Office Library
A library every agent reads.
Six layers, bottom to top. Files canonical. URL carried at every layer. The final story is what the agent reads on its way in.
The context window of the agent is not memory. A million tokens is not enough. Agents work for months on the same files, the same people, the same casefiles. They need an office — files, folders, post-its, stories — that any agent can walk into.
The six layers — read bottom-up
Raw files — ground truth
Bytes on disk. Never moved. Never rewritten.
Readable
Convert once, read forever. Every readable file points back to its raw byte.
Analysis · optional
Structured extraction on top of the readable file.
Post-its
Many perspectives. Each writes one short note from its lens.
Stories
Post-its cluster per casefile. Stories cross-link.
Memory
What the agent reads on its way in.
source_url back to the readable file; the readable file points to the raw byte. Every claim resolves to a byte.How agents come to work
An employee walks into the office. They open the casefile. They glance at the post-its from yesterday. They read the story so far. Then they work.
That is exactly what an agent does every morning. The library is what makes it possible across months and years — not just one session.
Why this is the moat
Most memory frameworks extract facts and embed them as vectors. The fact gets stored. The file that produced it is forgotten. We do it the other way around.
vs Mem0
vs Letta
vs Zep