BlitzGraph is an AI-native graph database backend, designed to let agents model the real world as a graph and operate on it through typed JSON queries and mutations, thereby avoiding stitching together SQL, JOINs, and custom ORM code.
Purpose
BlitzGraph's core purpose is to serve as the backend data layer for AI agents. It allows agents to programmatically compose typed JSON queries, rather than generating SQL strings or guessing the shape of results. Its query language, BQL (BlitzGraph Query Language), is structured JSON that agents can write correctly.
Main Capabilities
- Multi-kind entities: A single entity can have multiple kinds at the same time; for example, a User can be both an Admin and a Moderator, with no role table or migration required. Entities can evolve by gaining or losing kinds.
- Bidirectional relationships: Relationships are stored in both directions; querying "who wrote this article" and "what did this user write" have the same cost and index, both O(1), with no reverse lookup table or additional query required.
- Typed JSON queries (BQL): Agents compose query objects rather than SQL strings. Supports filters, nested expansion ($expand), projection, and full-text search, completed in a single request with zero N+1 problems.
- Rich content types: Supports types such as EMAIL, URL, DATE, JSON, FLEX, and more, not just varchar, with validation built in at the database level.
- Referential integrity: Cardinality constraints and onDelete policies (cascade, restrict, unlink) are enforced at the engine level, keeping the graph consistent by default.
- Built-in full-text search: A native BM25 engine, supporting typeahead, prefix, exact, and all-stems modes, with no Elasticsearch or external service required, and usable within graph traversals.
- Smart transactions: Mutation operations are topologically sorted and validated on the final result, rather than row by row. Business rules check the final state of the entire transaction, supporting complex multi-entity operations and guaranteeing all-or-nothing.
- In-database business logic: Validation, computed fields, transformations, and effects are all defined in the schema; business rules live together with the data store, rather than being scattered across middleware and application code.
How to Use
- Connect via MCP server: Add BlitzGraph as a remote MCP server, and you can use the same live backend from agents such as Claude or Codex. After adding the MCP server, authentication runs automatically; the user only needs to log in once in the browser, and afterward the agent can use the tools.
- Query language: The agent or application code builds typed JSON objects; the query itself is the data structure, with no generation step or parsing ambiguity. It supports a GraphQL-like read approach, but in the form of typed JSON, allowing you to fetch root entities, expand relationships, and project nested fields in a single request.
- Schema definition: Define entities, kinds, relationships, validation, computed fields, and more in the schema.
Applicable Scenarios
- AI agents need to interact with structured data but do not want to deal with the complexity of SQL or ORMs.
- Need to model complex relationships, multi-kind entities, and dynamically evolving data models.
- Applications that need built-in full-text search, referential integrity, and transactional consistency.
- Scenarios where you want to centralize business logic at the database layer and reduce scattered application code.
Limitations
- The material does not provide specific information about pricing, platform support, performance benchmarks, scalability limits, and so on.
- The material does not provide detailed installation steps or code examples (other than adding the MCP server).
- The material does not provide information about operations aspects such as data migration, backup, and security policies.
