>_
SSE

LLM Stream (SSE) Parser

Paste a raw streaming response (Server-Sent Events or NDJSON) from OpenAI, Anthropic Claude, Google Gemini, Ollama or any OpenAI-compatible API, and get the final text, reasoning, tool calls and token usage.

Runs locally — your input never leaves this page.

Samples
Final text 
  
 
 Tool calls 
 Usage 
 
 
 
  
 
 
 
How to use
 
     
  1. Copy the raw stream: from browser DevTools (Network → EventStream / Response), from curl -N, or from your server logs.
  2. Paste it in. The format is detected automatically: OpenAI Chat Completions, the OpenAI Responses API, Anthropic Messages, Gemini or Ollama. OpenAI-compatible APIs (DeepSeek, Qwen, Groq, vLLM…) work too.
  3. Read the reassembled answer, the reasoning or thinking text, each tool call with its complete JSON arguments, the stop reason and token usage.
    4.  
  
FAQ
 
 
  Why debug streams by hand?  
When a streamed answer is cut off, a tool call’s arguments fail to parse, or token usage looks wrong, the raw chunks are the source of truth. Reassembling them shows exactly what the API sent.
 
  My tool-call arguments are invalid JSON. Why?  
Arguments arrive as string fragments spread across many chunks. If any chunk is missing — for example, because a proxy buffered or dropped data — the concatenated JSON breaks. The parser highlights arguments that don’t parse.
 
 
 
  Report a bug or suggest a feature →