Ground the conversation
Recall relevant preferences and prior statements so the agent does not force the caller to repeat everything.
recall → conversational contextMemory for voice agents
Let the conversation flow. Gate the action.
Callers think aloud, change their minds, and refer to old conversations. ContextDB helps a voice agent remember all of it without treating every remembered sentence as permission to book, refund, or update.
The problem
“Maybe Friday” should not overwrite “Thursday is confirmed.”
A voice model needs context to sound coherent. The booking system needs a stricter standard. Use ordinary recall for the conversation and action recall before the tool call.
tool call · book_visit
ask
heard now
“Friday might be easier, maybe.”
0.4
confirmed
“Yes, Thursday afternoon.”
0.95
next step
Ask which day to book
Where it fits
One memory layer across the call lifecycle.
Evaluate action evidence
Check the specific booking, refund, reschedule, or escalation against source, confidence, confirmation, and policy.
recall_for_action → act | ask | abstainKeep the decision trace
Store which evidence and policy supported the result so an operator can review the action without replaying a full prompt.
decision → evidence + policy + reasonAction map
Not every voice-agent step needs the same trust bar.
| Step | Memory mode | Why |
|---|---|---|
| Answer a general question | recall |
Relevant context can help, while the response remains conversational. |
| Book or reschedule | recall_for_action |
The call creates a commitment in another system. |
| Issue a refund | recall_for_action |
The action changes money or account state. |
| Escalate to a person | Host policy | Missing or conflicting evidence should fail closed. |
Questions
Voice-agent memory, without the hand-waving.
Do I keep call transcripts?
No. Keep transcripts and recordings in their appropriate systems. ContextDB stores action-relevant memories, their provenance, and the decisions made from them.
Does ContextDB make the tool call?
No. Your voice-agent host receives the decision and must enforce it before calling booking, billing, CRM, or scheduling tools.
Can the agent still use tentative context in conversation?
Yes. Ordinary recall and action recall serve different jobs. Tentative context may help the conversation while remaining blocked from action.
Test your first booking guard locally.
Store one confirmed preference and one tentative wish, then inspect the result.