How a Pillowcase system works.
You don't need to understand the technology to work with us. But if you want to know what happens behind the scenes — here it is.
Three layers: knowledge, rules, tools.
Behind every Pillowcase system is a structured agent environment. What matters to you isn't the technical name but the result: a system that prepares work and makes it traceable.
The three layers
Layer 1 — Knowledge (knowledge base)
The foundation. The agent works only with documents you've approved: manuals, guidelines, contracts, templates, notes. Not your entire archive — only what you've explicitly selected.
More detail
The knowledge base is searchable via a RAG core (Retrieval-Augmented Generation): the agent searches your documents for relevant passages and answers with source references. If it finds nothing, it says so — it invents nothing.
Layer 2 — Rules and approvals
The agent acts by rules we define together. Which topics are sensitive? Which answers may be drafted automatically? Who approves? When is escalation triggered?
More detail
These rules aren't rigid — they adapt anytime your requirements change.
Layer 3 — Tools and log
The agent has tools: search, classification, draft creation. Every use is logged. You can reconstruct at any time what the agent did, when, and why.
Knowledge search with sources — not invented answers.
The heart of the knowledge base is the RAG core. "RAG" stands for Retrieval-Augmented Generation — a technical term for something simple:
The agent doesn't answer from memory, but from your documents. It finds the relevant passage, shows it, and says where it sits. If no suitable source exists, it admits: "I don't know that."
That's the difference between an agent that hallucinates and one you can trust.
We don't pick the model — we configure the agentic framework on top of it. Validation contracts check each step. When the model is unsure, the agent asks, not guesses. The result is work that's actually ready, not drafts your team has to redo.
Example
"What's our policy for order approval on amounts over €5,000?"
"Orders over €5,000 require management approval."
purchasing guideline §3.2"I have no approved source on this. Please contact [responsible person]."
no approved sourceSeparate systems. No shared data.
Every Pillowcase customer gets their own environment. Your data isn't shared with others. Systems are operated to German and European data protection standards.
Separate per customer
Every customer = own environment
GDPR-aware
Planned to European standards
No full-archive access
Only approved documents
Agent log
Every step traceable
The stack (if you want the details)
For the technically interested — others can skip this section.
Technical overview
- Agent framework
- Hermes (open source)
- Knowledge search
- RAG core with vector search (Qdrant)
- Language models
- Configurable, depending on requirements and data protection
- Backend
- FastAPI + PostgreSQL
- Infrastructure
- Docker, on own or customer infrastructure
- Logging
- Complete agent log for every step
Questions about the technology? Let's talk.
Technical details are best clarified in the Free AI Review. If you have questions in advance, write to us.