Company Brain with SurrealDB Agent Memory

A shared organizational memory the whole team talks to. What one person teaches it, everyone can recall — decisions, customer details, tribal knowledge, the stuff that usually lives in someone's head and leaves when they do.

Built on SurrealDB Agent Memory (Spectron), the hosted memory platform, via @surrealdb/mastra-ai. Fact extraction, embeddings, semantic recall, and document search all run server-side: there is no database, vector store, or embedding model to operate. The agent's process holds nothing durable — restart it and everything the team taught is still there, because it never lived in the process.

What you can try

• Teach it something in one conversation — "The staging environment resets nightly at 02:00" — then ask about it in a different thread, as a different teammate. The memory is the common ground, not the chat history.
• Ask a policy question — "Do I need approval for a $150 expense?" — and get an answer grounded in the ingested team FAQ.
• Correct the record — "We moved the weekly demo to Thursdays" — and watch it forget the stale fact and store the new one.
• Kill the dev server and start it again: fresh process, same brain.

Quick start

1. Create the project (or clone this repo directly)

npx create-mastra@latest my-company-brain --template template-surrealdb-company-brain
cd my-company-brain && npm install

2. Add your keys

cp .env.example .env

Set ANTHROPIC_API_KEY plus your Spectron credentials — get those at surrealdb.com/spectron.

3. Seed and run

npm run seed
npm run dev

Open Studio at http://localhost:4111, pick Company Brain, and ask: "What did the Northwind renewal call decide?"

How it works

teammate A ──► ┌───────────────┐        ┌─────────────────────────┐
teammate B ──► │ Company Brain │ ◄────► │ SurrealDB Agent Memory  │
teammate C ──► │    (agent)    │        │ facts · semantic recall │
               └───────────────┘        │ profile · documents     │
                                        └─────────────────────────┘

• src/mastra/agents/company-brain.ts — the agent. SpectronMemory is its Mastra memory provider (facts extracted from every turn, server-side), and createSpectronTools gives it explicit control: spectronRemember, spectronRecall, spectronForget, spectronContext, and spectronSearchDocuments.
• src/mastra/spectron.ts — the one client and context every conversation shares. That single shared context is the company brain.
• scripts/seed.ts — ingests the sample FAQ into the document corpus and plants a few starter facts so the first question has an answer.

Environment variables

Variable	Required	Notes
ANTHROPIC_API_KEY	yes	LLM for the agent
SPECTRON_ENDPOINT	yes	API endpoint origin
SPECTRON_CONTEXT	yes	The memory context — one per brain
SPECTRON_API_KEY	yes	Bearer token

Requirements

• Node.js 22.13+
• Spectron credentials (request access)