Skip to content

Memory and context

Memory settings decide how much of the previous dialog a virt reuses in each response and whether new turns are written to history. Tune them to balance cost, privacy, and quality.

Preconditions

  • Agreement with legal on retention limits
  • A test virt where you can replay dialogs
  • Metrics that define when the virt should escalate to a human

Steps

01

Open the Memory tab

Select the virt, choose AI, and open Memory. Confirm the environment toggle so you edit the right workspace.

Memory configuration tabMemory configuration tab
This page shows the context window and memory write mode
02

Pick a context window

Set a short window (up to 20 messages) for FAQ flows, a medium window (30-50) for consults, or a long window when you troubleshoot complex issues. Remember that longer context increases spend and latency.

Context window sliderContext window slider
Hints describe which window works for each use case
03

Configure context window and write mode

Use Last messages to cap how much recent context is retained per thread and keep spend predictable. If needed, enable read-only behavior so the virt can use existing memory but does not write new history during sensitive flows.

Context window and memory mode optionsContext window and memory mode options
Retention controls are based on last-message window and write mode
04

Test the new window

Send two or three linked questions in the preview channel and confirm the virt references earlier answers correctly. If context gets lost, either expand the window or update the instruction to restate key facts.

Preview dialog showing retained messagesPreview dialog showing retained messages
Preview shows which turns will be passed to the model

WARNING

Keep only the context you need. If policy requires minimizing personal data, reduce Last messages and enable read-only mode in sensitive flows.

Practical Example

Example

Practical Example

A fintech support virt keeps a 40-message window and enables read-only mode for sensitive flows. This preserves continuity while preventing unnecessary data from being written to history.

Руководство пользователя Worken AI