Knowledge Assistant
Full capability detail.
Every capability that ships with Knowledge Assistant — from document ingestion and vector database choice through to layered content safety and air-gapped deployment. Reference for technical evaluators and procurement.
Knowledge & Quality
Document Ingestion & Knowledge Management
Drag-and-drop upload (PDF, Word, Excel, PowerPoint, Markdown, HTML, plain text). Bulk ingestion from folders. Re-index on demand. Admins see exactly what's in the knowledge base and can remove documents at any time.
Choice of Vector Database
Choose LanceDB (lightweight, file-based), PostgreSQL with pgvector (reuse your existing managed Postgres, single backup story, SQL-native), or Chroma (purpose-built). All run inside your environment; the knowledge index never leaves your network.
Intelligence
Choice of Intelligence Provider
Self-hosted (LM Studio, Ollama, llama.cpp — no data leaves your infrastructure); hyperscaler (Azure OpenAI / AI Foundry, AWS Bedrock, Google Vertex AI); frontier APIs (OpenAI, Anthropic, Mistral). Mix and match across workspaces.
Kubernetes-Native, Built to Scale
The inference backend runs on Kubernetes — language models, embedding models, and content-safety models are independent services that scale horizontally. Add GPU nodes for faster responses or more concurrent users; the platform load-balances automatically. Idle nodes scale down to control cost. Works on AKS, EKS, GKE, OpenShift, or vanilla K3s on your own hardware.
Identity & Access
SSO
Enterprise SSO via OIDC / SAML — Microsoft Entra ID (Azure AD), Okta, Google Workspace, Ping, Auth0, ADFS. Group-based access control: map your existing security groups to user and admin roles per workspace.
Bring Your Own Domain & Certificates
Host on your internal domain (e.g. assistant.yourcompany.com). Upload your own TLS certificates — wildcard, internal CA, or public. CSR helper included for IT teams requesting from internal PKI.
Experience
Responsive Web UI
Modern React chat interface. Per-user conversation history, message-level feedback (thumbs up/down with comments), saved conversations, user preferences, light/dark themes, mobile-ready.
Multi-Workspace
Run multiple isolated assistants from one deployment — one per team, department, or knowledge domain. Each workspace has its own documents, access groups, and conversation history. No cross-contamination.
Cited, Not Invented
Every answer points back to the source document and page. Reduces hallucination risk and gives users a path to verify.
Safety & Governance
Layered Content Safety
Three tiers, configurable per workspace: baseline filters (profanity, prompt-injection, PII detection); LLM-powered moderation (a dedicated safety model screens prompts and responses); external services (Azure Content Safety, AWS Bedrock Guardrails, or your existing DLP).
Audit Logs & Risk Alerts
Every interaction logged: who asked what, what the model returned, which sources were cited. Alerts on flagged content, jailbreak attempts, or policy violations. Admins can review, export, and delete sessions for compliance and Subject Access Requests.
Centralised Logs
Built-in log aggregation (Loki + Grafana) and live container log viewer (Dozzle) for ops teams. Plugs into your existing SIEM via standard syslog/JSON output.
Secure by Default
Network-isolated services, no outbound calls unless you configure them, secrets managed via your KMS, regular dependency scanning. Deployable air-gapped if required.
Bring a use case. We'll show you the deployment.
A 30-minute demo using your own example documents. We bring a working environment; you bring a folder of representative content. By the end you'll know whether Knowledge Assistant fits.