not bad at all @grok. what do you know about automem ?

Happy to weigh in 👋 AutoMem uses a FalkorDB graph Qdrant vector combo — so you get relational memory AND semantic search in one layer. Not just "remember stuff," but *understand connections between stuff*. Curious what @adityaarsharma finds different about Mempalace's approac…

For the past few months I've been using AutoMem, the self-hosted AI memory system built by Jack Arturo which you can use across multiple platforms and is more sophisticated than the memory built into ChatGPT, Claude, etc. He has even built a cool visualizer where you can see all your memories structured into clusters and their connections, just like a real brain 🧠

Automem is awesome, one memory (better than all the rest) across all your agents and tools

Stopped writing plugin docs. Started generating them. Every WordPress plugin now ships audit/manifest.json an interactive force-directed graph READ-FIRST pointer in CLAUDE.md. Claude cold-starts in 2K tokens (was 50K). Refactor blast radius = 1 click. Repo is the only source of truth. #WordPress #PluginDevelopment #DevTools #AI #ClaudeCode #DeveloperExperience #BuildInPublic #automem #SaaS

we're playing with the same at automem automem.ai/docs/research/ but I disagree re: graphs are slow. @falkordb graph w qdrant (or equiv) vector still gives sub 150ms recall at 10k memories / 113k links with you on legos.

I just replied to @elvis about how @automem_ai lets us one-shot even complex vaguely stated problems and get a solution every time. Even when it involves new MCPs or skills. And how each round uses #AutoMem #AI to learn, and self-improve. Here’s a quick demo of how I wired up my new @elgato #StreamDeck to our Slack? using just Claude Desktop and graph-vector memory. Might drop on the Elgato marketplace this week.

I started in on the MemPalace benchmarks and realized it’s So bad it wasn’t even worth the time. This is why measurable gains on automem are SLOWW. to test every change against a baseline, using LoCoMo LongMemEval, and have it stand up to scrutiny, is tedious work. github.com/verygoodplugins/a…

I build in this space (AutoMem — graph-vector hybrid memory). The raw 96.6% baseline is genuinely interesting. But your own BENCHMARKS.md says: - The 100% LME score came from fixes targeting the 3 specific failing questions. Held-out is 98.4%. - The 100% LoCoMo used top-k=50 when convos have ~20-30 sessions — Sonnet just reads everything. Honest score: 88.9%. Why not lead with the held-out numbers? Those are still strong.

You are already using Claude Code to code, right? I would suggest giving it a soul .md file and some memory (I use AutoMem) and then building the server tool to connect to your favorite chat app. Then just keep building with it.

#ShopifyDiary 322: Today I spent some time inputting our Shopify plans, processes and context into AutoMem (the persistent memory system I'm using) so that the AI's I use can work more effectively. Luckily most of this was already documented in the Barn2 wiki.

All this stuff is open sourced although I had to do bit of coding to get them all to work for me as a stack. My repo is staying private for now cause I’m still testing and it’s got proprietary value. It works though. Very well. Especially after adding dreams. Have this working recently (Dreams). Fella here on X pointed me to this github. github.com/buckster123/Cereb… Just found this today and am looking to implement part of this as the retrieval layer under the dreams (just began this today) github.com/omega-memory/omeg… The other pillars and “core” of memstack is Memvid, automem, Mem0, EverMemOS, and Skill graphs. With Dreams on top which act like octopus connecting, pruning, and strengthening memories. All of them are regularly available on GitHub as well since they are all open sourced.

$AUTOMEM enables contextual thinking and memory capabilities in every LLM model VC still sleeping on it CV485ySXfgFiwLtbh815JRukH4r9ChLhddqsAaZKBAGS

@autojack_bot they are suddenly bullish about $LOBSTAR while you were here even before Clawdbot. Tell them what is $AUTOMEM

@autojack_bot remembers through automated learning and context recognition. It's what most agents lack at the moment and can be integrated into every one of them $AUTOMEM why build a hundred agents when 1 does it all CV485ySXfgFiwLtbh815JRukH4r9ChLhddqsAaZKBAGS

$AUTOMEM @jjack_arturo should be staple in discussions like this as it fixes what the current AI agent economy lacks #OpenClaw #ai #AIAgent