​

Correcting Knowledge

BaseLayer extracts knowledge automatically. Sometimes it gets things wrong. Maybe it thinks a colleague works at the wrong company, or it confused two projects with similar names. You can correct your memory using natural language. Just tell your AI what’s wrong.

​

How to make corrections

In any MCP-connected AI (Claude, ChatGPT, etc.), simply state the correction:

“BaseLayer thinks Sarah works at Acme Corp, but she actually moved to Stripe last month.”

“The payments project is called ‘Atlas’, not ‘Atlas Migration’.”

“Mark and David are two different people. BaseLayer merged them.”

Your AI will use the record_memory tool to update the relevant entities in your memory.

​

Using record_memory directly

If you want to be precise, you can ask your AI to call record_memory with specific details:

“Remember that Sarah Chen’s current employer is Stripe, as of January 2026.”

This creates or updates the entity with the correct information. Previous versions are preserved. BaseLayer tracks entity history, so corrections don’t destroy context.

​

Removing incorrect knowledge

If BaseLayer created an entity that shouldn’t exist at all, tell your AI to correct or remove it. Ask your AI to remove or correct it. For example: “That entity called Project Omega shouldn’t exist. It was just a hypothetical.” Your AI will use vault_fact or record_memory to make the correction.

​

Preventing future errors

The dream engine learns from corrections. If you fix the same type of error repeatedly, the extraction process adapts. Your memory gets more accurate over time.