Databricks Buys Electric to Give AI Agents Local Databases

Databricks has acquired Electric, the creator of the WebAssembly-based Postgres project PGlite and the Electric sync engine, for an undisclosed sum. The Electric team will integrate with the Neon team inside Databricks, which was previously acquired for approximately $1 billion to power the company's Lakebase platform. While Electric Cloud is shutting down, its open-source projects—including PGlite, Postgres Sync, TanStack DB, and Durable Streams—will remain open to the public.

This acquisition addresses a massive shift in how autonomous AI agents interact with data. Unlike traditional applications that rely on a single, centralized database, agentic workloads require rapid, local context. Databricks reports that AI agents on its Lakebase platform already generate about four times more databases than human users. A typical project maintains around 10 database branches, with some scaling to more than 500 branch iterations, and certain workloads feature database compute sessions lasting under 10 seconds.

To handle these ephemeral workloads, Electric developed PGlite, a complete Postgres database compiled to WebAssembly. Gzipped to under 3MB, PGlite runs directly inside agent sandboxes without requiring an external server, eliminating the latency of remote database connections. Packaged as a TypeScript library, it supports the browser, Node.js, Bun, and Deno. The technology has seen explosive growth, with weekly downloads surging from 1 million to 13 million over the past 12 months.

For AI developers, this acquisition means a shift toward decentralized, per-agent database architectures. Instead of managing complex, high-latency connections to a remote server for every reasoning step, practitioners can spin up thousands of disposable, local databases that run in-process. These lightweight instances can then seamlessly sync their state back to a central lakehouse, allowing parallel agent groups to collaborate without relying on stale data or duplicating efforts.