Streaming Enhancements

Stream chunk types, markdown healing, and table buffering utilities.

Stream Chunks

Stream chunks represent pieces of a streaming response.

import {
  createStreamChunk, parseStreamChunks, serializeStreamChunks,
  isTextChunk, isToolCallChunk, isToolResultChunk,
  healMarkdown, bufferTableCells,
} from "@fluxy-chat/sdk";

Creating Chunks

const text = createStreamChunk("text", { content: "Hello " });
const toolCall = createStreamChunk("tool-call", { toolName: "get_weather", toolArgs: '{"city":"London"}' });
const toolResult = createStreamChunk("tool-result", { toolResult: '{"temp":20}' });
const done = createStreamChunk("done");

Parsing & Serializing

Uses Vercel AI SDK wire format prefixes (0: for text, 8: for tool calls, 9: for tool results, 3: for errors):

const raw = '0:Hello\n8:{"toolName":"get_weather","args":{"city":"London"}}\n9:{"temp":20}';
const chunks = parseStreamChunks(raw);
// -> [{ type: "text", content: "Hello" }, { type: "tool-call", toolName: "get_weather" }, { type: "tool-result", ... }]

const serialized = serializeStreamChunks(chunks);
// -> same as raw

Type Guards

if (isTextChunk(chunk)) console.log(chunk.content);
if (isToolCallChunk(chunk)) console.log(chunk.toolName);
if (isToolResultChunk(chunk)) console.log(chunk.toolResult);

Markdown Healing

Closes unclosed code fences that may result from streaming interruption:

healMarkdown("```js\nconsole.log('hi')"); // -> "```js\nconsole.log('hi')\n```"
healMarkdown("hello **world**");           // unchanged (valid)

Table Buffering

Adds a separator row to pipe tables:

bufferTableCells("| Name | Age |\n| Alice | 30 |");
// -> "| Name | Age |\n| --- | --- |\n| Alice | 30 |"

Types

StreamChunkType: "text" | "tool-call" | "tool-result" | "error" | "done"
StreamChunk: \{ type, content?, toolName?, toolArgs?, toolResult?, error? \}

Streaming Enhancements

On this page