Powering Mr Fabry in rossum.ai

From Words to
Functional Workflows

AI agent toolkit that transforms natural language into working Rossum configurations — schemas, hooks, automations, and more.

View on GitHub Documentation
Mr Fabry is currently in alpha inside Rossum.
Named after the Chief Engineer in Karel Čapek's R.U.R.
56
MCP Tools
Thin wrapper around the Rossum API
6
Agent Skills
Precise workflows for complex tasks
3
Sub-Agents
Offload work, keep context clean

See It in Action

Mr. Fabry inside Rossum — from initial prompt to detailed analysis

Mr. Fabry welcome screen inside Rossum with quick-start prompts for analyzing queues, debugging hooks, and setting up workflows
Start a conversation Ask about documents, queues, workflows, or anything else
Mr. Fabry displaying a comprehensive queue analysis with 12 active extensions, validation rules, email settings, and automation configuration in a structured response
Get detailed insights Queue analysis with extensions, validation rules, and configuration
Mr. Fabry walking through Duplicate Handling extension setup step-by-step, showing live tool calls to create hooks and configure matching rules through natural conversation
Configure extensions through conversation Set up hooks, webhooks, and automation rules — Fabry handles the implementation

What It Does

Transform complex Rossum operations into conversational workflows

Organization Setup

Create queues, configure schemas, add validations, hooks, and email notifications—all through natural conversation.

Workflow Analysis

Analyze document processing workflows with detailed insights into extensions, rules, and configurations.

Workflow Diagrams

Generate visual workflow diagrams showing document flow, hook interactions, and processing stages.

Hook Debugging

Diagnose and fix hook issues with intelligent analysis and sub-agent Python code debugging.

Sandbox Testing

Test and validate changes safely in sandbox environments before applying to production.

Knowledge Base

Connected to Rossum Knowledge Base for contextual documentation and best practices.

Schema Skills

Sub-agents and skills for schema patching, field updates, and intelligent schema modifications.

Read-Only Mode

Enable read-only mode to explore and analyze your setup without any risk of changes. Browse freely, zero destructive access.

Image Understanding

Upload and analyze images directly—the agent sees and understands visual content for richer document insights.

Architecture

rossum.ai Mr Fabry
rossum-agent Opus 4.5 + Skills
rossum-mcp 49+ API Tools
Rossum API Document Processing

Quick Start

Get running locally for development

# Install
uv pip install rossum-agent

# Set AWS credentials (for Bedrock access to Claude)
export AWS_ACCESS_KEY_ID="your-access-key"
export AWS_SECRET_ACCESS_KEY="your-secret-key"
export AWS_REGION="eu-central-1"

# Run streamlit test-bed UI
rossum-agent

Frequently Asked Questions

Can I use a different model than Opus 4.5?
No. Each model requires a custom-tuned system prompt to ensure reliable performance. To move fast and maintain quality, we optimize exclusively for Opus 4.5. Supporting multiple models would fragment our testing and slow down iteration.
Is it safe to let the agent modify my Rossum configuration?
Yes. You control what the agent can do:
  • Read-only mode: Browse and analyze your setup freely — no changes possible
  • Read-write mode: The agent can create and update queues, schemas, hooks, and more
Every action is logged and visible in real-time, so you always know what's happening.
What data is sent to the LLM?
Only conversation context and Rossum metadata needed for the current task. Document content is not sent unless you explicitly ask about specific document data. We use AWS Bedrock, which complies with our security standards.
Can I extend it with custom tools or skills?
Not currently. The system is designed as a cohesive toolkit optimized for Rossum workflows. We may open extensibility in the future based on demand.
How do I know the model isn't just hallucinating?
Every tool call is logged and visible to you in real-time. For example, when you ask to set up an Invoice queue with a custom "Net Terms" field, you can watch the agent:
  • 📂 Load the queue setup skill
  • 📋 Fetch available queue templates from Rossum
  • 🏗️ Create the Invoices queue from a template
  • 🌳 Retrieve schema tree structure
  • ✂️ Prune unnecessary fields (payment instructions, delivery address)
  • 🔧 Load patching skill and add "The Net Terms" field
  • 🧮 Call Rossum API for formula suggestion (Date Due − Date Issue → Net 15/30/Outstanding)
  • ✅ Validate and apply the final configuration
You see exactly what happens at each step — no black box.

Want to Learn More?

Ping me on Slack or drop me an email — happy to chat.

daniel.stancl@rossum.ai