Three knowledge layers
Controls information density and token use through one-line, structured-summary, and full-text levels.
SOURCE LINK UNAVAILABLE · PROJECT NOTE
The design splits documents into L0, L1, and L2 information layers so an agent can scan summaries before drilling into full text, with session memory retaining useful context.
Controls information density and token use through one-line, structured-summary, and full-text levels.
Organizes documents with ctx:// paths and combines semantic and keyword search for progressive drill-down.
The design includes extracting memory from sessions and connecting agent clients through MCP.
Useful for readers studying how agents may browse knowledge by information density and combine retrieval with long-term memory.
The former repository is unavailable, so this page is a design note only. It cannot establish that code is installable, runnable, or maintained.
The former GitHub URL returned 404 on 2026-09-23. The page no longer says that public source is available.
The available evidence is first-party: deployment records, product documentation, and endpoint status at the review time. No independent evaluation was available. Reachability confirms access at that moment, not performance, accuracy, savings, or long-term availability.