Blog
Benchmarks, engineering notes, and the thinking behind giving agents a filesystem instead of a vector database.
Guide · July 28, 2026
Add AI to your Fumadocs docs
Give your documentation an agent and your readers can ask it questions instead of hunting for the right page. It answers by reading your docs with ls, find, cat, and grep, and links the pages it used. Two files, no vector database.
5 min readRead →
One question, three readsNo embeddings · no index to rebuild
grep
find
cat
Perspective · July 24, 2026
Everyone is building a filesystem for their agent
Mintlify, Anthropic, and Turso each built a filesystem for their agents, independently. Here is why the same primitive keeps reappearing, and why you should not build it yourself.
6 min readRead →
Three backends, one interfacels · find · cat · grep
Benchmark · July 23, 2026
Agent filesystem beats naive RAG on retrieval accuracy
We put a grep-over-filesystem agent against top-k and agentic RAG on an adversarial 18,340-file corpus. It answered 97.9% of held-out questions correctly, 16.7 points ahead of naive RAG. Here is how we ran it.
5 min readRead →
Retrieval accuracy+16.7 pts vs naive RAG · Vela corpus
grep-agent97.9%
agentic-RAG90.6%
naive-RAG81.3%