Player not loading? Watch on YouTube
Brandon Barros and Chiru Boggavarapu explain how Mastra integrates with SurrealDB Agent Memory through a fictional company called Comet Labs. The prepared demo contains five agents. Live queries to HR, finance and concierge agents retrieve shared context about equipment problems, spending rules and meeting notes.
Boggavarapu describes Agent Memory as a managed layer above SurrealDB's multi-model database. He says it turns conversations into typed facts, records their origins and keeps superseded facts rather than deleting them. An explicit forget tool can remove selected context. Directory-like scopes and API-token grants control which agents or users can read and write particular information.
The workshop distinguishes Mastra's memory capability from the SurrealDB adapters that supply its storage. Developers can use the database adapter to build their own knowledge graph, or the Agent Memory adapter for a managed experience. SurrealDB Studio displays entities and graph relationships; Mastra Studio exposes tool calls and timing traces.
Barros says Mastra supports local or cloud models and remains TypeScript first. A local LLM does not make the demonstrated memory service local: Boggavarapu describes Agent Memory as cloud-only, with a self-hosted version under consideration. He directs viewers to a waitlist for access during the workshop.
