Open source Local-first MCP-powered

Your data.
LLM-ready.
Privately.

DEVAN Agent transforms Excel spreadsheets and PDF documents into structured, queryable, AI-ready data — entirely on your machine. Zero cloud. Zero compromise.

100% local processing
Zero data telemetry
8 specialised MCP tools
Windows · macOS · Linux
PII detection built-in
The challenge

Your most valuable data is
invisible to AI.

Businesses hold years of operational knowledge locked inside Excel files and PDF reports. That data is too structured for LLMs to read directly, and too sensitive to send to the cloud.

Data trapped in silos

Excel files with dozens of sheets, formulae, and references. PDFs with tables, charts, and embedded data. None of it natively digestible by language models.

Privacy prevents cloud upload

Financial records, HR data, client information — you can't upload it to a third-party service. GDPR, compliance, and plain common sense stand in the way.

Custom pipelines take months

Writing custom ETL pipelines to extract, clean, and structure this data takes engineering teams months — before a single LLM query can run.

DEVAN bridges the gap

Upload your files. Ask questions in plain English. Get structured data, charts, and SQL queries — all running locally in minutes, not months.

How it works

From raw file to
AI-ready insight in three steps.

No configuration. No cloud sign-up. No data leaving your machine.

01

Upload

Drop any Excel workbook or PDF document into the DEVAN interface. Multi-sheet workbooks, scanned PDFs, and complex layouts all supported.

02

Ask

Type a question in plain English. DEVAN's AI selects the right MCP tool, extracts structured data, and runs queries — automatically.

03

Export

Download clean CSV, JSON, Excel, or Parquet. Copy the generated SQL. Share a chart. Your data — your format.

System overview

  ┌─────────────────────────────────────────────────────────┐
  │   Your Machine                                          │
  │                                                         │
  │  ┌──────────────┐      ┌───────────────────────────┐   │
  │  │  Your files  │─────▶│     DEVAN Agent (Docker)  │   │
  │  │  Excel · PDF │      │                           │   │
  │  └──────────────┘      │  ┌─────┐  ┌──────────┐   │   │
  │                        │  │ LLM │  │ 8 MCP    │   │   │
  │  ┌──────────────┐      │  │local│  │ servers  │   │   │
  │  │  Browser UI  │◀────▶│  └─────┘  └──────────┘   │   │
  │  │ localhost:   │      │                           │   │
  │  │    5001      │      └───────────────────────────┘   │
  │  └──────────────┘                                       │
  │                              Zero data exits ──▶ ✗      │
  └─────────────────────────────────────────────────────────┘
Capabilities

Eight specialised tools.
One unified agent.

Each MCP server is purpose-built for a specific data task, orchestrated by DEVAN's AI to answer your questions end-to-end.

Core

Excel Retriever

Multi-sheet extraction, formula resolution, data type inference, and relationship detection across complex workbooks.

Core

PDF Extractor

Layout-aware text extraction, OCR for scanned documents, table parsing, and chunking for RAG pipelines.

Data

Data Modelling

Infer relational structure from flat files. Export to JSON, Parquet, Arrow, or SQLite. Run DuckDB queries directly.

AI

Prompt Engineering

Curated prompt templates for document summarisation, data dictionary generation, and structured extraction tasks.

Secure

Guardrail

LLM output safety validation with PII detection, content filtering, and audit logging for compliance-sensitive workflows.

AI

Web Design

React component generation from natural language descriptions. Ideal for rapid prototyping data dashboards.

Data

Document Intelligence

Universal document understanding with Apache Arrow / Parquet compute tools. Handles mixed-format datasets: Excel, PDF, CSV, and JSON in a unified pipeline.

Secure

Audit & Compliance

Full audit trail of every LLM call, tool invocation, and data access event. Exportable log for enterprise compliance teams.

Privacy by design

Your data never
leaves your machine.

DEVAN runs entirely in a local Docker container. No API calls for your data. No telemetry. No third-party storage. Your files are processed and discarded on your own hardware — full stop.

  • Local Docker container — all compute stays on your hardware
  • LLM runs via Ollama (local) or your own API key (BYOK)
  • PII detection flags sensitive data before any LLM call
  • Full audit log exportable for compliance review
  • Open source — inspect every line of code
live · local processing

Outbound network requests

0

100%
local compute
0 KB
data uploaded
8
MCP servers
file size
Get started

Up and running
in three commands.

No account required. No credit card. Just Docker.

Step 01

Pull the image

Grab the latest DEVAN Agent image from Docker Hub.

docker pull m2laborg/devan-agent
Step 02

Run it

Mount your data folder and start the agent.

docker run -p 5001:5001 \
-v ~/data:/data \
m2laborg/devan-agent
Step 03

Open the UI

Navigate to the agent in your browser and start asking questions.

open http://localhost:5001

Ready to make your data speak?

Star the repo, pull the image, or read the docs — everything is free and open source.

Star on GitHub Docker Hub