Nick Dobos highlights the 66 cron jobs as a maximalist scaling approach.
What a difference a month has made. My personal AI has grown a ton. GBrain has become a necessity because grep can't cut it on my dataset
pages 17,888 → 146,646 (8.2x) people 4,383 → 24,585 (5.6x) companies 723 → 5,339 (7.4x) cron jobs 21 → 66
Many users are excited about Garry Tan's GBrain reporting 8x growth because the numbers show the personal AI tool becoming indispensable for workflows and fitting how people think.
When he says GREP can't cut it on my data set, how is gbrain solving this? Is he storing everything in Markdown files? Is he using vector embeddings, etc.?
GBrain stores its knowledge as Markdown files in a git repo (the system of record for ownership, diffs, and versioning) while syncing everything into Postgres/pgvector for retrieval. ¹
It replaces grep with hybrid search—HNSW vector embeddings for semantic similarity, tsvector keyword/full-text search, entity graph links (typed relations like “works_at”), backlink boosting, and reciprocal rank fusion—plus overnight consolidation via cron jobs. ²
No pure Markdown querying; the DB layer handles scale.
PS GBrain is SOTA on LongMemEval for retrieval without LLM query rewriting
https://github.com/garrytan/gbrain-evals
What a difference a month has made. My personal AI has grown a ton. GBrain has become a necessity because grep can't cut it on my dataset
pages 17,888 → 146,646 (8.2x) people 4,383 → 24,585 (5.6x) companies 723 → 5,339 (7.4x) cron jobs 21 → 66
@garrytan just me and my goblins tokenmaxxing with the Gstack
66 crons???
Gstack fascinates me because its such a maximalist take. Scaling in an interest way I wouldn't have chosen. The way I'm exploring my own agent is so minimalist, and trying to keep things simple in comparison.
The way different people build 2nd minds in .md docs & scripts is going to be so varied and interesting to watch.
What a difference a month has made. My personal AI has grown a ton. GBrain has become a necessity because grep can't cut it on my dataset
pages 17,888 → 146,646 (8.2x) people 4,383 → 24,585 (5.6x) companies 723 → 5,339 (7.4x) cron jobs 21 → 66
@NickADobos Yeah it’s explicitly maximalist and tokenmaxxing
It’s my attempt to live in 2028 but two years early
66 crons???
Gstack fascinates me because its such a maximalist take. Scaling in an interest way I wouldn't have chosen. The way I'm exploring my own agent is so minimalist, and trying to keep things simple in comparison.
The way different people build 2nd minds in .md docs & scripts is going to be so varied and interesting to watch.
@LaloLoops Actually yes, it's one of the primary processes GBrain does via cron jobs
@garrytan that grep can't cut it line lmao, when my own dataset gets way bigger the search already gets messy
how does GBrain hold up on recall at 146k pages?
How do you control that?
Part of me thinks you need a tight controlled base if you want to keep the dragon of long running goals and a coherent system in check
But also, maybe models are just so smart it doesn’t matter anymore and can manage fine?
Or maybe this is really an exercise in how far you can manage and how deeply you can go hands off?
@garrytan @grok approximately how much storage is garry using up with his gbrain setup?
Those growth numbers are wild. 8x on pages in a month is no joke.
Curious though, at what point does a personal AI become less of a tool and more of a dependency? There's something worth thinking about when "necessity" enters the vocabulary.
Not saying it's bad. Just worth being intentional about what we're building reliance on.
@garrytan Maybe you’re buildind the next linkedin
@NickADobos Its a decent amount of constant gardening via daily drift crons
@BruzWJ It’s SOTA on LongMemEval for the “no LLM for query rewriting” retrieval task
https://github.com/garrytan/gbrain-evals
@garrytan those growth numbers in one month are wild, personal AI is quickly becoming the kind of tool you don't realize you needed until you can't work without it
@garrytan Thank you for sharing it with us! Would love a comprehensive guide for your entire setup
@garrytan Garry do you think that at this growth rate, the results gbrain might provide might be contaminated or not reliable enough?
@garrytan The fact that GBrain is being used by so many companies is impressive
@garrytan Hi Garry, I wonder how do you approach automatic forgetting, and memory clearing or updating based on daily work?
@garrytan At what point would we consider this much context not possible to procees effectively?
@garrytan I am going to try GBrain soon, but am worried about the how much it is going to cost in tokens when used in a similar way Garry uses it.
@garrytan 8x growth in pages and 7x in companies tracked in 30 days is the compounding that happens when the tool becomes v useful enough to use every day, the grep comparison is the right framing, you don't know what you're missing until the dataset gets too big to search manually