rushdb.com

What It Is

RushDB is a "persistent memory layer" designed for AI Agents and AI applications. It allows developers to directly write JSON or CSV, converting messages, tool results, documents, entities, events, etc., into Records, while automatically preserving field types, nested structures, searchable text, and graph relationships. Its core value is not simply replacing vector databases, but rather unifying structured data, semantic retrieval, relationship traversal, and live schema into a single backend, thereby reducing the need for synchronization and stitching across application databases, Redis, vector databases, and graph databases.

Core Capabilities

In terms of AI capabilities, RushDB supports semantic search and managed embeddings, while also allowing the use of external vectors. During queries, you can first apply exact filtering using labels and where clauses, then perform similarity sorting, and further traverse relationships. The Ontology API can expose the current project's labels, field types, sample values, ranges, relationship directions, and vector index status in Markdown or JSON. This is ideal for letting an Agent plan queries based on the actual structure before calling tools, rather than guessing fields via prompts. It's worth noting that parent-child relationships within nested JSON can be automatically preserved, but domain relationships typically still need to be explicitly written by the application. For relationships in flat data, it is recommended to rely on the project's configured LLM, and they only take effect after approval.

Pricing & Integration

Pricing is metered by KU (Knowledge Unit), emphasizing "pay for writes, reads are free." The Free plan offers 100K KU/month, 2 projects, with no time limits and no feature restrictions; Pro starts at $24/month, including 10M KU/month, with overages at $3/M KU; Scale starts at $73/month, targeting high-volume scenarios with SLA and priority support. For integration, it supports REST API, TypeScript/Python SDK, MCP Server, SearchQuery, and Relationship API, along with self-hosting, BYOC, and connections to Neo4j or Aura Vector.

Pros, Cons & Who It's For

The main advantage is a clear engineering path: simply writing JSON creates a searchable, traversable context that Agents can understand. Simultaneous graph and vector querying is highly practical for GraphRAG, knowledge bases, product search, and Agent memory. The limitation is that KU costs can grow with attributes, links, embeddings, and deep queries, so actual bills need to be evaluated based on your data patterns. The quality of the Chinese interface, Chinese documentation, and Chinese semantic retrieval is not disclosed. Additionally, writes are paused once the free tier limit is reached.

Access from China

The article does not provide information regarding access from mainland China, payment methods, or local compliance, so china_access can only be determined as unknown. If network or payment restrictions apply, consider self-hosting a combination of traditional databases + vector databases + graph databases, or opt for domestically available vector retrieval, knowledge graph, and Agent Memory solutions as alternatives.