Server
Compose helper inside the agency-server repo (also reachable as ./dev up from the umbrella Workspace).
Container name resolution
Friendly aliases map to agency_* container names:
Alias |
Container |
|---|---|
django, wsgi |
|
asgi |
|
work |
|
beat |
|
flower |
|
nginx, web |
|
postgres, db |
|
redis, queue |
|
chroma, store |
|
ollama, llm |
|
comfy, comfyui |
|
voice |
|
Commands
cd agency-server
./dev up # start stack (foreground)
./dev down # stop
./dev build # build images and start
./dev enter [svc] # bash in container (default django)
./dev logs [svc] # follow logs
./dev log [svc] [name]# tail file under /usr/src/logs/
./dev restart [svc]
Documentation
./run document # build Sphinx → static/documentation/
Product URLs (default)
SPA: http://localhost/ (after http://localhost/login/)
Admin: http://localhost/admin/
Sphinx HTML: http://localhost/static/documentation/
Voice (STT/TTS)
Service voice (agency_voice) exposes faster-whisper + Piper.
Nginx strips the prefix: /api/voice/stt → container POST /stt.
cd agency-server
docker compose build voice
docker compose up -d voice
docker compose up -d --force-recreate web # pick up /api/voice/
Env: WHISPER_MODEL, WHISPER_DEVICE (cuda/cpu), PIPER_VOICE.
Models cache in volume agency_voice. Details: context/voice/README.md.
Agent chat (AgencyAgentTest) uses mic → STT and optional auto-speak TTS
when /api/voice/health is OK.
ComfyUI (image gen)
Service comfy (agency_comfy) is local GPU txt2img. Django uses
COMFYUI_URL (http://comfy:8188). Host UI: http://127.0.0.1:8188
cd agency-server
docker compose build comfy
docker compose up -d comfy
First boot downloads SD 1.5 fp16 (~2 GB) into volume agency_comfy.
Default --lowvram so it can share the GPU with Ollama. Details:
context/comfyui/README.md.
Compose file and env templates live in agency-server root. Package source is mounted from sibling agency-django.
Stack topology and services: Deployment.