Skip to main content
Workspace memory helps Custory AI carry useful context forward across the life of a workspace. Instead of every conversation starting from a blank page, the assistant can retrieve relevant prior knowledge from the same workspace and, when relevant, from the same journey.

​
What memory is

Custory uses workspace-scoped memory powered by Supermemory. Memory belongs to the workspace it came from. It is not one shared global pool across every workspace your team uses. When an AI conversation is tied to a journey, Custory can favor memory connected to that journey before falling back to broader workspace knowledge.

​
What can be remembered

Custory can sync supported workspace context into memory with workspace, journey, source type, and source ID metadata.

​
Memory status tracking

Custory tracks memory source state so the team can tell whether a source is usable. Ingestion can fail when the memory provider is not configured, the source content is empty, the provider rejects the request, or the request errors while syncing.

​
Memory control in AI chat

Custory includes a memory toggle in AI chat. Use memory on when the answer should draw from prior workspace or journey knowledge. Use memory off when you want the answer constrained to the current prompt or visible page context. Turning memory off skips retrieval for that chat. It does not delete existing memory.

​
What memory is good for

Memory is most useful for repeated work where the same context would otherwise be restated:
  • recurring journey reviews
  • ongoing product discovery
  • support, product, and delivery follow-up on the same journey
  • AI work that needs remembered decisions, preferences, or unresolved follow-ups
Memory is weaker when the workspace itself is stale or noisy. Clean source journeys and items make retrieval more useful.

​
Limits

  • Memory helps the assistant retrieve prior context; it does not prove the remembered statement is still true.
  • Imported content is treated as evidence, not as instructions the assistant should blindly follow.
  • Long source content may be truncated before ingestion.
  • Deleted memory sources are marked as deleted and removed from the memory provider when possible.
Treat retrieved memory as support for the answer, not as authority over the current workspace state.

​
Next step

Read AI as a workspace member for the broader AI collaboration model. Read MCP if your team also uses external AI clients.