Question 1

What does Personize actually do?

Accepted Answer

Personize gives AI agents durable memory. Instead of re-fetching the same customer, lead, deal, location, or asset context through Claude, ChatGPT, or another LLM every time, agents recall compact, governed memory from Personize.

Question 2

Why does this save money?

Accepted Answer

Raw agents often send full record context back to an LLM on every task. Personize memorizes raw data once, then returns compact recall payloads when agents need context, reducing repeated token usage.

Question 3

What counts as a record?

Accepted Answer

A record is anything your agents need to know about: a customer, lead, account, deal, product, location, asset, member, project, support case, or another entity in your business. Personize memorizes what agents capture on each record, then serves compact context back whenever they need it.

Question 4

What counts as a page of raw data?

Accepted Answer

A page is a simple estimate for context size: about 250 words. One email is usually around 0.5-1 page; search results, notes, support history, product usage, documents, or call transcripts are often 5-10 pages.

Question 5

What is a recall?

Accepted Answer

A recall is when an agent asks Personize for context about a record. Instead of searching every source again or sending full raw history to the LLM, the agent retrieves the relevant memory it needs.

Question 6

How is pricing calculated?

Accepted Answer

Pricing is based on memorizing raw data into Personize and recalling memory when agents need context. The calculator estimates both from records per month, pages per record, and recalls per record per month.

Question 7

Should I use managed keys, BYOK, or BYOC?

Accepted Answer

Managed keys are best for small usage and simple setup. BYOK is best when usage becomes meaningful because you bring your own model-provider keys. BYOC is for enterprises with larger databases, strict data controls, or predictable high volume.

Question 8

What if I do not want to manage LLM keys?

Accepted Answer

Personize can assign managed keys at signup. Additional per-token and per-call fees apply when managed keys are used for memorization or recall. For larger volumes, BYOK or BYOC usually gives you better control over model and infrastructure spend.

Question 9

How are savings estimated?

Accepted Answer

Savings estimates compare raw-context LLM usage against Personize compact recall and cached or batched memorization. Actual savings vary by model choice, retrieval depth, content size, and how often agents need the same context.

Question 10

Is this only useful at large scale?

Accepted Answer

The savings grow with volume, but the architecture matters early. Shared memory lets every new agent read from richer context instead of each tool building its own isolated memory.

Question 11

Can I still use Claude, ChatGPT, OpenAI, or other LLMs?

Accepted Answer

Yes. Personize sits underneath your agents as the memory layer. Agents can still run on Claude, ChatGPT, OpenAI, custom models, MCP clients, or your own app workflows.

Frequently Asked Questions

Honest answers about memory economics