What is a space (agent workspace)?

A space is the unit of privacy. Every end customer gets one, and everything about that relationship — what they asked, what they uploaded, what the agent remembers — lives inside it.

Why it is not just a filter

The tempting way to build multi-customer AI is to keep everything in one pile and add a customer id to each query. It works until one query is written without the filter, and then it fails silently and in the worst possible direction: someone reads someone else's documents.

Isolation belongs one layer down, in the database, so that a query missing its scope returns nothing rather than everything.

How it is enforced here

Isolation is not a WHERE customer_id = ? that every query has to remember. In agent4.io a customer's conversations, uploaded documents and memory are keyed to their space and enforced by row-level security in the database, so a query that arrives without its scope returns nothing rather than everything. The document bodies are additionally encrypted with a key belonging to that customer.

This is the same boundary the customer knowledge base sits behind — which is why customers can be invited to upload passports, payslips and refusal histories without that becoming a problem you now own.

What this buys a business

Per-customer memory that is safe to keep. An agent can remember that this client prefers WhatsApp after six, that their case involves a spouse, that they already sent the bank statements — precisely because that memory cannot leak into a conversation with someone else.

A defensible answer to the security question. "How do you keep our clients' files apart" is the question every serious buyer asks, and "we filter by customer id" is not a comfortable answer.

Somewhere for sensitive material to go. Passports, refusal histories, financials — these arrive whether or not you designed for them.