DispatchIQ Backend — v3 (Bug Fixes)

What was fixed

Fix 1 — AttributeError: 'AgentSession' object has no attribute 'room' In LiveKit Agents v1.5+, context.session is an AgentSession. It does not have a .room attribute. The room is only available in entrypoint via ctx.room.name.

Solution: room_name is now passed as a constructor argument to every agent class and stored as self.room_name. No more reliance on context.session.room.

Fix 2 — 500 errors on GET /sessions/{id} The livekit.api.LiveKitAPI constructor changed in newer versions. Replaced with the correct AccessToken + VideoGrants pattern that works with livekit>=0.17. Also fixed the update endpoint to not 404 when the agent pushes before the frontend polls — it now logs a warning and returns gracefully.

Fix 3 — model "qwen3:32b" not found Ollama cloud model names differ from local names. Default changed to llama3.1 which is stable and well-supported on Ollama cloud.

Fix 4 — silero deprecation warning livekit-plugins-silero is deprecated in v1.5+. VAD is now bundled inside AgentSession automatically. Removed the explicit vad=silero.VAD.load() arg and the silero import entirely.

---

Files

main.py        FastAPI — session management + LiveKit token generation (fixed)
agent.py       Voice agent — three agents, room_name fix applied (fixed)
tools.py       Dispatch tools — unchanged
escalation.py  Escalation logic — unchanged
requirements.txt
.env.development
.env.production

---

Setup

python -m venv venv
source venv/bin/activate    # Mac/Linux
# venv\Scripts\activate     # Windows

pip install -r requirements.txt

cp .env.development .env
# Fill in LIVEKIT_URL, LIVEKIT_API_KEY, LIVEKIT_API_SECRET,
#         DEEPGRAM_API_KEY, OLLAMA_API_KEY

Run

# Terminal 1
uvicorn main:app --reload --port 8000

# Terminal 2
python agent.py dev

Ollama Cloud models that work

llama3.1              ← default, recommended
llama3.2
mistral

Do NOT use qwen3:32b — that format is for local Ollama only. On Ollama cloud check available models at ollama.com/library.

Deploy to Railway

Service 1 — FastAPI:

uvicorn main:app --host 0.0.0.0 --port $PORT

Service 2 — Agent worker:

python agent.py start

Add all vars from .env.production to both services. Set BACKEND_URL in Service 2 to the Railway URL of Service 1.