Databricks buys Electric to power AI agent sandboxes

Databricks has acquired Electric, bringing its WASM-based Postgres technology into Databricks' push to support AI agent sandboxes.

Electric's technology centres on PGlite, a lightweight version of Postgres designed to run inside an application, browser tab, user device or agent sandbox rather than on a separate server. The acquisition extends Databricks' Postgres offering beyond its lakehouse environment into the more distributed settings where AI agents operate.

The move reflects a broader shift in software development as companies experiment with agentic applications. Unlike traditional applications, AI agents may decide at runtime what data they need, update their working context frequently, and run in sandboxed environments where constant calls to a central cloud database can slow execution.

Databricks is positioning Electric's software as a way to address that problem. Under the approach it outlined, each agent can run its own local Postgres instance through PGlite while synchronising state back to a central Lakebase Postgres system.

Sync model

A key part of Electric's offering is a real-time synchronisation engine designed to keep distributed agents aligned with a central record. The model is aimed at multi-agent systems in which several agents work in parallel and need to share context quickly to avoid acting on stale information or duplicating work.

Databricks argues that this architecture fits a new generation of software in which developers are moving beyond conventional application designs. In that model, a single managed database may no longer be enough because agents need both durable, governed data and a fast-changing local context.

Electric had already gained traction among developers through PGlite. Weekly downloads grew from 1 million to 13 million over the past 12 months, suggesting significant uptake for a product that pushes Postgres into environments closer to where code runs.

Postgres focus

The acquisition also deepens Databricks' investment in Postgres as it expands its database strategy. Lakebase, its managed Postgres offering, is designed to provide centralised database infrastructure, while Electric adds a smaller-footprint implementation for edge and embedded use cases.

Both products are tied to the growing use of Postgres in AI software. Databricks describes Postgres as an increasingly common foundation for AI agents, particularly where developers want a standard database model across central infrastructure and local execution environments.

Electric's technical heritage also connects to the wider Postgres ecosystem. Databricks said PGlite was built on foundational WASM Postgres work by Stas Kelvich, a co-founder of Neon, before Electric turned the concept into an embeddable Postgres product used in millions of projects each week.

Infrastructure race

The transaction comes as database and cloud software providers compete to define the infrastructure stack for AI systems. Much of that contest has focused on models, orchestration tools and vector databases, but the handling of state, context and synchronisation between agents is emerging as another area of competition.

For Databricks, bringing Electric in-house adds technology aimed at a specific operational challenge in multi-agent software: letting agents work locally with low latency while preserving a central source of truth. That could prove relevant for developers building collaborative systems in which many agents operate at once across different environments.

Databricks said the combination of Lakebase and Electric would allow developers to build agentic applications on a single Postgres standard, run Postgres directly inside the agent sandbox, and synchronise data back to central infrastructure stored on object storage.

Databricks framed the deal as part of a broader transition in software architecture: "The world is building a new era of agentic applications which require distributed state and real-time data synchronisation between teams of agents working in sandboxes."

It added: "That vision now continues at Databricks as we bring WASM Postgres to AI agent sandboxes, extending Databricks' Postgres capabilities from the lakehouse to the edge."