Version Packages

[github-actions]

This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.

Releases

@mobile-reality/mdma-agui@0.3.0

Minor Changes

• 16fe1f2: Surface agentic activity — tool calls, run steps, and reasoning streams — as an ordered feed via the
  onActivity option and bridge.activity (also returned from useMdmaAgentStream). Activity is
  deliberately kept out of the MDMA document store, so agent chatter and rendered components stay
  decoupled: render it as a timeline beside the documents, or ignore it entirely. Each MdmaActivity
  carries a stable id (tool-call id / step handle / reasoning message id), a kind
  ('tool' | 'step' | 'reasoning'), a label, a status ('running' | 'done'), and streamed
  detail — accumulating tool args, the tool result, or the reasoning text.

• 16fe1f2: Accept MDMA on a dedicated CUSTOM event channel alongside inline assistant text. A backend that
  would rather not interleave documents into visible prose can emit
  { type: 'CUSTOM', name: 'mdma', value } (the name is exported as MDMA_CUSTOM_EVENT_NAME), where
  value is either the markdown string or { messageId?, markdown }. Both channels feed the same
  parse/store/render pipeline — out-of-band text is parsed immediately rather than throttled, since it
  arrives complete — and each message reports where it came from via message.source
  ('text' | 'custom'). Keeping documents off the prose channel means no markup leaks into the chat,
  which is what a tool-calling agent wants when the document lives in a tool argument.

• 16fe1f2: Resume parked runs through AG-UI's native interrupt primitive. When a run finishes with an
  interrupt outcome the bridge exposes the pending set as bridge.interrupts and fires
  onInterrupt; answering the component an interrupt refers to now resolves that interrupt with
  runAgent({ resume }), so the parked run continues with its state intact instead of starting a
  fresh turn.

  The new resumeMode option selects the strategy:

  • 'auto' (default) — resolve a matching interrupt if the run is parked on one, otherwise fall back
    to a fresh user turn.
  • 'interrupt' — only ever resolve a matching interrupt; if none matches, do nothing.
  • 'user-turn' — always open a fresh user turn (addMessage + runAgent).

  Behavior change: a user decision previously always opened a fresh user turn. Under the new
  'auto' default it will resolve a matching interrupt when the run is parked on one. Pass
  resumeMode: 'user-turn' to keep the previous behavior. Returning false from onAction, or
  supplying resume, still overrides resumption entirely.

• 16fe1f2: Track the agent's shared state (STATE_SNAPSHOT / STATE_DELTA, including JSON-patch deltas) as a
  componentId → values map, exposed via the onState option and bridge.state, and use it to
  hydrate MDMA stores. MDMA components are headless — a document describes intent and takes its values
  from state — so this is what lets a form the agent renders come up pre-filled from what it
  already knows.

  Hydration is reactive: state arriving after a component is already on screen is dispatched into
  that live store too, so the agent can set a field the user is currently looking at without
  re-rendering the component.

@mobile-reality/mdma-demo-native@0.0.4

Patch Changes

• Updated dependencies [16fe1f2]
• Updated dependencies [16fe1f2]
• Updated dependencies [16fe1f2]
• Updated dependencies [16fe1f2]
  • @mobile-reality/mdma-agui@0.3.0

mdma-example-agui@0.0.1

Patch Changes

• Updated dependencies [16fe1f2]
• Updated dependencies [16fe1f2]
• Updated dependencies [16fe1f2]
• Updated dependencies [16fe1f2]
  • @mobile-reality/mdma-agui@0.3.0