ai-core/ag-ui-protocol

To stream AI chat responses as Server-Sent Events, you wrap TanStack AI chat iterables in an SSE helper. This converts backend streams into AG-UI compliant typed events for web clients to consume directly over HTTP.

SSE (Server-Sent Events) and NDJSON (newline-delimited JSON) are both supported formats for streaming AI chat events. SSE maintains a persistent HTTP connection for real-time updates, while NDJSON sends structured JSON lines for simpler event parsing.

Handle tool call failures and aborts during HTTP event streaming by utilizing built-in abort handling and RUN_ERROR reporting. These features ensure streams remain stable and properly report errors when tool failures occur.

Expose typed state snapshots and text deltas by converting backend iterables into a typed StreamChunk event catalog. This documents full event suites including tool call arguments, steps, and custom messages for precise frontend reactions.

Yes, you can use TanStack AI chat streams behind reverse proxies. The streaming mechanism includes header customization and proxy-friendly diagnostics to keep SSE and HTTP streams stable in restricted network environments.

AG-UI adapters fail to render tool calls from custom streams because they expect compliant event formats. Converting your backend streams to the AG-UI typed StreamChunk catalog ensures proper synchronization of tool calls and state updates.