Enterprise Data Intelligence#

GitHub Readme

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, retrieval-augmented generation (EC-RAG in this case), 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 — EC-RAG indexes uploaded documents in a Milvus vector store and serves grounded answers through embedding, reranking, and vLLM-based generation.

  • 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_PDF_generator

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 (SKILL.md + query_rag.sh)

knowledgebase

Generic RAG query skill — retrieves any information from the local EC-RAG knowledge base via a curl-based ecrag wrapper and generates structured reports, summaries, comparisons, or Q&A responses

Shipped (SKILL.md + ecrag)

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 EC-RAG knowledge base, while an LLM router (with a prompt compressor) fronts local and cloud models.

                 user
                  │
                  ▼
          ┌───────────────┐
          │      UI       │  :7000
          └───────┬───────┘
                  │
                  ▼
          ┌───────────────┐        Skills (competitive_analysis_PDF_generator)
          │   OpenClaw    │  :18789
          │   agent       │ ◄──────────────┐
          └───┬───────┬───┘                │ query_rag.sh
              │       │                     ▼
   model calls│       │            ┌────────────────┐
              ▼       │            │  EC-RAG        │  :16011
      ┌──────────────┐│            │  (retrieval +  │
      │  Router +    ││            │   vLLM answer) │
      │  compressor  ││ :8000/:8001└────────────────┘
      └──────┬───────┘│
             ▼        ▼
      local vLLM   cloud models
      :8086        (MiniMax, …)

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.

  • EC-RAG — Edge Craft RAG: embedding + reranker + vLLM answer generation over an uploadable knowledge base (Milvus vector store).

  • 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#