Control Plane Graph Intelligence

Atulya's Control Plane now helps teams understand a memory bank as a living system, not just a list of stored items.

The main question this feature answers is simple:

What changed, why does Atulya think that, and what should I inspect next?

This is useful for support teams, ops teams, knowledge teams, and anyone who needs to review evolving memory without manually tracing every raw fact.

What You Get

Graph Intelligence adds two connected views:

• State Graph shows the current state of important entities and topics.
• Evidence Graph shows the raw memories and links behind that state.

Together, they let a human move from:

1. a high-level summary
2. to a focused graph
3. to the exact evidence behind it

The Fast Mental Model

If you only remember one thing, remember this:

• State Graph tells you the story.
• Evidence Graph shows you the proof.
• The graph assistant helps you move between the two without getting lost.

Words We Use In This Page

To keep the docs easy to scan, we use the same words everywhere:

Term	Meaning
State	The current view of a topic or entity
Evidence	The raw memory items behind that state
Proof	The exact evidence path that supports an answer
Focus area	The small part of the graph you are reviewing right now
Filter	A way to show less noise (for example, anomaly-only nodes)

If you want to...	Start here	Why
Understand what changed	State Graph	It starts with meaning, not noise
Check whether something is still true	State Graph	It highlights changed, stale, and conflicting state
Verify the raw source	Evidence Graph	It shows the actual memories and links
Explain a conclusion to another person	Analyst answer + Evidence Graph	You get the short answer and the supporting path

Why This Replaces the Old Graph

The older raw node-link graph was useful for debugging, but primitive for real operational use.

It forced humans to read a dense network of low-level memories and infer the story themselves.

The new model is more useful because it starts with meaning:

• show important state instead of every raw memory first
• highlight changes, contradictions, and stale assumptions
• explain why Atulya believes something
• let the user open proof only when needed

The raw graph still exists, but it is now a supporting investigation tool instead of the default landing view.

The Two Main Views

State Graph

Use State Graph when you want to understand the bank quickly.

It emphasizes:

• what changed recently
• which entities or topics matter most
• where evidence conflicts
• which states may be stale

Each card is designed to be readable in plain language, with confidence, evidence count, and a short explanation of the current state.

Typical questions:

• What changed about this customer, repo, or workflow?
• Which belief is now outdated?
• Where do we have conflicting evidence?

State Graph helps with...	What the user sees
Recent change	A state card and change signal
Contradiction	Evidence that does not agree
Stale understanding	A belief that has not been refreshed while related signals moved
Priority	The most important entities or topics first

Evidence Graph

Use Evidence Graph when you need proof.

It shows the underlying memories, connections, and supporting paths behind a state or answer.

Typical questions:

• Which raw memories support this conclusion?
• Which document or chunk introduced this belief?
• How are these two memories connected?

Evidence Graph helps with...	What the user sees
Raw proof	Memory cards and their links
Provenance	Which memory, chunk, or document supports a conclusion
Connection tracing	How one memory led to another
Human review	The exact path behind a state or analyst answer

New Controls for Faster Human Review

The latest graph update focuses on simple controls that reduce clutter without hiding useful information.

Right-click node menu

Right-clicking a node opens quick actions. This lets users focus the graph without opening multiple panels first.

Surface	Right-click actions
State Graph node	Focus neighbors, filter to entity/topic, reset filters
Evidence Graph node	Focus node, show connected focus area, anomaly-only view, clear focus, reset filters

Anomaly-focused review

Evidence Graph now includes anomaly review controls so teams can quickly inspect risky areas.

Control	What it does
Filter Anomalies	Shows only nodes linked to anomaly events
Severity Overlay	Colors nodes by highest anomaly severity
Top anomaly type chips	Shows the most common anomaly categories in the current result

Link meaning improvements

Links are still grouped into semantic, temporal, entity, and causal, but causal links now keep their subtype meaning.

Link type	Color in UI
semantic	red
temporal	dark red
entity	amber
causes	violet
caused_by	indigo
enables	green
prevents	red

A Simple Human Workflow

The intended flow is:

1. Open a memory bank.
2. Start in State Graph to see the important changes.
3. Ask the graph assistant a question such as What changed about X?
4. Inspect the answer and the focused subgraph.
5. Open Evidence Graph when you want the raw proof.

This keeps the experience text-first and human-readable, while still keeping proof easy to verify.

Step	What Atulya is doing for the human
Open bank	Surfaces the most useful state first
Ask a question	Narrows the graph to the relevant part
Inspect answer	Explains what changed in plain language
Open evidence	Lets you verify the proof without losing context

Built for Large Banks

Graph Intelligence is designed to scale without turning the browser into a wall of cards.

For larger banks, the Control Plane uses progressive exploration instead of trying to render everything at once.

Detail Mode

For smaller focus areas, Atulya shows full cards and rich interaction.

Compact Mode

For medium-sized focus areas, cards become lighter and more condensed.

Overview Mode

For larger graphs, Atulya switches to an overview that summarizes the bank using:

• clusters
• bundled paths
• top-use memories
• top-use chunks
• focused area expansion

This is important because the goal is not "show every node." The goal is "help a human understand the bank quickly."

Graph size	What Atulya does	Why it helps
Small	Shows rich cards directly	Full detail is still easy to read
Medium	Uses lighter cards	Keeps the canvas readable
Large	Starts with an overview	Avoids turning the browser into a wall of boxes
Very large	Expands only the useful focus area	Keeps the experience fast and understandable

What the Analyst Layer Adds

The graph is not just a visual renderer.

Atulya also builds a read model that lifts raw memory evidence into higher-level state:

• stable state
• changed state
• contradictory state
• stale state

The graph assistant then returns:

• a short answer
• the focused nodes and edges
• change events
• an evidence path
• recommended checks

That is what makes the feature useful in industry: the graph is connected to explanation, not just drawing.

Analyst output	What it means in plain language
answer	The short human-readable conclusion
focal_node_ids / focal_edge_ids	Which part of the graph matters for this question
change_events	What changed, conflicted, or went stale
evidence_path	The proof trail behind the answer
recommended_checks	What a human may want to review next

API Surfaces Behind the UI

The Control Plane is backed by dedicated graph endpoints:

• GET /v1/default/banks/{bank_id}/graph/intelligence
• POST /v1/default/banks/{bank_id}/graph/investigate
• GET /v1/default/banks/{bank_id}/graph/summary
• GET /v1/default/banks/{bank_id}/graph/neighborhood
• POST /v1/default/banks/{bank_id}/anomaly/intelligence

The first two power the meaning-first workflow:

• build the state graph
• investigate a change or question

The second two support scaling:

• summarize large graphs
• expand a focused area without loading the whole bank into the canvas

Endpoint	What it is for
GET /graph/intelligence	Build the meaning-first state graph
POST /graph/investigate	Ask a question and get a focused answer
GET /graph/summary	Summarize a large graph into a readable overview
GET /graph/neighborhood	Load the focus area a human actually needs
POST /anomaly/intelligence	Fetch anomaly events and correction history for overlays and filters

Good Use Cases

This feature is especially useful when a memory bank represents:

• customer history
• product operations
• support cases
• engineering incidents
• personal knowledge that evolves over time
• shared team memory across tools and conversations

In each of these cases, a human usually does not want a primitive raw graph first.

They want:

• the story
• the proof
• the next thing to check

Real use case	Why Graph Intelligence helps
Support operations	See what changed about a customer or incident before reading every note
Product operations	Track shifts in issues, requests, and state over time
Engineering incidents	Move from a summary of change to the exact evidence trail
Shared team memory	Understand evolving knowledge without reading every raw memory
Personal knowledge banks	Review how beliefs and facts changed over time

Related Docs

• Services
• Installation
• Configuration
• Brain and Dream