One of the most persuasive moments in an AI agent demo is when it answers a question using your own material: your price list, last year's proposals, the notes on a long-time customer. It feels like the agent "knows" your business. For an owner, that's exactly the moment to ask a less exciting question: where did the agent put the copy?
Full disclosure: I built Agent Ready Rankings (https://rankings.leffke.com), a site that scores AI agent tools out of 10. The scores below are overall scores from my November 9 data run, not a same-day refresh. Browsing the site is free; the full score breakdown is part of Pro, $10/month.
How agent "memory" usually works
In many projects, an agent doesn't learn your documents the way a new employee would. Instead, the builder splits your files into small pieces, turns them into a searchable form, and stores them in a database. When a customer asks a question, the agent looks up the most relevant pieces and reads them before answering. This pattern is often called retrieval-augmented generation, or RAG.
That design is practical, but it means a second copy of your business information now exists somewhere, and it has a life of its own.
Three things it changes for you
Data location and terms. If the database is a hosted service, your documents sit under that provider's terms as well as the model provider's. Ask which company holds them and what their retention terms say.
Deletion. When you update a price list or a customer asks to be removed, the old pieces don't disappear on their own. Someone has to delete or refresh them. Ask who does that and how you'd confirm it happened.
Cost. Storage and search usually have their own plan, separate from the model API. A free tier can cover a small pilot; a growing document library may not stay free.
The Memory & RAG tools I track (overall scores)
Pinecone: 6.65 out of 10. A managed vector database for semantic search and agent memory. Free tier with paid plans. The latest change I've recorded is pg_pinecone, a Postgres extension, dated October 27, 2026.
LlamaIndex: 6.55 out of 10. A data framework for connecting documents to LLM apps, building indexes, and running RAG workflows. Free tier with paid plans. The last release I've tracked was September 30, 2026, which is part of why it sits a little lower right now.
These are tools your builder would use, not products you'd buy yourself. That's why it's worth knowing what to ask.
A short owner checklist
Before an agent starts "remembering" your business, ask:
- Which documents will it store, and is anything sensitive (contracts, customer records) excluded?
- Which service holds the stored copy, and where are its data terms written down?
- When we change or remove a document, how does the old version get deleted?
- What does storage and search cost per month at our expected size?
About the rankings
Agent Ready Rankings covers 22 AI agent tools across coding agents, LLM APIs, agent frameworks, voice, memory, and evaluation and safety. I built it, so treat every score as one informed opinion and tell me where you think I'm off.
Memory & RAG category: https://rankings.leffke.com/categories/memory-rag
Top comments (0)