Skip to content

Stories

L0 · RawL1 · ReadableL2 · AnalysisL3 · Post-itsL4 · StoriesL5 · Memory

Post-its cluster. Stories cross-link.

This is what makes the library think like human memory does.

A story is where the dots get joined. Many post-its from many files, gathered around one casefile. Many casefiles, gathered into one narrative on top.

L4 — A story per casefile

LAYER 4 · STORIES

Many post-its, one woven narrative.

A casefile collects all the post-its tied to one thing — an engagement, a vessel, a person, a project. The story-builder weaves those post-its into a single readable story.md on disk. It’s rewritten as new post-its arrive — not appended.

How the dots get joined

ca
law
coach
found
arch
ca
law
found
arch
ca
coach
found
casefile A
casefile B
casefile C
all active casefiles

A story — anatomy

One casefile
An engagement, a vessel, a person, a project. The thing the story is about.
Many post-its
Every post-it tied to the casefile, across every perspective.
Woven
A story-builder synthesises post-its into a single readable narrative.
Rewritten
When enough new post-its arrive, the story is rewritten — not appended.
On disk
story.md lives in the casefile folder. The DB stores a hash.
Versioned
Old versions are archived. The story arc itself is recoverable.

L5 — One final story on top

LAYER 5 · MEMORY

”Where are we right now?”

The final story weaves every active casefile’s story into one calm narrative. It is rewritten daily — or whenever something major lands. When an agent wakes up, the final story is loaded first. That is how an agent knows the state of everything before it answers a single question.

Each casefile

has its own story.md. Updated as new post-its arrive.

Daily routine

A higher-quality model reads every active story.

One narrative

Stories woven into final-story.md. The top of the stack.

Cold-start inject

Every agent reads the final story on its way in.

Ready to work

The agent already knows the state of the office. No human briefing needed.

Why this is closer to human memory

Humans do not recall facts as isolated rows. We recall stories. A name reminds us of a place, which reminds us of a meeting, which reminds us of a decision. Many things, thought of together, cross-linked.

A pile of vector embeddings cannot do that. Stories can. The library is built to think the way the brain does.

Every claim resolves to a raw byte

If the final story makes a statement, you can walk down the chain until you reach the original byte.

final-story.md makes a claimL5 · “vendor invoice acknowledged, payment scheduled”
story.md cites the post-itL4 · note_paths: [postit:12345]
post-it carries the source URLL3 · source_url: file:///inbox/2026-05-20-vendor.eml
readable .md points to the raw fileL1 · 2026-05-20-vendor.md (sidecar)
raw byte — ground truthL0 · 2026-05-20-vendor.eml — bytes preserved as they arrived

No floating facts. No invented citations. Every claim has a thread back to a real file.

story.mdfinal-story.mdcross-linkingcold-start injectprovenancefile-canonical