Your company knowledge, searchable by AI - without it ever leaving the hive.

MadHoney RAG indexes Confluence, Jira, Azure DevOps, GitHub, GitLab, Forgejo, SharePoint, OneDrive, Google Drive, Slack, databases and local disks into a private vector database on your own hardware, and exposes it to any AI agent or assistant over MCP.

localhost:3333 · RAG Inspector · workspace acme-finance
embedding 29 ms (local bge-m3)vector search 5 ms (LanceDB)total 34 ms
  1. [1]Q3 Financial Performance & Sales Commissions

    localscore 0.914

    Finance / FY25 / Quarterly Reports / Q3

    Q3 Enterprise ARR closed at $2.4M (112% of target). Total sales commission payouts reached $185,000 across 8 Senior Account Executives. Top performer Marcus Vance achieved 142% quota, triggering Tier 3 accelerators ($34,500 payout).

  2. [2]Q4 Headcount Budget & Compensation Plan

    gdrivescore 0.831

    HR & Finance / Headcount / Sales Org

    Approved Q4 budget includes headcount for +3 Senior AEs and +2 Solutions Engineers (filtered by Sales Org / Enterprise NA). Base salary range fixed at $130k-$150k with 50/50 OTE split and quarterly commission schedules.

  3. [3]Sales Compensation Policy & Quota Thresholds

    confluencescore 0.692

    Revenue Ops / Policies & Guidelines

    Quarterly payout rules: Reps must reach a minimum 80% quota threshold in Q3/Q4 to unlock variable pay. Commission payouts are filtered and audited by RevOps on the 10th business day following quarter end.

Why MadHoney RAG

Private by design

Zero data leakage

Documents, embeddings and the vector index never leave your machine. The cloud portal only handles your licence.

$0 API ingestion costs

bge-m3 runs locally via ONNX - index millions of chunks without paying per token.

On-premise performance

LanceDB on local disk answers in milliseconds, with strict per-workspace isolation (workspace_id pushdown filters).

Audit-ready for ISO 27001 & SOC 2

Stored on your hardware, ready for your auditors. On every plan.

How it works

The RAG Pipeline

Connectors feed one private vector index on your hardware. AI agents query it over MCP for grounded, up-to-date context.

Everything runs inside your infrastructure: connectors for files and S3, databases, REST APIs, Confluence and Jira, Azure DevOps, GitHub and GitLab, SharePoint, OneDrive, Google Drive and Slack run in parallel and feed the ingestion pipeline of the MadHoney engine on your hardware. A rule-based quality filter removes noise and duplicates, then content is chunked, embedded locally with bge-m3 and indexed in the local vector database. AI agents search it through the MCP server, optionally only with an access token you issue. The hosted control plane at rag.madhoney.cz only handles your account, billing, Engine Token and engine downloads and never receives your documents, queries or embeddings.
Your infrastructure

Connectors

  • Files & S3

    Local disk, S3 - MD, PDF, DOCX, XLSX

  • Databases

    PostgreSQL, MySQL, SQL Server

  • REST APIs

    Internal systems - JSON, CSV, HTML

  • Confluence & Jira

    Spaces, pages, issues

  • Dev platforms

    Azure DevOps, GitHub, GitLab

  • Workplace

    SharePoint, OneDrive, Drive, Slack

Run in parallel, each with its own credentials.

MadHoney engine

Ingestion & Embedding Pipeline

A rule-based quality filter drops noise and duplicates; content is then chunked and embedded locally with bge-m3 (cloud embeddings only if you opt in).

Core knowledge store

Vector Database

LanceDB on your disk - workspace-isolated index for vector or hybrid search in milliseconds.

MCP Server

Model Context Protocol endpoint exposing search_knowledge_base as a tool. Optional access tokens - you decide who searches

Agents

AI Agents

Autonomous agents & assistants

  • Semantic search
  • Context retrieval
  • Grounded answers

rag.madhoney.cz - Control plane

Hosted by MadHoney: your account, billing, Engine Token and engine downloads. It never talks to your engine or stores your documents, queries or embeddings.

  • Account
  • Engine Token
  • Downloads
Tour

See the engine at work

Five short chapters, from connecting your sources to a cited answer in your AI assistant - sample screens of the local Admin UI.

Admin UI · Sources · workspace acme-finance
SourceDocumentsStatus
  • Confluence0syncing
  • Jira0syncing
  • GitHub0syncing
  • SharePoint0syncing
  • Google Drive0syncing
  • Local folder0syncing

Chapter 1 of 5

Add your sources in the local Admin UI. Every connector runs as its own process with its own credentials, all of them in parallel.

Business value

One engine, value for every role

Pick your role to see which layers matter to you and what they deliver.

Primary focus

Zero trust & IP protection

On-premise boundary

0 documents out

Documents, vectors and search stay on your network. What your AI assistant sends to its provider stays under your control.

Control plane isolation

Account & billing only

rag.madhoney.cz handles your account, billing, Engine Token and engine downloads. It never receives documents, queries or embeddings.

Token-gated AI access

Revocable tokens

Switch on MCP access tokens and only assistants holding a token you issued can search. Validity 1 to 12 months, only hashes stored, revoked access ends on the next request.

Your infrastructure
  1. Input

    Multi-source sync

  2. Processing

    Quality filter & bge-m3

  3. Storage

    LanceDB on your disk

    Vector and keyword index with millisecond retrieval - no cloud database.

  4. Integration

    MCP knowledge server

    search_knowledge_base for your AI assistants and coding agents - optionally gated by access tokens.

rag.madhoney.cz - control planeAccount, billing & Engine Token only - never your documents.
Connectors

Every source, one private index

Parallel connectors pull data from all your tools, automatically filter out noise, and build a unified local vector index - keeping your enterprise data completely private.

ConfluenceSpaces and pages, converted from XHTML to clean Markdown.

Get started

From plan to first answer

  1. Choose a plan

    Subscribe on rag.madhoney.cz and generate your Engine Token.

  2. Download & start

    Run start.bat or start.sh from the engine bundle - the engine runs as a Docker image on your hardware.

  3. Activate offline

    Paste the Engine Token and create your Super Admin in the local Admin UI.

  4. Connect your assistant

    Add sources, run ingestion and add the generated .mcp.json to your AI assistant.

Docker image

Runs on Docker Desktop (Windows, macOS), Docker Engine (Linux) or Podman.

Windows installer (no Docker)

Runs as a native Windows service on Windows 10 / 11 and Server 2019+ - for servers where containers are not an option.

Runs on your hardware and is activated offline - your documents stay with you.

See plans