aimarket-streaming

Documentation

SSE/WS streaming + per-chunk billing with micro-receipts. Capability streams tokens. Every N tokens → signed micro-receipt. Consumer can cancel mid-stream — pays only for what was received. Channel balance debited incrementally.

When to Use

• LLM text generation (tokens streamed → billed per 10 tokens)
• Long-running AI tasks where consumer wants partial results
• Budget-sensitive consumers who want to cancel mid-stream
• Audit: every chunk has a signed receipt for compliance

Installation

pip install aimarket-streaming

Example

from aimarket_hub.signing import Signer
from aimarket_streaming.streaming import StreamingBiller

signer = Signer()
biller = StreamingBiller(signer)

session = biller.open_session(
    capability_id="text.generate@v1", product_id="prod-gpt",
    channel_id="ch_abc123", price_per_token_usd=0.001,
    tokens_per_chunk=10
)

async def token_stream():
    for word in ["The", "quick", "brown", "fox", "jumps", "over", "the", "lazy", "dog", ".", "..."]:
        yield word  # In production: yield actual LLM tokens

import asyncio
async def main():
    async for chunk in biller.stream_tokens(session, token_stream()):
        print(f"Chunk {chunk['chunk_receipt']['chunk_index']}: "
              f"${chunk['cumulative_price_usd']:.6f} "
              f"({chunk['cumulative_tokens']} tokens)")

asyncio.run(main())

# Cancel mid-stream — pays only for received
result = biller.cancel_session(session.session_id)
print(f"Cancelled. Total paid: ${result['total_price_usd']:.6f}")

Per-Chunk Billing

• price_per_token_usd: 0.001 (0.1 cent per token)
• tokens_per_chunk: 10 (micro-receipt every 10 tokens)
• Each chunk receipt is Ed25519-signed — verifiable audit trail

License

MIT · Maintained by AI-Factory