VanillaDB

We explore agents,
memory, and what
data can become.

VanillaDB is an open-source project building database infrastructure for the agentic era. We're interested in how agents read, remember, and act on operational data, and in making that infrastructure simple enough to actually understand.

Agent memory
Agents that can't remember degrade over time. We build the infrastructure layer that lets agents persist, retrieve, and reason over structured context across sessions, locally and without accounts. Memory shouldn't require a cloud service.
VanillaGraph · Memory Layer
Operational intelligence
Raw operational data (logs, traces, decisions, documents) is the most underused resource in agent systems. We explore how to turn it into queryable, structured knowledge that agents can navigate and reason over in real time.
Data Normalization · Benchmarking
Open tooling
Everything we build is MIT-licensed and public. Our releases are intentionally vanilla: minimal opinions, clear interfaces, designed to be extended. We believe infrastructure that stays simple is infrastructure people can actually own and customize to their needs.
MIT Licensed · Fork anything
How we work

Intentionally vanilla.

01 Simple interfaces over rich ones.
02 Composable parts over monolithic systems.
03 Code you can read over code that just works.
04 Open source, always.

The name isn't arbitrary. Our releases ship minimal by design. We don't add opinions you didn't ask for, abstractions you can't see through, or defaults you can't change. Every component we build is meant to be understood in full, then customized to fit your actual problem.

We believe the right abstraction makes a hard problem feel obvious. If something we ship doesn't make sense on a first read, that's a bug in the design, not an acceptable tradeoff.

Everything is public. We discuss roadmap decisions openly, share early builds with the community first, and welcome contributions. If you're building on or with VanillaDB, or thinking through similar problems, we want to hear about it.

Our releases are starting points. We expect teams to fork, extend, and replace pieces as their needs evolve. That's the intended use. Build something vanilla, then make it yours.