コンテンツにスキップ

Conversation Memory

このコンテンツはまだ日本語訳がありません。

v1.7.0

Conversation Memory lets your AI assistant remember durable facts about each verified customer — their name, preferences, recurring issues, purchases, and feedback — so future conversations feel personal instead of starting from scratch.

When a conversation is resolved (closed by an agent or admin), Clienta.ai runs a durable background job that:

  1. Generates a short AI summary of the conversation.
  2. Extracts structured facts using a fast, low-cost language model.
  3. Stores the facts, linked to the verified contact.

Extraction is non-blocking — resolving a conversation is instant; the memory work happens asynchronously and never delays your agents.

Each fact has a category and a short key:

CategoryExamples
personalName, timezone
preferencePreferred language, contact channel
issueRecurring problem (“billing via PromptPay”)
purchaseProduct or plan owned
feedbackStated likes / dislikes

Sensitive data is filtered out — payment card numbers, government IDs, and similar patterns are discarded before anything is stored. Facts below a confidence threshold are also dropped.

On a returning conversation with a verified customer, the AI is given a compact, read-only customer context block (capped to a small token budget). It references this data to personalize greetings and responses — for example, addressing the customer by name or recalling a prior issue.

Memories expire automatically. Each organization sets a retention period (default 365 days) under Settings → AI Settings. A daily cleanup job hard-deletes any memory past its expiry.

  • Re-extracted facts upsert by (contact, category, key) — the newest value always wins.
  • Withdrawing consent immediately wipes all of that contact’s memories.

See Privacy & Compliance for export, erasure, and consent controls.