KrakenD vs Liveblocks

Liveblocks shipped a six-week run of directional change: a brand-new React Flow SDK for collaborative node-graph editing, a Chat SDK adapter, a Feeds API explicitly framed for "Agent Workflows," a Python SDK for backend use, and v3.18 with breaking changes plus a mutateFlow() API that lets servers mutate diagrams from Node.js. The accompanying blog stream — "AI agents are becoming native users of software," feeds and APIs for agent workflows, n8n nodes — leaves no doubt about where the team is pointing. Underneath, the v1→v2 storage engine migration is in the long tail.

Liveblocks is no longer just "realtime presence and comments for documents." It is positioning itself as the substrate where humans and AI agents share state — diagrams, chat threads, feeds — with the same primitives. The room.history.disable() API for hiding agent-generated writes from the undo stack is the small detail that gives the strategy away: the team is designing for a world where most edits in a room come from agents, and humans need to be shielded from that noise.

The next two minors should bring more agent-shaped APIs (auth scoping for agents, read-only agent participants, server-driven LiveObjects mutations from non-Node runtimes) and broader framework adapters beyond React Flow and Handsontable. Expect Liveblocks to pitch the Chat SDK adapter as a path off OpenAI's hosted threads when teams want agent conversation state stored on their own infra.