How iKB answers.
How you stay in control.
Retrieval-augmented generation (RAG) is how iKB works under the hood: it finds relevant source material first, then writes the AI answer. See how iKB searches your knowledge, applies access and model rules, and keeps a record your auditors can review.
01Sources & processing
Connect, process and index approved informationView details
Add documents, JSON data records and timestamped audio transcripts, including video audio tracks. Import supported public shared files when enabled, or use authenticated connections for private sources. Public-link imports check access, file type, size and allowed destinations before processing. Keep versions, data types and links between files and their attachments. The current version stays searchable during reprocessing. Each connector action needs topic approval.
02Search & evidence
Find relevant passages and keep referencesView details
Search exact phrases and related meanings, rank results and use supported cross-language search. GraphRAG adds relationships between facts. Keep document evidence separate from table calculations. Configured Research adds web and literature sources, with outgoing-query privacy checks. Evidence checks can monitor results or filter irrelevant material while protecting required evidence. Flash mode uses a focused search-and-answer process.
03Policy & models
Access, privacy and model selectionView details
Apply access rules, approved prompts, moderation, privacy checks and retention. Choose models from Efficient, Balanced and Advanced groups. Configured decision models can classify tasks and fall back if needed. Flash topics skip these decision tasks. Strict-PII topics apply stricter personal-data rules and exclude remote decision models.
04Answers & delivery
Conversations, Studio, channels and APIView details
Use native chat, the widget, Slack, Teams, Chatwoot or the API. Enabled native Analytics supports saved dashboards shared through topic access. Raw-row access and SQL editing have separate permissions; review detailed rows before sharing. Configured Chatwoot channels can hand over to an agent. Studio saves output versions and supports branded exports when configured. Interactive cards are opt-in for native chat; the embedded widget keeps plain answers. WhatsApp and Telegram are also messaging options.
Check the answer.
Keep the supporting record.
Decision Records keep the sources, policy checks and model use behind each answer. They cover native chat, the widget, Slack, Teams, Chatwoot and API. Export JSON or PDF packages with integrity checks and access and retention controls. These records do not expose hidden model reasoning.
What supported the answer?
Source documents, citations, transcript timestamps, structured records and calculated results. Keep source details and the topic that was searched.
Which policies applied?
Prompt and configuration versions, privacy checks, approved connector actions, and whether a decision task was accepted, skipped or used a fallback.
How did the system perform?
Processing times, model choices and results. Audit events are append-only and hash-chained to help detect changes. A stored usage ledger records costs and explains missing price information.
Choose where data
is stored and processed.
Run storage, indexing, document search, databases, caching and background jobs on your own infrastructure when required. Review hosted models and external services separately.
Compare deployment optionsHosted
A managed starting point for pilots and agreed business uses.
Private cloud
A dedicated environment for your organisation. Review which model endpoints and external services are enabled.
On-premise
Run core services and supported models on your infrastructure. Operators have a command-line interface and 18-signal health probes for system checks. Installation, Docker and recovery workflows support maintenance. Shared capacity limits control heavy workloads.
Air-gapped
Air-gapped means isolated from external networks. Core knowledge search and answering use local services and models. Live web research and externally hosted services are unavailable.
Questions for IT and risk teams.
What does AI governance mean in iKB?
How is RAG different from fine-tuning?
Does private deployment mean no data leaves our network?
What should we test before production?
Review your security and IT requirements.
Check data flows, model endpoints, access rules and audit records with your technical team before deployment.