Architecture & governance

Review how iKB connects data sources, retrieves evidence, applies policies and delivers answers. Each stage contributes to a shared audit trail.

iKB reference architectureLogical flow / configuration varies by deployment
Enterprise sourcesDocumentsStorageDatabasesWebsites & APIs
01

Data & ingestion

Connectors, processing, and indexingControl details

Capture metadata, content hashes, versions, and provenance. Keep document storage and indexes inside the selected deployment boundary.

02

Evidence & retrieval

Hybrid search, anchors, and GraphRAGControl details

Combine exact-phrase retrieval, reranking, cross-lingual support, and graph evidence in a ranked pool. Verify grounding against the retrieved material.

03

Policy & models

Permissions, guardrails, and routingControl details

Apply topic policies, prompt versions, moderation, DLP, redaction, and retention. Route to approved hosted providers or local model endpoints.

04

Answers & delivery

Workspaces, channels, and APIControl details

Deliver controlled answers through web chat, Slack, Teams, WhatsApp, Telegram, and API, with voice and document generation where enabled.

Continuous audit & evidenceSources / policy events / model usage / outcomes / exports
The evidence record

Understand what happened.
Keep the record for review.

The audit trail describes the system's actions and supporting evidence without exposing a model's hidden reasoning.

01 / Sources

What supported the answer?

Source documents, citations, provenance, retrieved evidence, and the knowledge domain queried.

02 / Controls

Which policies applied?

Prompt and configuration versions, triggered rules, and whether an external action was allowed, modified, or blocked.

03 / Operations

How did the system perform?

Stage timing, model selection, token usage, cost, and outcomes in append-only, hash-chained events.

Deployment boundaries

Place the system
where your data belongs.

Local storage, indexing, retrieval, relational and semantic data layers, caching, and background processing support operation within your own infrastructure.

Compare deployment paths
01

Hosted

A starting point for pilots and standard enterprise knowledge use cases.

02

Private cloud

Dedicated infrastructure and environment isolation for sensitive programmes.

03

On-premise

Run close to internal systems with local inference, an operator CLI, and 18-signal health probes.

04

Air-gapped

Keep the core knowledge experience local with locally hosted models. Live web research and externally hosted services are unavailable inside a closed boundary.

Bring IT, security, and risk to the same table.

Review the data flows, model endpoints, evidence records, and operating boundary before committing to deployment.

Talk to our team