BlitzGraph says it is an AI-native backend that models data as graphs and lets agents send typed JSON queries instead of SQL. The project includes built-in validation, full-text search, transactions, and a remote MCP server for Claude and Codex.
BlitzGraph is a new backend project that says it is designed for LLM agents as much as for humans. The company describes it as an “AI-native backend” where developers model data as graphs and agents issue typed JSON queries instead of SQL strings.
According to the project page, BlitzGraph is free to try with live data and no account needed. It also offers a remote MCP server, which means agents such as Claude or Codex can connect to the same backend through the Model Context Protocol.
This is news because it shows another attempt to build a database backend specifically for AI agents, not just for human developers. MCP, or Model Context Protocol, is a standard that lets tools plug into agents like Claude and Codex. For non-specialists, the key idea is that the system tries to make data access safer and more structured for software that generates its own queries.
If you build self-hosted agents, pay attention to the typed-query approach here: it is a cleaner fit than having agents assemble SQL strings. For your own stack, prefer tools that accept structured JSON or schema-checked requests, because that reduces parsing mistakes and bad query generation. If you expose a backend to agents, test whether it can enforce validation, relationship rules, and transaction boundaries at the data layer instead of relying on prompts or app code.
The product positions itself as an alternative to systems that still think in tables, columns, or documents. BlitzGraph says it models “reality, not tables,” using entities that can belong to multiple kinds at once. In its example, a user can be a user, admin, and moderator at the same time without requiring extra role tables or migrations.
The company also says relationships are bidirectional by default. That means a query can ask both “Who wrote this post?” and “What did this user write?” with the same cost and index, rather than relying on reverse lookup tables or extra joins.
Queries are built as JSON objects, not SQL strings, according to the documentation. BlitzGraph calls this language BQL. The company says BQL supports filters, nested expands, projections, and full-text search in one request, and that it avoids the “N+1” query problem common in app code and ORMs.
BlitzGraph also says it includes built-in data types such as EMAIL, URL, DATE, JSON, and FLEX, plus validation at the database level. Referential integrity rules, including cardinality and onDelete behavior such as cascade, restrict, and unlink, are enforced by the engine.
For search, the project says it has a native BM25 engine with typeahead, prefix, and exact modes. BM25 is a common ranking method used in search systems. BlitzGraph says this search runs inside the graph engine without needing Elasticsearch or another external service.
The backend also includes what it calls “smart transactions.” According to the site, mutations are topologically sorted and validated against the final state of the transaction, not checked one line at a time. The company says that makes multi-entity changes easier to handle while keeping data consistent.
BlitzGraph says business logic can live in the schema itself through validations, computed fields, transforms, and effects. The project contrasts that with application designs where business rules are spread across middleware and separate code paths.
The product page also compares BlitzGraph with Supabase, Convex, MongoDB Atlas, and Firebase. It says the system combines graph, document, and relational patterns in one engine, but it also says some areas remain behind those more established tools, including realtime queries, production history, and community size.
The project’s examples show how agents might use it directly. One example uses typed JSON to fetch entities, expand relationships, and calculate computed fields in a single request. Another shows sessions with related memories pulled through graph traversal and filtered by relevance.
BlitzGraph is available through its website, and the company says auth runs automatically after a user signs in once in the browser. After that, the agent can use the tools through the MCP server.
