SurrealDB Agent Memory

What it does

SurrealDB Agent Memory is a memory and knowledge layer for AI agents. It takes in conversations, documents and facts an agent learns, and turns them into typed entities, attributes and relations. Each stored fact keeps its source, span, confidence and validity dates. When a new fact contradicts an old one, a reconciler supersedes the old value while keeping its history, so an agent can answer questions as of a given date. Recall fuses seven retrieval signals into one ranking, and writes and answers are recorded as traces.

People use it to give agents memory that persists beyond a session, to query what an agent believed at a past date, and to trace an answer back to its source passage. Developers connect it through a CLI, a TypeScript or other language SDK, a REST API, or seven MCP tools that work with Claude Desktop, Cursor and Claude Code. Every fact lives in one SurrealDB transaction, so the memory and its history are not split across separate stores.

It suits developers and teams building agents that need auditable, correctable memory. A free Sandbox plan includes 3 million tokens one-off with no card required; paid Cloud plans start at $30 per month. Limits to note: the memory layer is closed source today, SurrealDB itself is source-available under the Business Source License 1.1, Enterprise pricing is custom, and the benchmark result is self-reported.

What it can do

• Typed memory: conversations and documents become attributes and relations on entities, each with source, span and confidence.
• Knowledge graph: turns and documents share one entity graph, with citations back to the passage.
• Recall: vector similarity, keyword matching, graph relationships and usage history fused into one ranking.
• Provenance: writes, ranked reads and chat answers record traces linked to the facts they touched.
• Temporal memory: a new value closes the validity of the old one rather than overwriting it.
• Repeated-question handling: cached chat answers are not reused once a cited fact is superseded.

Works with

• Claude
• OpenAI Codex
• Cursor
• VS Code
• Zed
• LangChain
• OpenAI Agents
• CrewAI
• Google ADK
• Pydantic AI

Before you choose

• The SPECTRON_* environment variables and X-Spectron-* headers still use the former project name.
• Enterprise pricing is custom and requires sales contact; self-hosting is an Enterprise-tier option.
• The 77.2% BEAM 1M score is self-reported with the judge and date stated on the page.

Listed under

• Coding
• MCP and agent tools

Get started

Free Sandbox needs no card; open Studio, create a context and copy its URL and API key.

1. Create a context in SurrealDB Studio and copy its URL, API key and context ID.
2. Install the agent-memory CLI and export SPECTRON_URL, SPECTRON_API_KEY and SPECTRON_CONTEXT_ID, then run remember and recall.
3. Add the context's /mcp endpoint with the API key and X-Spectron-Context header to Claude Desktop, Cursor or Claude Code.

Paid plans are monthly prices that include a monthly token allowance; tokens are counted on every language-model and embedding call. Usage beyond included tokens has overage rates not stated numerically on the page. Enterprise is an annual custom agreement. Separate SurrealDB Cloud instance pricing (hourly compute, per-GB storage) is on the pricing page and is not confirmed to apply to Agent Memory plans.

Source: surrealdb.com/pricing, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Agent Memory you can trust | SurrealDB ↗surrealdb.com
• Agent Memory | Get started | Agent Memory ↗surrealdb.com
• The context and memory layer for AI agents | SurrealDB ↗surrealdb.com
• Pricing and plans | SurrealDB ↗surrealdb.com

Company website

surrealdb.com: product, pricing and policy pages

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

manual_review

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.