Skip to content
AI Primer
release

Electric Agents introduces sync-based multi-agent platform with shared sessions and forking

ElectricSQL launched Electric Agents, treating agents as long-lived data entities that sync across shared coding sessions, swarms, and branches. The release matters for teams building collaborative agent systems that need durable state and coordination primitives, not just one-shot task runners.

4 min read
Electric Agents introduces sync-based multi-agent platform with shared sessions and forking
Electric Agents introduces sync-based multi-agent platform with shared sessions and forking

TL;DR

You can read the launch post, skim the product page, and jump straight to the quickstart. The weirdly useful bit is that Electric is selling agent sessions as durable streams you can observe, share, and fork, not just as background jobs that happen to call an LLM. The launch videos also make the product angle obvious: shared coding sessions, agent swarms, and branchable sessions.

Durable streams

Electric's big bet is that agent state should sit outside the running process. The launch post says that move makes sessions persistent, addressable, observable, resumable, shareable, composable, and forkable.

The docs overview makes that concrete: every agent is an entity at /{type}/{id}, and its session lives on a durable Electric Stream. That gives Electric a clean distinction between durable state and ephemeral execution, with agents sleeping when idle and waking when something happens.

Runtime primitives

The product page breaks the system into a small set of primitives:

  • Spawn and send: create an entity, then wake it by delivering a message.
  • Observe: subscribe to another entity's stream and react to new events without polling.
  • Shared state: let multiple entities read and write against the same stream-backed state.
  • Replay on wake: rebuild context from the stream instead of checkpointing a long-running process.

The docs add the wake conditions: entities run when a message arrives, a child finishes, state changes, or a scheduled time elapses. Under the hood, Electric says the runtime SDK sits on Electric Streams, TanStack DB, and pi-agent-core.

Shared sessions and forks

The launch materials are much more collaborative than most agent-platform announcements. Electric is pitching shared interactive coding sessions, swarm-style coordination, and branchable sessions as default patterns, not edge cases.

That lines up with the product claim in ElectricSQL's launch thread that agents are long-lived logical entities you can share with users, devices, and other agents. The demos matter because they show what Electric thinks the durable stream buys you: multiple participants watching the same session, parallel exploration through forks, and coordination without keeping every agent hot all the time.

Quickstart and built-ins

The quickstart fills in the practical details the launch thread skips. You need Node.js 18+, Docker, and an Anthropic API key, with an optional Brave Search key for web search.

One command starts Postgres, Electric, the runtime server, and a local web UI at http://localhost:4437. It also launches a built-in Horton runtime and registers three entity types out of the box: horton, worker, and coder.

The quickstart also exposes the initial CLI surface:

  • npx electric-ax agents quickstart
  • npx electric-ax agents spawn /horton/onboarding
  • npx electric-ax agents send /horton/onboarding 'Walk me through Electric Agents'
  • npx electric-ax agents observe /horton/onboarding

That last detail is new information compared with the launch videos: Electric is not just shipping a concept and a control plane, it is shipping a local, inspectable onboarding loop with a built-in assistant and observable streams on day one.

Share on X