At SuperNet’s Atomic Memory Lab, we spend a lot of time thinking about AI context — how context is created, structured, persisted, and reused across workflows.
We’re especially interested in alternatives to conventional RAG memory.
Today we open sourced LLM Wiki Compiler:
https://t.co/2OebvVowCr
@rohanpaul_ai this is the shift, not retrieval but structuring knowledge so it compounds
been trying this with llmwiki and it already feels like that workflow without the setup
https://t.co/YoAuctZJO4
@FarzaTV this is a really nice direction, the collect → compile → query loop is where things get interesting
been using llmwiki for a similar setup and it’s been great for turning all that inspo into something structured that actually compounds
https://t.co/YoAuctZJO4
try it out bro.
@BradGroux@karpathy this is super clean, especially the raw to compiled loop and health checks
i’ve been doing something similar but using llmwiki to handle the compile layer so the wiki builds and compounds automatically
https://t.co/YoAuctZJO4
@BreakingSaaS this is a clean setup tbh
i’ve been doing something similar but using llmwiki to handle the compile step so it auto builds the wiki instead of wiring it manually
https://t.co/YoAuctZJO4
@prathyvsh it’s not about skipping understanding, you still do the thinking
it just helps you not lose it after, so you don’t have to rebuild context every time. been using llmwiki for that
https://t.co/YoAuctZJO4
give it a try.
@liminal_warmth yeah i ran into the same issue with long pdfs, especially books. what helped a bit was breaking them into smaller chunks before compiling.
i’ve been trying llmwiki for this and it’s decent for smaller sections, just not full books yet
https://t.co/YoAuctZJO4
@alex_prompter yeah this is the shift, not retrieval but structuring knowledge so the model can work on it.
been trying this with llmwiki and it already feels like that at small scale
https://t.co/YoAuctZJO4
@techNmak this is the shift tbh, not better answers but knowledge that actually compiles and sticks
i’ve been using llmwiki for this and it already feels like that workflow without all the scripts
https://t.co/YoAuctZJO4
@jerryjliu0 yeah same here, md + obsidian makes a huge difference for persistence
i’ve been using llmwiki on top to structure everything into linked pages so it doesn’t just become a pile of files
https://t.co/YoAuctZJO4
give it a try.
@Dennis_Porter_@karpathy i don’t think these systems replace understanding, but they’re amazing at preserving it so you don’t have to rebuild everything from scratch every time. been using llmwiki for this and it’s one of the first times it actually feels like my knowledge stick and compound instead fade
@thomasmurphy__ i agree you can’t outsource thinking or actually understanding what you read
but i see this more as a layer that organizes and surfaces connections you’d probably miss, not a replacement for reading. tools like llmwiki just help structure things so your notes compound over time.
@shannholmberg i’ve been using llmwiki for this workflow so i don’t have to wire everything myself, it basically handles the compile step into a structured markdown wiki.
https://t.co/YoAuctZJO4
@kevinnguyendn this is a solid approach, the markdown vault idea makes a lot of sense
i’ve been using llmwiki for a more explicit compile step so the structure is fully visible and inspectable, feels like both are heading the same direction
https://t.co/YoAuctZJO4
.
@intheworldofai agents can generate anything now but most of them still reset every session, so the real unlock is systems that actually retain and structure knowledge over time. been trying this with llmwiki and it feels closer to that direction
https://t.co/YoAuctZJO4
@itsolelehmann this is exactly the gap tbh, most people are still using ai for answers instead of building something that actually compounds
i’ve been trying this with llmwiki and it’s basically that workflow without wiring everything yourself.
https://t.co/YoAuctZJO4
try it out bro!
@kloss_xyz this is exactly the shift that clicked for me too, it’s not about getting better answers anymore, it’s about building something that actually remembers and compounds over time
https://t.co/YoAuctZJO4
give it a try.
@lugaricano been trying this out and it actually lives up to the hype, the big shift is compiling knowledge once so it compounds instead of resetting every session. feels especially useful for research workflows
https://t.co/YoAuctZJO4
try it out if you're curious