Connect my existing real-time AI agent to Chert FaceTime using Bring Your Own Room (BYOR). Build and deploy the integration, including the webhook receiver, and return the exact public HTTPS webhook URL to enter in Chert.

Complete the implementation and deployment. Do not stop at a plan, sample code, or instructions for me to build the receiver.

1. Inspect and preserve my agent.

   Inspect this workspace and its instructions. Identify the conversation engine, audio input/output, optional avatar/video provider, session lifecycle, and deployment setup. Preserve existing prompts, workflows, tools, state transitions, timing, and completion behavior. Add FaceTime as another way to use this agent.

   Read the customer receiver reference at https://github.com/cherthq/chert-reference-agent, especially docs/CONTRACT.md, docs/OPERATIONS.md, source, and tests. It publishes test media; replace that fixture with my actual agent. Use the documented customer interface and current Chert onboarding configuration. Identify missing contract details rather than guessing. Do not require access to Chert's private source.

2. Connect the media.

   Reuse an existing LiveKit integration when available. Otherwise, connect the agent's supported media interfaces: route incoming caller audio to the agent and publish its generated speech into the room. If it has an avatar, publish its actual generated video from the same participant. Keep one agent audio publisher and preserve supported interruption and turn-taking behavior.

   Verify that the provider exposes a usable media path; a browser preview alone is insufficient. Do not assume caller video is available. If the provider cannot support this connection, identify the exact blocker without silently replacing my agent, provider, or avatar with a demo.

3. Build the webhook receiver.

   Implement a public HTTPS POST endpoint for Chert call events. On a valid call.incoming, prepare an isolated session and LiveKit room, ensure the agent is ready, and respond within the configured Chert decision deadline with action="accept", livekit_url, participant_token, and remote_participant_identity as documented in docs/CONTRACT.md.

   Use the agent's identity for remote_participant_identity. Supply Chert a fresh, room-scoped token with the required permissions and a distinct identity. Never share conversation state between callers. If startup is too slow, use bounded preparation ahead of calls and exclusive session assignment. Decline when preparation fails; never accept an unready agent or extend the deadline.

   Handle duplicate/concurrent events, call.started, call.ended, call.failed, connection timeout, expiry, and restart-safe cleanup. Release agent/provider sessions and rooms on termination. Use Chert's supported call lifecycle.

   For a new receiver, follow the reference default: CHERT_SIGNING_ENABLED=false with explicit webhook routes. If signing is requested or already enabled, preserve it and verify the documented HMAC signature over exact request bytes using the configured numeric key version. Never fall back to unsigned mode. Keep secrets server-side and out of chat, source control, and logs. Add health/readiness endpoints. Do not add recordings or log caller media, tokens, or webhook request/response bodies.

4. Deploy and verify.

   Use my existing configured hosting if it supports the runtime requirements. The reference service requires one always-on process with writable persistent cleanup storage. Follow docs/OPERATIONS.md; do not put its long-lived media process into a request-scoped serverless function.

   Use existing authorized deployment tools and securely configured credentials. Ask only for genuinely missing access or configuration, a hosting choice if none is available, or approval for new spending. Never ask me to paste secrets into chat.

   Test validation, the configured signing mode, timely acceptance, room/token/ identity binding, duplicate delivery, session isolation, and cleanup. Deploy and verify the actual public HTTPS POST route, health, and readiness. Run authorized synthetic lifecycle and LiveKit checks for outgoing audio, optional avatar video, incoming audio handling, and cleanup. Report what actually passed.

5. Return the onboarding result.

   Put the exact deployed webhook URL first, including its full route path. Do not substitute a frontend URL, LiveKit URL, localhost URL, placeholder, or deployment dashboard link. Then provide deployment/test status, the matching Chert signing setting, and any remaining setup or blockers.

   For unsigned mode, tell me to leave “Enable webhook signing” unchecked. For signed mode, securely install Chert's generated secret/version on the receiver. Changing the dashboard checkbox does not configure the server. Verify the deployed receiver's mode before reporting it.

   Tell me to use customer_decides, leave the Chert Assistant unassigned, and assign the integration after receiver setup is complete. Do not change line assignments or place a FaceTime call. Distinguish synthetic/LiveKit verification from a supervised real FaceTime test.

   If deployment is blocked, finish everything possible locally and identify the exact missing prerequisite. Never invent a webhook URL or claim an unverified integration is ready.