ShapleyCut API

Dedicated service for the ShapleyCut compression engine. This package exposes HTTP endpoints for optimizing memory and running the gateway mode.

Core engine: https://pypi.org/project/shapleycut-core/

Install

pip install shapleycut-api

Quickstart

python -m uvicorn shapleycut_api.main:app --host 0.0.0.0 --port 8000

If you are on Python 3.14 and see httptools errors, use:

python -m uvicorn shapleycut_api.main:app --host 0.0.0.0 --port 8000 --http h11

SDK snippet (Python)

from shapleycut import MemoryMessage, optimize_messages

messages = [
	MemoryMessage(role="system", content="You are a helpful assistant."),
	MemoryMessage(role="user", content="Summarize the project status."),
	MemoryMessage(role="assistant", content="We shipped MVP and added gateway mode."),
]

out = optimize_messages(messages, max_output_tokens=512)
print(out.optimized_memory)

Endpoints

• GET /health
• POST /v1/optimize
• POST /v1/chat/completions
• POST /v1/responses
• GET /v1/metrics
• GET /v1/metrics/export

Environment

LLM credentials (server side only):

• GITHUB_TOKEN or GH_TOKEN
• GITHUB_INFERENCE_BASE_URL (default: https://models.github.ai/inference)
• OPENAI_API_KEY
• OPENAI_BASE_URL (default: https://api.openai.com/v1)

Gateway settings:

• SHAPLEYCUT_GATEWAY_PROVIDER (github_azure, openai, anthropic)
• SHAPLEYCUT_GATEWAY_BASE_URL
• SHAPLEYCUT_GATEWAY_API_KEY_ENV
• SHAPLEYCUT_GATEWAY_API_KEY
• SHAPLEYCUT_GATEWAY_REQUIRE_KEY
• SHAPLEYCUT_GATEWAY_ALLOW_OVERRIDE
• SHAPLEYCUT_GATEWAY_LOG
• SHAPLEYCUT_GATEWAY_DB
• SHAPLEYCUT_GATEWAY_METRICS_MAX
• SHAPLEYCUT_GATEWAY_CACHE_MAX
• SHAPLEYCUT_GATEWAY_CACHE_TTL_S

Notes

• /v1/responses is not supported when SHAPLEYCUT_GATEWAY_PROVIDER=github_azure. Use /v1/chat/completions or set the provider to openai.