Knowledge base
Corpus

Curate the knowledge available to every workflow.

Review authoritative sources, licensing and indexing status before content becomes available for retrieval. Human admission and passage-level provenance help teams control what workflows can use and trace responses to supporting material. Each engagement has its own corpus; the one open here is default.

Context and RAG engineeringhow context reaches an agent · click any stagerunning here 11 of 18specified
Tools and telemetry5 stages
Knowledge sources9 stages
Agentic retrievalwhere an agent reads
Two lanes, one meeting pointTelemetry lands as facts an agent can query. Knowledge lands as passages it can cite. Both pass through a person before an agent can reach them, and both reach the agent only through retrieval, which is where the guardrails sit.
Curated list of internal and external sources, chosen for the agents planned for deployment1 · running here
Vector ingestion pipeline with an agentic retrieval system2 · running here
Data security guardrails against access misuse, with policy-driven data access3 · specified
Predefined data parsers for known security tools4 · specified
Optional human in the loop to validate the accuracy of vector ingestion5 · running here
Hardened infrastructure as code templates to provision the target cloud environment6 · specified
Curated—sources across 8 families
Waiting for a person—fetched and split · not embedded, not searchable
Indexed—chunks retrievable now, from 0 admitted documents
Store—Embedding—Provider—Dimensions—Sources indexed—
Integrity: reading…every admitted chunk, with the current provider
Ask the corpus what an agent wouldSearch runs the retrieval stages and shows the passages: hybrid search over meaning and words, a cross-encoder rerank against the question, and the guardrails in the query. Answer runs the whole loop: it judges the evidence, rewrites and retries the query when the evidence is weak, answers from the passages only with every claim cited, then checks each cited sentence against the passage it cites.
Searching the default corpus
Curated sourcesreading…
SourceLicenceUsed forIndexedCadence
Tools and telemetry sourcesthe estate an agent asks about12 specified
WizWizCloud securityCloud findings, misconfiguration and attack paths
Microsoft DefenderEndpointEndpoint alerts, device posture and vulnerabilities
HackerOneHackerOneBug bountyExternally reported findings and their triage state
BitSightBitSightThird-party riskVendor security ratings and observed issues
ProofpointProofpointEmail securityMessage threats, campaigns and user reporting
AkamaiAkamaiEdge and WAFEdge events, blocked requests and bot traffic
InvictiInvictiApplication scanningWeb application findings with proof
DigiCertDigiCertCertificatesCertificate inventory and expiry
BeyondTrustBeyondTrustPrivileged accessPrivileged sessions and entitlements
CyberArkCyberArkSecrets and PAMVaulted credentials and session records
BrinqaBrinqaRisk aggregationConsolidated findings and risk scoring
QualysVulnerability scanningHost and application vulnerability detections
Admission recordevery decision about what the estate may believe

Nothing has been admitted or refused yet.