TL;DR — Key Takeaways
- Databricks is acquiring Electric to bring lightweight, locally running Postgres databases and synchronization technology into its agentic AI database strategy.
- Electric’s PGlite lets AI agents run a full Postgres database inside browsers, Node.js processes and sandboxes without constantly connecting to a remote database.
- Electric will join Neon, strengthening Databricks’ Lakebase architecture for rapidly created, short-lived database environments used by AI agents.
Databricks is acquiring database startup Electric, adding technology that allows AI agents to run a lightweight Postgres database locally while synchronizing data with a central database.
The acquisition gives Databricks two key Electric technologies: PGlite, a Postgres database built with WebAssembly (WASM), and Electric’s synchronization engine. Together, they allow an AI agent to work with data inside its own application or sandbox instead of repeatedly connecting to a remote database.
The Electric team will join Neon, the serverless Postgres company Databricks acquired for about $1 billion in 2025. Neon became the foundation for Lakebase, Databricks’ Postgres-based database service, which launched in February.
Electric’s technology addresses a database challenge created by agentic AI. Traditional applications typically connect many users or processes to a central database. AI agents can operate differently, creating short-lived environments, performing numerous database operations and discarding those environments when a task is complete.
For Databricks, the strategy here is that AI agents will benefit from a database architecture built around large numbers of small, temporary environments rather than a single long-running database. Electric gives the company technology designed specifically for that model. Keeping data close to an agent could reduce network traffic and latency, particularly for complex tasks that require frequent database access. It could also lower costs by reducing remote database calls and the need to maintain a managed database instance for every agent.
Still, there are potential downsides to the architecture. If large numbers of agents maintain temporary local databases, companies will need policies governing which data can enter those environments and how data is deleted when an agent finishes its work.
Auditing will likely become more complicated. Compliance controls that were once focused primarily on a centralized database may need to extend into short-lived agent environments. Synchronization introduces another potential problem: multiple agents could act on different versions of data, leaving IT teams to identify and resolve conflicts.
Terms of the deal were not disclosed.
Agents Drive Database Branching
Databricks says agents on Lakebase already create about four times as many databases as human users. An average project has roughly 10 database branches, while some projects have gone through more than 500 branch iterations. For certain Lakebase applications, database compute lasts less than 10 seconds on average.
That usage pattern creates demand for databases that can be launched quickly and at low cost. PGlite provides a full Postgres database that can run in a browser, Node.js process or agent sandbox. It also supports extensions like pgvector. PGlite has seen remarkable growth, the companies claim: over the last year it zoomed from 1 million to 13 million weekly downloads.
Electric’s sync technology connects these local databases with a central Postgres system. In Databricks’ planned architecture, an agent could create a sandbox, run PGlite locally, perform dev or testing work, then sync relevant data with a Lakebase branch. Again, temporary databases can be discarded when they are no longer needed.
There is a technical connection between Electric and Neon as well. PGlite originated from work by Neon co-founder Stas Kelvich, who experimented with compiling Postgres to WASM. Electric subsequently developed that work into PGlite.
Electric’s founders said its open-source projects will remain open source, including PGlite, Electric’s sync engine, Durable Streams and TanStack DB. Electric Cloud, however, is being shut down, requiring customers of the hosted service to self-host or select another provider.