scoutwyze-compute

Client + LangChain/Haystack Tool wrappers for ScoutWyze Compute — a machine-readable GPU placement recommendation API for AI infrastructure agents and MLOps pipelines. Ranks live 8x H100 80GB (RunPod-backed) offers by price and data freshness; strictly separates what a provider reported from what ScoutWyze calculated. This is a routing/recommendation layer, not an execution platform — it never reserves, provisions, or runs anything.

Dual-rail, both real:

• Bearer — a prepaid API key, billed per successful match (a no_match result is never charged). Get one free with POST /v1/signup, fund it with POST /v1/checkout-sessions (real Stripe Checkout).
• x402 — pay per call in USDC on Base, no signup, no API key, no ETH for gas. Uses the real x402 "exact" EVM scheme (EIP-3009 TransferWithAuthorization): you only ever sign a message, this server's own facilitator broadcasts the settlement.

Install

pip install scoutwyze-compute                 # Bearer rail only, zero extra deps
pip install "scoutwyze-compute[x402]"          # + x402/USDC-on-Base rail
pip install "scoutwyze-compute[langchain]"     # + LangChain Tool wrapper
pip install "scoutwyze-compute[haystack]"      # + Haystack Tool wrapper
pip install "scoutwyze-compute[all]"           # everything

Usage

Plain client

from scoutwyze_compute import rank_via_bearer, rank_via_x402

result = rank_via_bearer("https://scoutwyze-compute.fly.dev", "sw_live_...", {"gpuClass": "H100"})
print(result["body"]["recommended"])

LangChain

from scoutwyze_compute.langchain_tool import make_bearer_tool

tool = make_bearer_tool(api_key="sw_live_...")

Haystack

from scoutwyze_compute.haystack_tool import make_bearer_tool

tool = make_bearer_tool(api_key="sw_live_...")
offers = tool.invoke(gpuClass="H100")

See the main repo for full API docs, llms.txt, and openapi.json.

License

MIT