Open Brain
VercelEngineering PostOfficial Source

Chat SDK agents can now stream generative UI with json-render

Chat SDK agents can stream typed JSON specs into platform-native cards through json-render. Schema validation and platform limits help catch malformed or oversized cards before they reach users.

Vercel · Nov 6, 2026
Open Source Open MarkdownOpen JSON
Source Summary

With **json-render**, a Chat SDK agent emits a JSON spec from a typed component catalog, which is rendered as a native card such as Slack Block Kit or Teams Adaptive Cards. Cards can include tables, charts, and buttons.

Practical Implication

Builders can stream a card as content arrives; updates are sent at most every **500 ms** by default. Validate generated specs with `toCard`, and pass platform limits to reject cards that exceed an adapter’s capacity.

Agent-Ready Context
With **json-render**, a Chat SDK agent emits a JSON spec from a typed component catalog, which is rendered as a native card such as Slack Block Kit or Teams Adaptive Cards. Cards can include tables, charts, and buttons.

Builders can stream a card as content arrives; updates are sent at most every **500 ms** by default. Validate generated specs with `toCard`, and pass platform limits to reject cards that exceed an adapter’s capacity.

Without explicit limits, adapters may degrade excess content; for example, Slack converts a second table into a code block. Validation can return issues for correction, but does not guarantee the model will produce a valid replacement.
Connected Context · Feed7 Judgment

This gives Chat SDK agents a typed path from generated JSON to native interactive cards, with streaming, validation, and adapter limits. Validation can flag invalid or oversized specs, but does not ensure the model will correct them; platform-specific capacity still shapes what renders.

Context Map
toolscoding#agent-sdks#dev-ux#tool-use
Uncertainty
Without explicit limits, adapters may degrade excess content; for example, Slack converts a second table into a code block. Validation can return issues for correction, but does not guarantee the model will produce a valid replacement.