Skip to main content

Memory

Everything an agent "knows" and "remembers" is organised around context primary keys. A key groups the conversation history for one specific thing — a customer, a contract, a ticket, a GDPR article — so that when an expert is asked about that thing again, it can load everything it ever discussed under that key.

Three concepts to understand​

  • Context primary key — a unique identifier that creates an isolated memory space. Example: contract number C45653 groups every conversation the CRM expert has ever had about that customer.
  • Memory strategy — how the memory gets populated. Lazy (grows from real conversations), Hydration (pre-loaded synthetic conversations), or MCP (fetched live from an external system).
  • Threading phase — when an expert is first @-mentioned in a conversation, it runs a threading routine to identify which context primary key applies and lock it for the rest of the thread.

Getting started​

  1. Decide what makes a good key for your agent's domain (email? contract number? ticket ID?)
  2. Create the keys through the dashboard for small batches, or through the API for bulk
  3. Pick a memory strategy that matches your use case
  4. Configure the agent's Context Not Found prompt to recognise your key format
  5. Test in a real conversation

For developers​