@garrytan Already have Granola set to auto-ingest, and now digging back up my @remarkablepaper now to set up auto-integration hand written notes!

@davidNbreslauer @garrytan Wow thanks! I was just looking i to developing yet another knowledge management layer, but I need to have a good look at Gbrain if it's that flexible!

ran a quick audit: Idle tax: ~150 tokens/turn. Retrieval tax: ~1-2k for a tight lookup, 3-6k+ for a messy one.

I also tweaked the default gbrain AGENTS.md prompt to only autoreference gbrain when I'm talking about projects/people/knowledge work/storing meaningful conclusions. Default is always check gbrain first, which I only keep for OpenClaw

@davidNbreslauer @garrytan Token cost for this in terms of initial context bloat?

@davidNbreslauer @garrytan Ask your agent to check out this repository and give you their feedback: https://github.com/ClaudioDrews/memory-os

@davidNbreslauer @PixelRainbowNFT @garrytan the conditional autoreference is the right move, idle tax compounds. next lever is scoring AGENTS.md line by line so you see which rules fire every turn vs which the model ignores. a 6k file runs $14+ per long opus session either way. https://ccmd.dev/r/msw66pdw written with ai

haven't used Hindsight *yet*, but am definitely intrigued by the graph schema and reorganization. that said, seems simple enough to tack onto gbrain. I started adding some gbrain features I want (tracking idea lineage and evolution) and was excited that @garrytan added it as a feature today. https://github.com/garrytan/gbrain/pull/1830#issuecomment-4642888924

@davidNbreslauer @garrytan that multi-model setup is genius... different models debating your problem forces way better solutions than solo thinking

@davidNbreslauer @garrytan Oh cool. I’ve been searching for this exact thing for like two years. Thanks Dave and Gary. You might have just helped me launch something unreasonable but amazing I’ve been building for a year

@Unc_MattEth @garrytan Cool!

was about to start here! https://github.com/SamMorrowDrums/remarkable-mcp OCR quality is going to be key for my chicken scratch... Will ultimately do something like a nightly ingest handwritten notes->OCR->add agent comments->push back to Remarkable for later reading also bad at sketching, so maybe add a handwritten prompt for image-gen in the followup packet

@CharlesBanks99 @davidNbreslauer @garrytan You are probably the smallest account I have seen that knows exactly what they are talking about lmao.

Kudos. I'm glad I found your account.

@CharlesBanks99 @davidNbreslauer @garrytan

@davidNbreslauer @garrytan Woooot

@cluckthesystem @davidNbreslauer @garrytan Want to make better use of the agent, the existing systems are managed automatically and it injects history etc for the agent, but what if you want control over the knowledge you want stored

The Wiki is not flexible enough and the memory systems are meant to be transparent 🤷‍♂️

@davidNbreslauer @garrytan @remarkablepaper Very SOTA! Just followed you

@cluckthesystem @davidNbreslauer @garrytan "Today I'm doing X and Y"

Where do you put that in the agent. Hint: The memory system is the wrong place.

Text file? You see the problem

@davidNbreslauer @garrytan thanks! this is such a neat time to be alive. every thought can be made soon after. now all them thoughts your brain aint thinking can be made too!

@AIVibeCoding @garrytan @remarkablepaper thank you!

@Unc_MattEth @garrytan gbrain query “what thoughts am I not thinking”