Enterprise data engine for AI applications

Hermes Notebook

Turns enterprise information scattered across PDFs, Word, Excel, PowerPoint, images, local folders, and business web pages into citable, retrievable knowledge that can support customer-service bots, internal Q&A, and other business assistants through APIs.

Deployable system | makes enterprise data retrievable and citable by AI

A company's most valuable information is scattered across folders, Office documents, PDFs, images, and business web pages. Employees struggle to find it, while separate AI applications repeatedly prepare the same material
Hermes imports multiple formats, restores document structure, builds traceable knowledge nodes and retrieval indexes, and supplies evidence to downstream applications through permission-scoped query and retrieval APIs
Customers receive an enterprise knowledge engine they control. Customer-service bots, internal Q&A, and future business assistants can reuse the same prepared data instead of cleaning and importing it again for every AI application

The project currently supports local Docker Compose deployment, file, folder, and URL ingestion, multi-format preparation, traceable retrieval, citation-backed Q&A, and retrieval and query APIs restricted by tokens and knowledge-base scope.

Previous workflow

Enterprise information was spread across local folders, PDFs, Word, Excel, PowerPoint, images, and business web pages maintained by different teams and employees.
Employees relied on filenames, folders, and personal memory to find information, then manually checked which document supported an answer.
Each new customer-service bot or internal assistant required similar information to be collected, cleaned, and imported again, with little reuse between applications.

What the system takes over

Extract text, headings, tables, and image text by format while preserving the relationship between each passage and its original location.
Write prepared content into the relevant knowledge base and build retrievable knowledge nodes, source records, and completeness states.
When an application queries the system, Hermes first checks its token, API scope, and allowed knowledge bases, then returns evidence passages or generates a cited answer from that evidence.

What remains human

Administrators decide which folders and sources enter each knowledge base and review failed imports, duplicates, and source completeness.
Business owners confirm whether retrieved evidence and citations support the answer and decide which knowledge bases each application may use.
Customer-service wording, external responses, and business actions remain subject to review in the relevant application and by responsible staff.

Delivered outputs

Enterprise knowledge bases organized by business topic with sources, original locations, completeness, and processing records
Evidence retrieval results ready for application use and knowledge answers with source citations
Retrieval and Q&A APIs for customer-service bots, internal knowledge Q&A, and other business assistants

Usage scope

Hermes provides the enterprise knowledge preparation and retrieval engine; customer-service bots, WeCom or Feishu integration, interfaces, and business actions belong to downstream applications.
When source information is missing, outdated, or poorly scanned, the system cannot reconstruct facts and the enterprise must correct the source material.
Enterprise group and department identity cannot rely on model judgment and must be explicitly mapped to allowed knowledge bases by the integration layer.

Your business does not need to copy this project; diagnosis starts from the actual workflow.

Discuss your workflow