Skip to content

Get started

Agent Memory

Principles, architecture, quickstarts, and mental model for SurrealDB Agent Memory - memory and knowledge for AI agents on SurrealDB.

SurrealDB Agent Memory is a memory and knowledge layer for AI agents. It is an application tier in front of SurrealDB, and every durable record lives in that database: graph, vector, document, relational, and geospatial, written under ACID transactions. Provenance and trust sit on the records themselves. Traces of retrieval and decisions are graph nodes, not disposable logs. Belief history is tri-temporal, so the system can keep apart what was said, what is true now, and what used to be true.

Use this hub to go from principles to running code, then dive into the product sections (memory & knowledge, integrations, cookbooks, reference).

Note

SurrealDB Agent Memory was developed under the project name Spectron, and that name is retained throughout the shipped interface: the spectron and spectrond binaries, the SPECTRON_* environment variables, the spectron_* MCP tool names, and SDK packages such as @surrealdb/spectron. The rule of thumb is that prose uses the product name and anything you type or configure uses spectron. These names will be renamed in a future release.

What SurrealDB Agent Memory is built to do, what it is not, and how retrieval, traces, and time work:

Quickstarts

Building with AI coding tools? Start with Agent guide (AGENTS.md) - copy it into Cursor rules or a project skill so your agent can learn SurrealDB Agent Memory without reading the full docs.

How isolation, sessions, categories, and provenance fit together:

  • Memory & knowledge - authoritative and experiential ingest, unified retrieval, reasoning, operations, tuning.

  • Integrations - SDKs, MCP, framework adapters.

  • Cookbooks - end-to-end patterns.

  • Reference - REST, management API, CLI, configuration, errors.

Was this page helpful?