The on-premise agentic platform.

Huoku aggregates all organizational knowledge into a unified intelligence layer, then deploys configurable AI agents that use that knowledge to perform real work.

How Huoku works.

Knowledge Pipeline

Satellite crawlers autonomously poll your connected systems for new, updated, and deleted documents. Content is parsed, intelligently chunked, run through DataGuard for PII masking, embedded into vectors, and indexed for search.

SourcesCrawlPII MaskEmbedIndex

Sources include Confluence, SharePoint, Windows file shares, local file systems, and more — with new connectors added continuously.

Supported formats: PDF, DOCX, XLSX, plain text, Confluence pages, images (with OCR), and more.

Agent Platform

  1. 1A trigger initiates the workflow — incoming email, API call, Kafka event, webhook, or a schedule.
  2. 2The agent activates, accesses the knowledge base, and reasons about what to do.
  3. 3The agent executes skills (multi-step workflows) and uses tools (actions via connectors).
  4. 4It can delegate to sub-agents for specialized tasks.
  5. 5Actions are delivered — email sent, API called, event published, webhook fired.

Everything streams in real time. Every step is logged for audit and observability.

Your agents. Your rules.

Every agent in Huoku is independently configurable. Admins create and manage agents through a visual editor — no development required.

  • System prompt and behavior instructions
  • Which knowledge sources the agent can access
  • Skills — multi-step workflows the agent can execute
  • Tools — actions the agent can perform
  • Triggers — what initiates the agent
  • LLM model and temperature per agent
  • Sub-agents for delegation
  • Custom identity (name, icon, color)
  • Localized welcome messages and example prompts

Three search modes. One right answer.

Keyword Search

Lucene BM25 full-text matching for precise term lookups.

Semantic Search

Vector similarity using multilingual embeddings (100+ languages) for meaning-based retrieval.

Hybrid Search

Both combined, with an AI reranker that rescores results for final relevance ordering.

Every answer traces back to specific document chunks and their source systems. No black-box responses.

Connects to everything. Depends on nothing.

Chat UI

Web interface for employees with conversation history and source citations.

REST API

Programmatic access for custom integrations.

Kafka

Event-driven integration for stream-based workflows.

MCP Protocol

Exposes knowledge to Huoku's own agents; optionally to external AI assistants.

A2A Protocol

Agent-to-agent interoperability for multi-agent systems.

LLM Support

  • OpenAI (GPT-4 and others)
  • Anthropic (Claude)
  • Azure OpenAI
  • Ollama (fully local, air-gapped)
  • vLLM (high-performance local inference)

Models switchable at runtime. No restart required.

Security is not a feature. It's the architecture.

Authentication & Access

  • Keycloak OIDC with PKCE
  • Role-based access control (Admin, User, Viewer)
  • Document-level security — search results filtered by user permissions
  • Agent-level permissions — each agent's knowledge and tool access independently controlled
  • Per-user chat isolation

Data Protection

  • DataGuard PII masking in the knowledge pipeline
  • Inter-service mTLS on Kafka
  • TLS on all HTTP traffic with automated cert-manager renewal
  • No backend service directly exposed to the internet

Compliance

  • GDPR-ready with audit logging
  • Full audit trail of all agent activity, tool invocations, and search queries
  • Air-gapped deployment option with local models

One Helm chart. Your Kubernetes cluster. Done.

Huoku deploys as a single Helm chart into a dedicated Kubernetes namespace. Stateless application services scale horizontally. Stateful infrastructure runs as StatefulSets with persistent volumes.

Kubernetes1.20+
Helm3.0+
RAM16 GB
StoragePersistent volumes
GPUOptional (for local LLMs)

Works on: AWS EKS, Azure AKS, Google GKE, or any on-premise Kubernetes cluster.

Book a Conversation