TanStack AI now has subagents 🤖 Define agents once, pass them to chat(), and each one streams its own card in your UI. Jev routes every message to the right agents, in parallel or in order⚡

Find the docs here:

Feels like Tanstack AI is the only Tanstack library receiving updates 😞

love the parallel streaming cards way nicer than one long assistant blob how does Jev decide order vs parallel routing?

This is a clean example of orchestration moving into the framework layer. Define specialized agents once, route work by intent, run them in parallel or sequence, and keep their outputs visible in the UI. That’s a much more useful abstraction than “one agent does everything.”

Man @typesafeai was really the missing piece we needed all along

The UI card is useful only if it carries execution context: input slice, tools called, and exit condition. Otherwise parallel cards just turn one opaque blob into several smaller opaque blobs.

Jev picks parallel when no child needs another child's text, and sequence only when a later agent must read the earlier one.

fan-out is where determinism dies quietly: each subagent should read a frozen input slice taken at spawn, not the shared live conversation. otherwise results depend on who read when, and the same message lands in different cards on a rerun. freeze the inputs, merge the outputs.

One card per agent streaming in parallel is a nice default. The hard part for users is knowing which card to trust when two disagree. Any plans for a merged summary view?

Streaming each specialist’s card could make parallel agent workflows easier to debug.