Knowledge Health · Agent Context · Graph

Check the work before you rely on it.

Three surfaces answer the questions that decide whether you trust an answer: what is indexed, what an agent would actually retrieve, and how your knowledge connects.

Start free Read the docs

Atlas Launch · Knowledge HealthHealthy

Indexed sources

AI connections

Successful retrievals

Indexing and imports

Customer interviews.pdfPDFIndexed

Reliability review — Q3Web pageIndexed

Pricing teardown.pptxDeckQueued

Vendor SLA (archived)PDFStale

Imports remain available as metadata and lexical results even when semantic indexing is paused.

Authorized connections

Claude DesktopOAuthRead

Research assistantOAuthRead

knowledge:readclaims:read

An illustration of the Knowledge Health view, using fictional data.

What you can do

Three views, one question: can I trust this answer?

Indexed sources

See how much of a Library is semantically indexed, what is still queued, and what failed to import.

Authorized access

Every connection with access to this Library, and the exact scopes it holds, on one screen.

Successful retrievals

A running count of retrievals over the last seven days, so you can tell whether your setup is actually being used.

Agent Context

Run a query and inspect the exact chunks, memories, graph links, and sources an agent would receive for it.

Explicit warnings

When something in scope is stale, unindexed, or failed, the context view says so rather than quietly returning less.

Knowledge Graph

Explore how items connect, inspect a node's relationships, and edit the links that shape retrieval.

Imports always remain available as metadata and lexical results, even when semantic indexing is paused.

Before the answer

See the context an agent would actually get.

Agent Context is a dry run. Type the question you were going to ask elsewhere and Prism shows the retrieval trace: which chunks matched semantically, what the graph pulled in, which durable memories applied, and what it had to warn you about.

Choose an essentials, standard, or deep context profile.
Every block is attributed to the source it came from.
Warnings surface stale or unindexed material in scope.

Atlas Launch · Agent ContextInspect

Retrieval trace

Semantic matches

6 chunks above threshold across 4 sources

Step 1

Graph expansion

2 linked artifacts pulled in via relationships

Step 2

Workspace memory

3 durable observations applied

Step 3

Warnings

One source in scope has not finished indexing, so it contributed lexical matches only.

How it connects

Relationships are first-class, not inferred each time.

The Knowledge Graph shows how sources, notes, artifacts, folders, and companies relate. Links are explicit records you can inspect and edit — and authorized clients can propose them, subject to the scopes you granted.