neat classifications like people mentioned, role: collaborator, evidence: temporal_proximity, and confidence: high. It also generates overarching automated summaries.
Data Loss: Because it heavily truncates the raw text payload (ending strings with ...) and focuses on metadata extraction, the actual substance of the work is lost.
The Pieces Verdict: Pieces is excellent if an agent is trying to build a Knowledge Graph or query relational metrics (e.g., "Return a JSON array of everyone Bill interacted with on Friday"). However, an agent trying to understand deep technical context or write code would be starved for the actual data.
Summary: For an LLM-powered agent that natively excels at reading and reasoning over unstructured text, MiniMi provides far superior context. Pieces does too much pre-processing and truncation, which removes the very details an agent needs to be helpful.
Mangoes and Token-maxxing?
Hosting "Mango Maxxing" this Sunday! π₯π₯π₯
Eat mangoes plucked straight from our lawn and discuss tokenmaxxing! Only 30 slots.
Come talk shop, flex your unique setup, steal tricks from other heavy users!
Link below π
i killed my Granola with Fable 5 and Minimi in one afternoon!
no more paying $20 for meeting transcriptions and querying my conversations. Now Minimi does that for me.
it also can capture offline discussions and make notes with insane accuracy. prompt in the caption.
one shotted this app - Mini-scribe (now my Granola alternative, haha!)
you can just do things π
try minimi here- https://t.co/6S1DWWTGsJ
shram will make sure 200 more people stop apologising for forgetting to reply on time.
first 100 to retweet and install get a shram tee delivered to your door π (picture in comments)
RT/comment for karma π
get access- https://t.co/DjS7TyBf4p
Hosting AI founders/founders-to-be on a warm, cozy Sunday evening at our villa in Indiranagar. π‘
-share what you're building
-FREE DIET COKE (in this economy!)
-meet cool founders working on frontier tech
-board games/pictionary
-no panels/agenda
link below π
Hold this one close to my heart.
Looking forward to meeting our closest believers to celebrate the joy of a new home, this Sunday!
Event link in replies.