Recall past interactions: your AI agent's memory across conversations

"Recall past interactions" is a tool designed to let an AI agent search a contact's own history for earlier conversations relevant to the current question — for example when a customer says "the issue I reported before" or "my last order".

How it's designed to work

1. When a conversation closes, the system generates a summary and stores it as a "memory" tied to that contact.
2. When the agent needs context, it turns the customer's question into a semantic vector and compares it against stored memories.
3. It returns the most similar summaries (5 by default, up to 20) for the model to use while replying.

Current status

This tool exists in Omnifox's technical catalog but is globally disabled by a product decision: every AI agent is scoped to the current conversation only, without cross-referencing other chats from the same contact. That's why you won't find it toggleable under the Actions tab of the agent editor — it shows up listed as "not available".

How to keep continuity in the meantime

• Use Add internal note so the agent leaves notes visible only to your team (e.g. "customer already reported this bug on 7/3").
• Use Update contact fields to save relevant info directly on the contact record (company, plan, preferences).
• Human agents always see the contact's full conversation history in the Inbox, even though the bot doesn't automatically "recall" it.

Example

A returning customer writes: "did you confirm the invoice thing we talked about last week?". Today the AI agent can't look up that past conversation on its own — the right move is to escalate with Assign to person or team, after leaving an internal note with whatever context is already available.

Troubleshooting

• "The agent doesn't remember anything from earlier chats": that's expected behavior right now — cross-conversation semantic memory isn't active.
• "Will it be enabled?": the underlying code is kept in place for a future re-launch; if your operation needs it, reach out to support and describe your use case.