📖

KB

The tree that remembers everything you tell it.

Tell it things. Ask it things. One person maintains, everyone benefits. The tree organizes knowledge into a hierarchy. The AI answers from stored notes with citations. The coworker who never forgets and never gets interrupted.

Two actions. Tell and ask.

1

Tell it what you know

~ > kb "Row A has servers 1-12. Servers 1-6 are Dell R740.
Servers 7-12 are R760. All on circuit breaker panel 3A."

Filed under Server Rack Layout / Row A.
Created note with hardware specs and power info.

The AI parsed the input. Detected it's about server racks. Found or created the right branch. Wrote a note. One LLM call. If it can't categorize, it files to Unplaced.

2

Ask it anything

~ > kb "what do I do if the temperature alarm goes off?"

Check the HVAC panel on the east wall. If ambient is above
85F, call facilities at ext 4401. If above 95F, begin
emergency shutdown starting with non-critical servers in Row D.

[Source: "Temperature Thresholds" on Alert Procedures/Temperature, updated 3 weeks ago]

The AI answers from the notes in the tree. It cites which note and which branch the answer came from. The source is always visible. The knowledge is always auditable.

3

Correct and update naturally

~ > kb "actually servers 7-12 in Row A are R760 not R750"

Found existing note on Server Rack Layout / Row A.
Updated: R750 to R760 for servers 7-12.

The AI reads existing notes before writing. If the new info contradicts or updates something, it modifies rather than duplicates. The old version stays in history.

The tree grows from what you tell it.

Not from a template. Not from a schema. You say things. The AI organizes them. The notes are human-editable. The structure is human-navigable. The AI answers from both.

Datacenter Ops
├── Log (where you talk)
├── Topics
│ ├── Server Rack Layout
│ │ ├── Row A (Dell R740/R760, panel 3A)
│ │ ├── Row B
│ │ └── Cable Management
│ ├── Alert Procedures
│ │ ├── Temperature (thresholds, escalation)
│ │ ├── Power Failure (UPS sequence, generator)
│ │ └── Network Down
│ ├── Vendor Contacts
│ └── Onboarding
│ ├── First Week (badge, access, parking)
│ └── Safety
├── Unplaced (can't categorize yet)
├── Profile (maintainers, access)
└── History (changes log)

It knows what's getting old.

Notes track when they were last updated. The AI flags anything over 90 days. Type be at the kb and it walks you through stale notes one by one. "Vendor Contacts / Cisco hasn't been touched in 6 months. Still current?"

Stale Notes
Vendor Contacts / Cisco
180 days since last update
Onboarding / Systems Access
95 days since last update

One maintainer. Everyone benefits.

Maintainers tell the kb things. Everyone else asks. A new employee joins the land, gets contributor access, types "what do I do on my first day?" The AI reads Onboarding/First Week. Answers with badge pickup, access requests, parking info. The employee never bothered a coworker.

Part of something bigger.

The first four are personal. KB is the first one that scales to teams.

🍎
Food
Fuels the body
💪
Fitness
Builds the body
🌿
Recovery
Heals the body
📚
Study
Builds the mind
📖
KB
Builds the team

Under the hood.

Three AI modes

Tell parses statements into knowledge. Finds or creates the right topic branch. Updates existing notes when corrections arrive. Ask searches the tree, reads matching notes, assembles answers with citations. Uses scout and embed if available. Review the guidedMode for be. Walks stale notes. Presents each one. Asks if it's current.

Intelligence integration

Understanding compresses branches bottom-up. The AI reads encodings to know what's in each branch without reading every note. Tree-compress consolidates when branches get too dense. Contradiction detects conflicting notes. Purpose checks coherence against the kb's topic. The kb extension doesn't implement any of this. It installs alongside the intelligence bundle and gets it for free.

Scales through the tree

500 notes across 200 nodes. 5,000 notes across 2,000. The tree handles its own scale. Understanding creates one-sentence encodings per branch. The AI reads the encoding to know what's there, then dives into the specific branch that matches the question. No vector database needed. The tree IS the index.

Commands

kb "statement"Tell the kb something new
kb "question"Ask the kb something
kb statusCoverage and freshness
kb staleNotes needing review
kb unplacedUncategorized items
kb reviewGuided review of stale notes
beStart guided review mode

What does your team need to know?

Plant a land. Create a KB tree. Start telling it things. Everyone else just asks.