Thoughts on writing.
Context compounds. Make sure you own it.
Models are becoming a commodity - the lasting asset is the accumulated context that tells them what to do. Which raises an uncomfortable question - if your company's context is the most defensible thing you'll build, why is it living inside someone else's product?
RAG is a patch, not a foundation
Retrieval was supposed to connect AI to your company's knowledge. In practice it bolts a search engine onto a language model and hopes the top results are enough. What agents actually need is not better retrieval - it's a source of truth they can read and write.
Smarter models won't fix your AI
The industry keeps shipping bigger models, and your team keeps getting confident nonsense. That's because the limiting factor was never intelligence - it's that your AI knows nothing about your company. Here's the gap, and its four symptoms.
Per-folder access, enforced on the server
Most knowledge tools hide what you can't see in the UI - the data is still on your disk. Monora's access check runs on the server, on every fetch and push. A folder you're not authorized for is absent from your clone and its history, not greyed out.
Why Monora: the missing middle between Drive and git
Drive has per-folder permissions but breaks the agent loop. Git has the agent loop but is all-or-nothing per repo. Monora is the missing middle - a real git tree underneath, with per-folder access enforced server-side.