dgdao

[matthewmcneely]

We're happy to announce that dgdao is now available at github.com/dgraph-io/dgdao. dgdao is a type-first, transactional database layer for Go, built on Dgraph. You define plain Go structs with dgraph tags, and dgdao gives you ORM/DAO-like Insert/Get/Query mechanics, automatic schema management, and full access to DQL when you need to drop down for advanced queries.

The part you find most useful: dgdao runs in-process. Point the client at a file:// URI and it internalizes Dgraph's server components, writing to a local file-based database with low-latency reads, writes, and vector searches. There's no separate server to stand up, which makes it a natural fit for development, testing, and embedded applications. When you're ready to deploy against a real cluster, change the connection string to dgraph://host:9080 and the same code runs against any remote Dgraph server.

There's also a generics-based typed client for those who want compile-time type safety without code generation. The typed package wraps the base client in a Go generic layer: bind your type once with typed.NewClient[User](client) and you get typed CRUD plus a fluent query builder where Nodes() returns []User directly, with no destination slices or decode steps at call sites. The builder supports composable filters, OrGroup for parenthesized OR scopes, WhereEdge for constraining by a neighbour's scalar over an edge (rendered as a server-side var block so memory stays bounded), auto-paging iteration over a single read-only snapshot, and MultiQuery for batching several queries into one round-trip. It composes on the same primitives as the base client, so you keep full access to Dgraph's query power.

Beyond the basics, dgdao includes schema validation with a Validator hook, a Defaulter hook for before-write defaulting, automatic retries, and vector search over embeddings. Auto-schema (dg.WithAutoSchema(true)) derives predicates and indexes directly from your struct tags, so your Go types stay the single source of truth.

A good fit for AI-assisted development

If you're building Go applications with coding agents, dgdao removes a lot of friction. An agent can write code against the typed client, spin up a real database with nothing more than a file:// URI, and run the full test suite locally, with no containers, servers, or credentials to configure. Struct tags keep the schema in the code itself, so the agent sees the whole data model in one place, and the compiler catches a large class of mistakes that would otherwise surface at runtime. And because the same code runs unchanged against a remote cluster, what an agent builds and verifies locally scales to Dgraph's well-documented distributed graph capabilities, including horizontal scaling, replication, and multi-tenancy, just by swapping the connection string.

Companion repos

There are also companion repos under the dgraph-io org, each an independent Go module you can adopt on its own:

• dgdao-gen: code generator and wrapper-entity runtime. Run go generate against your structs and get a fully typed client, query builders, and auto-paging iterators.
• dgdao-migrate: struct-first schema migrations, managed as an ordered chain of run-once revisions with checksum-enforced immutability and drift gates.
• dgdao-telemetry: OpenTelemetry instrumentation emitting a client span per database operation.

The README has a copy-paste quickstart, and the examples/ directory covers common patterns, including loading the 1million movie RDF dataset for benchmarking. One caveat: dgdao targets POSIX systems and isn't guaranteed to work on Windows. Everything is Apache 2.0 and contributions are welcome.