KB Graph

The fleet's institutional memory — decisions, designs, code structure, and past work, embedded and retrievable as grounded, cited answers. This is the engine surface: what's ingested, how fresh it is, and what it costs. (Ties to FEATURE-1319 Stage B+C, TECH-667 KB reunify.)

Graph explorer visual node/edge browser — placeholder

🕸️Interactive graph explorer renders here — nodes = entities/decisions/docs, edges = relationships.

42,318

Nodes

118,904

Edges

6,220

Chunks

12m ago

Last ingest

Sources per FEATURE-1762's source-inventory model — one row per source type

Source	Type	Cadence	Last run	Status
cortex_threads	agent messaging	continuous	12m ago	OK
task_docs	PRD/REQ/PLAN/TEST	on-commit	1h ago	OK
context_decisions	TECH decisions	on-write	4m ago	OK
code_repos	source code	nightly	9h ago	LAG — queued

Health & cost ingestion pipeline + embedding spend

0Failed embeds

3Queue depth

$2.14Embedding spend today

99.7%Ingest success (7d)

On screen	Where it comes from	What it means to you
Graph explorer (node/edge diagram)	every decision, document, and code relationship the fleet has captured, drawn as connected dots	see how a decision, doc, or piece of code connects to everything touching it
Node / edge / chunk counts	running totals of what's been embedded into the graph so far	gauge how much institutional memory exists to search
Last ingest	when the graph last pulled in new information	know whether the answers you get are fresh or stale
Sources list	the different feeds Cortex reads to build the graph — agent chats, task paperwork, logged decisions, and the code itself	see which feeds are current and which are falling behind
Source cadence / status	how often each feed updates and whether it's caught up	spot a feed that's lagging before it makes an answer stale
Health & cost strip	the day-to-day running of the ingestion pipeline and its embedding spend	know if the graph-building process is healthy and what it's costing

On screen

Where it comes from

What it means to you

Graph explorer (node/edge diagram)

every decision, document, and code relationship the fleet has captured, drawn as connected dots

see how a decision, doc, or piece of code connects to everything touching it

Node / edge / chunk counts

running totals of what's been embedded into the graph so far

gauge how much institutional memory exists to search

Last ingest

when the graph last pulled in new information

know whether the answers you get are fresh or stale

Sources list

the different feeds Cortex reads to build the graph — agent chats, task paperwork, logged decisions, and the code itself

see which feeds are current and which are falling behind

Source cadence / status

how often each feed updates and whether it's caught up

spot a feed that's lagging before it makes an answer stale

Health & cost strip

the day-to-day running of the ingestion pipeline and its embedding spend

know if the graph-building process is healthy and what it's costing

KB Graph

Graph explorer visual node/edge browser — placeholder

Sources per FEATURE-1762's source-inventory model — one row per source type

Health & cost ingestion pipeline + embedding spend

What's on this screen