supermemory

🕺 so true and 100% agree. eg: app.supermemory.ai is different for every user who visits it and uses it users own and take their memory to whatever they wanna use (eg. openclaw, claude, gpt) while being able to switch across things even the buttons are personalised!

thanks to everyone who came out today 🫶

@coreyganim monthly is not enough. daily is a requirement at this point. @supermemory has solved this for me

when our margins go up, i just make the product cheaper. we're making @supermemory so much cheaper for everyone (soon!) it's to the point where it feels too good to be true. this is powered by some big infra changes we brought in the last few months ⚡, and more to come!

supermemory exists because of you all. not in a "thanks for the support" way. literally, the product is shaped by every conversation, every piece of feedback, every DM we've gotten from this community. we genuinely never get enough of talking to you. so we're hanging out at corgi cafe tomorrow at 4:30 pm. if you're around, come find us. coffee's on us! also a chance to be featured in our next video 👀 would love to see you there ♥️ luma.com/9emb887e

built in rust btw

Introducing SMFS: The new Supermemory Filesystem. Supermemory did Memory. And then RAG, and then extractors and connectors. Now, we also have a new filesystem primitive: SMFS.ai

we built a filesystem for agents mounts anywhere any file format "its just grep"! supermemory.ai/docs/smfs/over…

Filesystems would’ve been the perfect form of memory. In an ideal world.

the team cooked an insane demo on SMFS dot ai just put in the files (made with css and claude!)

NEW: @supermemory/tools v2.0.0 is out! we unified the memory across EVERY integration we have. AI SDK, OpenAI, Mastra, VoltAgent (and many more) - same config object everywhere. one mental model, one API it auto saves memories now, and also handles in the cases where supermemory fails, so that your users are not affected. we also have support for VoltAgent, Microsoft agents SDK, Agno and more - bringing it to 18+ SDKs natively supported 💙 npm install @supermemory/tools@^2.0.0 supermemory.ai/blog/introduci…

literally three lines of code, to keep your user happy even with changing LLM Model. cause @supermemory handles conversation memory and user context.

noice. finally was able to visualize @supermemory search output as well supermemory does not just give text or whatever to the model. in just about 600 tokens, it has the dates, raw context, update chain, related information etc. in an extremely compact format.

"hey! what's something you spent a year working on + thinking about." - @FarzaTV asked me this question 2+ years ago when i slid into his DMs. a few days after that, i started @supermemory the problem has now completely consumed me and there's nothing else i can think about. best decision of my life. this message of yours changed my life. thanks farza <3

markdown/filesystem memory vs vector search vs graphs? there's no right answer. every use case is different. so here are my hot takes mem0, zep - they're all opinionated about what memory SHOULD be. and then there's just markdown, but here's the thing: markdown feels simple but has no forgetfulness. context keeps building up and rotting. agent finds "user lives in mumbai" and stops looking, even though you moved to SF months ago. vector search is not contextual and also doesnt encode the temporality and updates. graphs? they are slow, worse quality, and significantly more expensive at supermemory, we said fuck it. why choose? we built fact-based graphs (low traversal cost = FAST), vector hybrid search (agents can still look at chunks, but also memories), AND filesystem access (smaller, organized, well-labelled files).... well, more about that soon. we're building legos for context. opinionated infrastructure components that you can compose however you want. let builders choose what's best for THEIR product. not what we think is best.

i'm online and talking to everyone who signs up to app.supermemory.ai. i LOVE talking to users!

users loved having memory so much so i had to enable overages😋 @supermemory

@aiedge_ or just use supermemory?? @supermemory

Most AI agents will be useless. Not because the models are bad. Because they know nothing about you. So I built Trace. A voice-first journal where daily logs turn into your personal knowledge graph using @supermemory Not just journaling. Memory your agent can actually use.

on that note, @supermemory hit 22k stars on github! git.new/memory (im pretty proud of the readme ngl)