Retell turns repeat-caller context into a built-in setting for voice agents

Retell AI, a platform for building phone and chat agents, has added a built-in Memory feature that keeps a running brief on each contact. The company announced it in an October 8 blog post, and that post is the only source we could find. Our searches turned up no independent coverage or testing, so everything below is Retell's own description.

Until now, Retell says, giving a voice agent memory meant building a workflow to extract details after each call, store them and fetch them before the next one. The new feature moves that into the product behind a toggle. Retell says it maintains a brief for each contact covering preferences, what was discussed and what still needs follow-up, and clears items once they are resolved.

Retell describes the mechanism in detail. For agents set to save, a language model reads each completed phone call or SMS chat tied to a contact's phone number, along with the existing brief and contact fields, and rewrites the brief under a workspace-wide prompt. Web calls and web chats do not update it automatically. Saving and using are separate settings, so one agent can write the brief while another only reads it.

The second change affects conversation flows. Builders can mark selected nodes to be skipped when Memory or the current call already satisfies the node's condition. Retell says the agent still asks for missing details, and it advises builders to keep steps that need confirmation during the live call.

Retell also draws a limit. Memory carries conversational context, not live data, so an agent helping with a lost package would still need an order lookup to confirm shipment status.

On control, Retell says the brief can be reviewed, edited or cleared in Contacts. Reading the workspace prompt requires CRM view permission, and changing it requires CRM edit permission plus permission to edit the agent. When storage is set to exclude personal information, the company says the brief is built from the redacted transcript, and backfills skip conversations with missing transcripts or failed redaction. Retell lists backfilling and saving Memory at $0.005 per run.

What remains unknown is how well this works in practice: how accurate the rewritten briefs are, how often skipped questions cause errors, and how the redaction handling holds up.

Retell pitches the feature as letting a returning caller pick up where they left off without re-explaining. Our read, not Retell's: moving memory from custom workflow plumbing into a product setting should reduce build effort, while leaving the privacy and review decisions with the builders who turn it on.

Revised 16 October 2026: Guth re-read the original sources and rewrote this article. The earlier version remains available as revision 1.