sse-llm-parser

Zero-dependency parser for streaming LLM responses — the Server-Sent Events format (data: {…}\n\n … data: [DONE]) used by OpenAI- and Anthropic-style APIs. Robust to arbitrary chunk boundaries (events split mid-stream are buffered and reassembled), with helpers to pull the incremental text out of each event.

Install

npm install sse-llm-parser

Use

const { LlmSseParser, textDelta } = require('sse-llm-parser');
const parser = new LlmSseParser();

for await (const chunk of responseStream) {        // chunk: string
  for (const event of parser.push(chunk)) {
    if (event.done) break;                          // data: [DONE]
    process.stdout.write(textDelta(event));         // incremental text
  }
}
for (const event of parser.flush()) { /* trailing event, if any */ }

• parser.push(chunk) → array of completed { data, raw, done } events.
• parser.flush() → any trailing buffered event at end of stream.
• textDelta(event) → incremental text for OpenAI (choices[0].delta.content) and Anthropic (content_block_delta.delta.text) shapes; '' otherwise.

Handles multi-line data: fields, \r\n, non-JSON payloads (returned as-is) and the [DONE] sentinel.

More AI-coding & privacy guides

alexi.sh — independent guides on AI coding tools, LLMs and developer privacy.

License

MIT