I see this when using some OpenAI compatible models.
Likely root cause:
Interactive chat sessions were intermittently aborting when the UI attempted to
stream assistant messages. The React SDK delegated streaming to
@tambo-ai/typescript-sdk’s advanceStream, which tries to JSON.parse each
newline-delimited SSE chunk as it arrives. Under normal network conditions the
server can split a single JSON payload across multiple TCP frames, so the client
occasionally received incomplete JSON and threw Failed to parse JSON after multiple chunks.—terminating the stream and surfacing an error toast.