Spectron
Spectron is SurrealDB’s persistent agent memory layer, embedded directly into the SurrealDB multi-model database.
Publisher review
Spectron is SurrealDB’s persistent agent memory layer, embedded directly into the SurrealDB multi-model database. It is designed for developers building AI agents that need memory to be transactional with application state, avoiding the consistency and latency problems of external memory middleware. The target audience includes teams working on conversational AI, autonomous agents, and event-driven systems that require ACID guarantees across documents, graphs, vectors, time-series, and relational data. By unifying memory and data in one database, Spectron aims to eliminate the operational burden of managing separate vector stores, graph databases, and authentication services.
Spectron operates as a native layer within SurrealDB, allowing agents to write memory and application state in a single ACID transaction. It supports hybrid retrieval methods that combine vector similarity, graph traversal, BM25 full-text search, and keyword matching, enabling nuanced context retrieval without external services. The platform includes built-in RBAC and record-level permissions, so the same auth policies govern both memory and data. SurrealDB Cloud offers a free tier with 1GB storage and 0.25 vCPU/512MB compute, scaling to production clusters with multi-region disaster recovery. The database also supports tri-temporal facts (valid time, transaction time, and decision time) and multi-tenant scopes, which are useful for audit-heavy or compliance-sensitive agent workloads.
Spectron competes directly with Mem0, Zep, and Letta (MemGPT), which all implement agent memory as middleware sitting between the agent and a database. SurrealDB argues that this architecture introduces a “fragmented memory tax”: writes to memory and state can diverge (no atomicity), every memory operation adds a round-trip through middleware (two hops instead of one), and developers must manage separate auth models. Spectron eliminates this by making memory a first-class database feature. However, Mem0 and Zep offer more specialized memory extraction (e.g., LLM-driven entity extraction and temporal knowledge graphs) that SurrealDB does not yet match out of the box.
The honest trade-off: Spectron is still maturing. SurrealDB’s embedded mode has been reported as “incredibly slow compared to SQLite (like many orders of magnitude)” per a Reddit user, though the company has published new benchmarks for v3.x claiming significant improvements. Some features (database branching, object-storage-backed data, distributed live queries) are marked as “coming soon.” Documentation and community resources are thinner than those for established databases, and the Scale tier pricing is not yet published. Teams that need a proven, high-performance memory store today may find Mem0 or Zep more battle-tested, while those willing to bet on a unified transactional memory layer will find Spectron’s architecture compelling.
How it works
-
Unified multi-model database
Supports document, graph, vector, time-series, and relational data models in a single database, eliminating the need for separate stores.
-
ACID transactions across models
All data models participate in atomic, consistent, isolated, and durable transactions, preventing data divergence between memory and state.
-
Persistent agent memory layer
Spectron provides a native memory layer where agents write memories directly into SurrealDB, with ACID guarantees and unified permissions.
-
Hybrid retrieval methods
Combines vector similarity, graph traversal, BM25 full-text search, and keyword matching for nuanced context retrieval without external services.
-
Built-in authentication and authorization
Includes RBAC and record-level permissions that apply to both memory and data, eliminating the need for a separate auth service.
-
Real-time and event-driven capabilities
Supports live queries and event-driven triggers, enabling agents to react to data changes in real time without polling.
-
Tri-temporal facts and multi-tenant scopes
Stores valid time, transaction time, and decision time for each fact, and supports isolated multi-tenant namespaces for compliance.
Strengths and trade-offs
Strengths
- Unified transactional layer across five data models (document, graph, vector, time-series, relational) means no separate stores to sync.
- Spectron eliminates middleware latency by allowing agents to write memory and state in one ACID transaction with a single round trip.
- Hybrid retrieval combining vectors, graph, BM25, and keywords provides flexible context retrieval without external vector or search services.
- Built-in RBAC and record-level permissions apply uniformly to memory and data, removing the need to manage separate auth models.
Trade-offs
- Embedded mode performance has been reported as significantly slower than SQLite, though SurrealDB claims improvements in v3.x benchmarks.
- Several features (database branching, object-storage-backed data, distributed live queries) are still marked as 'coming soon' and not yet available.
- Public documentation and community resources are thinner than those for established databases like PostgreSQL or specialized memory tools like Mem0.
- Scale tier pricing is not yet published, making it difficult to estimate production costs for fault-tolerant deployments.
Pricing context
Free tier: $0/month with 1GB storage, 0.25 vCPU/512MB compute, and free network (then $0.021/hr for burst). Scale tier: pricing coming soon, with multi-AZ, horizontal scaling, and multi-region disaster recovery. Enterprise: custom pricing for on-premises or private cloud deployments with FIPS-compliant cryptography and SLA-backed uptime.
Getting started with Spectron
-
Sign up for SurrealDB Cloud
Go to surrealdb.com and create a free account. Select the Free tier to get 1GB storage and 0.25 vCPU with 512MB compute. This gives you a cloud-hosted instance to start building with Spectron's agent memory layer.
-
Connect your database instance
From the SurrealDB Cloud dashboard, copy your instance endpoint and authentication credentials. Use the SurrealDB SDK for your language (e.g., JavaScript, Python) to establish a WebSocket or HTTP connection to the database.
-
Define memory schema and scopes
Create a table for agent memories with fields for content, embedding, and metadata. Use SurrealQL to define multi-tenant scopes and RBAC roles so that memory access is governed by the same permissions as your application data.
-
Write agent memory in a transaction
In your agent code, start a SurrealDB transaction. Insert a memory record with a vector embedding and application state in one atomic operation. Commit the transaction to ensure both memory and state are saved together with ACID guarantees.
-
Query memories with hybrid retrieval
Use SurrealQL to perform a hybrid search combining vector similarity, graph traversal, and BM25 full-text search. For example, run a query that finds relevant memories by embedding distance and keyword match, then use the results to inform your agent's next action.
Frequently Asked Questions
What is Spectron in SurrealDB?
Spectron is SurrealDB's persistent agent memory layer embedded directly into the multi-model database. It lets AI agents write memory and application state in a single ACID transaction, avoiding the consistency and latency issues of external memory middleware.
How does Spectron handle agent memory differently from Mem0 or Zep?
Unlike Mem0 and Zep, which run as middleware between the agent and a database, Spectron makes memory a native database feature. This eliminates the fragmented memory tax by ensuring atomic writes, single round trips, and unified permissions across memory and data.
What retrieval methods does Spectron support for agent context?
Spectron supports hybrid retrieval combining vector similarity, graph traversal, BM25 full-text search, and keyword matching. This allows nuanced context retrieval without needing external vector stores or search services, all within a single ACID transaction.
What are the pricing options for Spectron and SurrealDB?
SurrealDB Cloud offers a free tier with 1GB storage and 0.25 vCPU/512MB compute. The Scale tier pricing is coming soon, with multi-AZ and multi-region disaster recovery. Enterprise plans include custom pricing for on-premises or private cloud with FIPS compliance.
Is Spectron ready for production use compared to Mem0 or Zep?
Spectron is still maturing. Embedded mode performance has been reported as slower than SQLite, though v3.x benchmarks show improvements. Some features like database branching are coming soon. Teams needing proven performance may prefer Mem0 or Zep today.
What are the main benefits of using Spectron for AI agents?
Spectron unifies memory and data in one database, eliminating separate stores. It provides ACID transactions, hybrid retrieval, built-in RBAC, and tri-temporal facts. This reduces operational complexity and ensures consistency between agent memory and application state.
Alternatives
How Spectron compares
Direct head-to-head against 2 competitors. Picked by 7wData.
Spectron
- Pricing
- Free tier: $0/month with 1GB storage, 0.25 vCPU/512MB compute, and free network (then $0.021/hr for burst). Scale tier: pricing coming soon, with multi-AZ, horizontal scaling, and multi-region disaster recovery. Enterprise: custom pricing for on-premises or private cloud deployments with FIPS-compliant cryptography and SLA-backed uptime.
- Target
- Spectron is SurrealDB’s persistent agent memory layer, embedded directly into the SurrealDB multi-model database.
- Strength
- Unified transactional layer across five data models (document, graph, vector, time-series, relational) means no separate stores to sync.
- Watch for
- Embedded mode performance has been reported as significantly slower than SQLite, though SurrealDB claims improvements in v3.x benchmarks.
The Micromanipulator Company
- Pricing
- Custom/Contact sales
- Target
- Semiconductor testing equipment
- Deployment
- Industrial probe stations
- Strength
- Specialized probe tips for microelectronics
- Watch for
- Limited consumer audio applications
Nexgenix
- Pricing
- Custom/Contact sales
- Target
- Global electronic components
- Deployment
- B2B distribution networks
- Strength
- High-reliability semiconductor sourcing
- Watch for
- Focus on commercial/industrial markets
User reviews
No user reviews yet. Be the first to write one.
Sources
Reporting on this tool draws on these publicly available sources.