Vivarium Proxemics

Quick Start

1. Install Docker Desktop (with Compose) and make.
2. I start in stacks/r0-dev-exploration.
3. I open that stack README.md.
4. Copy env templates and add required API keys.
5. I run make start.
6. I open the local app URL from that stack README (usually http://localhost:3000).
7. I stop with make stop.

Stacks

• stacks/r0-dev-exploration: baseline LiveKit + agent-runner + web-client.
• stacks/r1-eval-s2s-openai: OpenAI realtime speech-to-speech eval.
• stacks/r2-eval-s2s-gemini: Gemini Live speech-to-speech eval.
• stacks/r3-livekit-meet-lab: LiveKit Meet + desk tooling.
• stacks/r4-zoom-join-leave-lab: Zoom join/leave automation lab.
• stacks/r5-deployment-webrtc-test: SmallWebRTC deployment test stack.
• notebooks: experiments and analysis.

Daily Workflow

1. I enter the stack: cd stacks/<stack-name>.
2. I start it: make start.
3. I inspect logs: make logs or make logs SERVICE=<service>.
4. Edit code (server/ or agent-runner/ for backend, client/, web-client/, or meet/ for frontend).
5. I stop it: make stop.

Notes

• I keep env files stack-local.
• If env or Dockerfiles change, I restart: make stop && make start.
• I treat each stack README.md as source of truth for ports and required keys.

Known Issue (r3)

• In stacks/r3-livekit-meet-lab, LiveKit + Pipecat sessions can still hit memory growth/OOM in some camera-on flows.
• Symptom: bot drops after a few minutes, then bot start/proxy calls fail (fetch failed / 502) because agent-runner exited.
• I keep video ingest disabled by default and ignore video subscriptions unless explicitly enabled.