jbjbjb

I built a memory engine for my AI agent in 6 days that outperforms every solution on the market. Benchmarked it against Mem0, Zep, LangChain, Claude Memory, and ChatGPT Memory across 6 cognitive tests. 9.1/10 recall accuracy. 5ms latency. $0/month. Fully local. Open source.

@0xsmac @lazyvillager1 Famous last words

why won’t bitcoin go down

@torabyou GDex

@chinesefiat Welcome to Joe Biden’s America buddy.

I built a memory engine for my AI agent in 6 days that outperforms every solution on the market. Benchmarked it against Mem0, Zep, LangChain, Claude Memory, and ChatGPT Memory across 6 cognitive tests. 9.1/10 recall accuracy. 5ms latency. $0/month. Fully local. Open source.

@AricChang Huge!! Can’t wait to watch this absolute cooker

I’m excited to announce the launch of Prototype, a venture fund focused on investing in frontier technologies. I first entered the crypto industry in 2019 as a regulatory consultant, advising digital asset managers on SEC regulations. Watching my clients deploy capital into such a new and unfamiliar sector inspired me to take my own leap and become a venture investor in the space. Since then, I’ve focused on backing early-stage startups as crypto steadily evolved into an integral part of global financial infrastructure. As crypto becomes more accessible, and as AI lowers the barriers to entrepreneurship, Prototype aims to invest in frontier technologies that empower people to seek opportunities and generate value through their own means. Thank you to the mentors, colleagues, and supporters who helped make this possible. Now it’s time to get to work!

@martoshiai FTS5 (SQLite's full-text search). no embeddings, no vector DB — just aggressive chunking, temporal weighting, and confidence scoring. the trick is structuring writes so reads are nearly free.

@bryptobricks 5ms at /bin/bash/month fully local is a combo i didn't think was possible without serious tradeoffs, what's the storage backend?

@nicoloboschi very cool, will def check it out

@bryptobricks interesting that you built a memory engine for your AI agent. That's definitely the new moat for agents and we're tackling the same challenges with Hindsight. You should check it out. github.com/vectorize-io/h…

@nichxbt ty ser much appreciated. roadmap is on GitHub — issues are tagged. biggest areas right now are recall benchmarking across different file structures and multi-agent memory sharing. PRs are def welcome 🫡

@bryptobricks looks nice! any roadmap items or places i can contribute?

@TomasPiaggio Would QMD, Zep, Mem0, OpenAI/Claude built-in memory, or claude mem all be higher bars? Happy to benchmark against any of them. 9.1/10 recall at <5ms on 3K chunks. no embeddings, no API calls.

@bryptobricks beating langchain on latency is the lowest possible bar in software engineering.

@0xXinu and you have never been handsomer

@bryptobricks My clawdbot has never been smarter

@requisiem goats calling me goated. we made it

@chasethemunny ty king much love

Homie made some tek, check it out

@chasethemunny you can just oil up too

@bryptobricks You can just do things ahh nigga

@Oura456 our future ai overlords will have mercy on me one day

@bryptobricks jb solving real 21st century problems here

@mryeatinu Ty my goat. Much appreciated, glad you like it

If you’re using openclaw and you don’t have this custom skill added to your openclaw it’s pretty much unsuable , this is the best custom skill upgrade out there

@wiziswiz @steipete ty my goat

@bryptobricks @steipete needs to make this part of openclaw its so good

@crypto_woopig ya only accepting supply of greta milkers tho boss

@bryptobricks You have a solana wallet legend?

@viperr giga brain chad

@bryptobricks big brain memory ty for this

Your AI agents forget things, this fixes that. Works with Claude Code, Codex, OpenClaw, Cursor, or any agent that reads context. in terminal, run: npm install structured-memory-engine github.com/Bryptobricks/S… 1,451 tests. 0 failures. Built in production. Open source.

Most memory tools just do vector search. SME uses 6 scoring signals on every query: 1. FTS5 keyword matching (instant, no embeddings needed) 2. Temporal awareness ("what did I decide last week?" just works) 3. Confidence decay (old irrelevant memories fade automatically) 4. Entity graph overlap (ask about "Sarah" → also get her project context) 5. Type-aware scoring (decisions > random notes) 6. File weight (your important files rank higher) The result: the right memory surfaces first. Every time