docker-compose.yml

# project mocha — no containers.
#
# Mocha used to run her own vLLM (Hermes-4-70B) and a gesture service here.
# After co-locating inside "opus trading", both were removed:
#
#   • LLM       — Mocha now SHARES opus trading's already-running vLLM
#                 (Qwen/Qwen3-32B-FP8 on http://127.0.0.1:8000/v1). Putting a
#                 second LLM on the GPU is unnecessary and risks contending with
#                 Corvus's trading inference. See config.yaml `llm.base_url` and
#                 the rate-limit/circuit-breaker in bridge/llm_client.py.
#   • gesture   — the build context (gesture/Dockerfile) never existed; the
#                 service was already broken. Body-language is driven entirely by
#                 the web app's clip-based animation controller (no GPU service).
#
# Mocha's own services (STT :8091, TTS :8092, bridge :8090, web :8080) run as
# plain processes via ./start.sh — no Docker needed.
services: {}