chutes-call

Shared Docker FastAPI service (port 8630) that centralizes ALL Chutes.ai LLM calls. Enforces the account-wide 5-concurrent-connection limit via a global asyncio semaphore, handles retries with tenacity, implements a circuit breaker (CLOSED/OPEN/HALF-OPEN), and returns structured request/response objects with cleaned JSON.

Why

• Chutes has a hard 5-concurrent-connection limit per account.
• Without centralized coordination, multiple agents cause ~50% failure rate.
• scillm's retry logic has shared mutable state bugs and 30s timeouts.
• 31 skills make bespoke Chutes calls — one gateway replaces all of them.

Usage