Enterprise Data Intelligence - Multi-Agent Enterprise Knowledge Assistant#
An agent-native automation platform that combines a local knowledge base with autonomous agents to complete real enterprise tasks end-to-end. It wires together a UI service, knowledge base retrieval (the Data Intelligence Engine), an LLM router with prompt compression, and an OpenClaw agent runtime — agents run reusable Skills that query the knowledge base and produce professional deliverables (e.g., competitive-analysis reports).
Features#
Multi-agent task automation — OpenClaw runs a main agent that can spawn sub-agents to complete multi-step enterprise tasks without manual intervention.
Local knowledge base retrieval — The Data Intelligence Engine indexes uploaded documents in a Milvus vector store and returns the most relevant chunks through embedding and retrieval; the agent writes grounded answers from them.
Hybrid model routing with cost savings — The Router and compressor front local (vLLM) and cloud models (for example, MiniMax), shrinking prompts before dispatch to cut token usage and latency.
Token consumption monitoring — The UI shows a live dashboard of local versus cloud token usage, latency, and tokens saved by compression.
Reusable Skills — Agents load self-contained Skills, such as the competitive-analysis report generator, to produce professional deliverables like HTML and PDF reports.
Conversational UI — A chat interface with slash commands (for example,
/model,/session,/tool) and streaming responses for controlling agents and models mid-conversation.Multi-language support — The UI and generated reports support multiple languages, including English and Chinese.
Skills#
Skill |
Description |
Status |
|---|---|---|
|
Competitive-analysis report generator — gathers product info from the local RAG knowledge base plus web search, then produces a professional Chinese HTML/PDF comparison report |
Shipped ( |
|
Generic RAG query skill — retrieves any information from the local knowledge base (Data Intelligence Engine) via a curl-based |
Shipped ( |
See the skills folder for the shipped Skills and the Get Started guide for how to install and enable a Skill in OpenClaw.
Architecture#
The platform is built around a UI service that talks to the OpenClaw agent runtime. OpenClaw orchestrates work through Skills. A Skill retrieves grounded facts from the knowledge base served by the Data Intelligence Engine, and the agent model writes the answer from them, while an LLM router (with a prompt compressor) fronts local and cloud models.
Components#
UI — browser-based front end for sending tasks to OpenClaw and viewing generated results.
Router + compressor — LLM router that fronts local (vLLM) and cloud models, with a LinguaCompressor front end that shrinks prompts before dispatch.
Data Intelligence Engine — knowledge base service: embedding + retrieval over an uploadable knowledge base (Milvus vector store). It returns the retrieved chunks to the Skills and does not generate answers.
vLLM — local LLM serving (
Qwen/Qwen3.5-35B-A3B) used by OpenClaw and the router.OpenClaw — the agent runtime that loads Skills, calls models via the router, and executes tasks (web search, RAG query, PDF generation).
Skills — reusable, self-contained task recipes under the skills folder that agents load at runtime.
Additional Resources#
Inference Router microservice — the microservice that implements the Router and compressor.
Data Intelligence Engine microservice — the microservice that serves the knowledge base.
Get Started guide — step-by-step instructions for setting up the platform and running the demo.
Release Notes — a changelog of updates and improvements to the platform.