Company
We're building the memory layer for AI
Make persistent AI memory as easy to add as a database.
We were building an AI chatbot and spent two weeks setting up vector databases, chunking, embeddings, and retrieval. Then we rebuilt the exact same thing for the next project. And the next. That's when we started Databaset: so no developer has to build AI memory infrastructure twice.
2025
Founded
12M+
Memories processed
41ms
Median recall
Remote-first
Team
Timeline
Q3 2025
Prototype
First internal chatbot with pgvector + custom chunking.
Q4 2025
Private beta
10 teams onboarded. Dashboard and Node SDK shipped.
Jan 2026
Public launch
Free tier, Python SDK, docs site, and self-host preview.
H1 2026
Roadmap
Webhooks, memory expiry controls, Go SDK, SOC2 Type I.
Team
Founding team
Engineering
Previously built AI products at startups where memory was always the bottleneck.
Developer experience
Docs & SDKs
Obsessed with API design, error messages, and copy-paste quickstarts.
Infrastructure
Platform
pgvector, embedding pipelines, and the boring reliability work.
Values
Developer first
If we wouldn't use it ourselves, we don't ship it.
Radical simplicity
Two methods beat twenty knobs. Complexity is a bug.
Transparent
Pricing, limits, and failures are visible. No dark patterns.
Open core
Core SDK is MIT. Managed cloud funds the hosted product.