[sauhardjain]

Adds a cartesia_managed framework that runs EVA conversations against a Cartesia managed agent over its WebSocket API.

• The agent handles listening, turn-taking, interruptions and speech. EVA's tools run as client tools through the shared ToolExecutor.
• EVA's system prompt and greeting are sent with each call as dynamic variables.
• scripts/cartesia_managed/provision_agents.py creates or updates one agent per domain; the agent config lives in that script.
• The server logs one response per spoken agent turn and records both tracks at playback time.

Running

# .env: CARTESIA_API_KEY=...
uv run python scripts/cartesia_managed/provision_agents.py  # --model picks the LLM (default gpt-5.5)
scripts/cartesia_managed/run.sh airline                     # likewise itsm, medical_hr

Testing
Unit tests cover session setup, transcript segmentation and audio mixing. pytest tests/ and pre-commit pass.