didww/ai / skills/didww-to-openai-realtime
didww-to-openai-realtime
VerifiedPoint a DIDWW phone number directly at OpenAI's Realtime API over SIP, with a generated webhook backend for call control. Use whenever the user wants gpt-realtime answering a real phone number via DIDWW, mentions sip.api.openai.com, realtime.call.incoming webhooks, Realtime SIP, or debugging calls between DIDWW and OpenAI. This is the developer-grade pairing: the agent behavior is code the user owns.
DIDWW to OpenAI Realtime
The most direct wiring in the family: the DIDWW trunk points straight at
OpenAI, no platform in between. The tradeoff is that there is no agent
object to configure; every incoming call fires a webhook and YOUR code
accepts it with instructions. This skill therefore generates that
backend (scripts/webhook_server.py) rather than pretending
configuration exists. Not a no-code path; say so in intake and offer the
hosted platforms if the user has no way to run a small public HTTPS
endpoint.
The flow
- Confirm in one message: number (or country to buy), what the agent should do (becomes the instructions), transfer target, where the webhook backend will run (any host that can serve public HTTPS: a five-dollar VPS, a serverless function, a tunnel for testing).
- OpenAI side: in platform settings, create a project webhook for
realtime.call.incomingpointing at the backend URL; note the webhook secret. The project ID (proj_...) becomes the SIP address. - DIDWW side (MCP, OAuth): inbound trunk with destination
sip:{project_id}@sip.api.openai.com;transport=tls, port 5061, TLS signaling and SRTP media (both REQUIRED; EU data residency usessip-eu.api.openai.com). Attach the DID. Setcapacity_limit= expected concurrent calls (it caps how many calls run at once, and so the cost). No DIDWW outbound trunk is involved, so there is no 24-hour limit to set. Note: no digest auth exists on OpenAI's ingress, and anyone who knows the project ID can send an INVITE with any To number. The server's DID match is a routing filter, not authentication: keep the project ID private (treat the SIP URI like a credential) and rely on the signed webhook plus the reject-by-default for unknown numbers. - Deploy the backend:
scripts/webhook_server.pyis a stdlib-only starting point: verifies the webhook signature (it refuses to start withoutOPENAI_WEBHOOK_SECRETandOPENAI_API_KEY), drops duplicate deliveries, matches the To number exactly againstAGENTS, accepts with the model and instructions, and for a DID withtransfer_togives the model atransfer_calltool, watches the call websocket (wss://api.openai.com/v1/realtime?call_id=...) and on the tool call refers the call (POST /v1/realtime/calls/{call_id}/referwith{"target_uri": "tel:+..."}). Adapt, deploy behind HTTPS, set both variables in its environment. - Verify: inbound call from a mobile; watch the server log accept the call; test the transfer phrase. Outbound calls are not part of OpenAI's SIP interface as of August 2026; originate outbound from the DIDWW side or a platform if needed.
Failure map
- Call rings once and dies: TLS or SRTP not set on the DIDWW trunk (both are mandatory), or the project ID in the SIP address is wrong.
- Webhook never fires: webhook not configured for the project, or the backend URL is not publicly reachable HTTPS.
- Call connects then silence: the accept API call failed; check the server log, the API key, and that the accept happens within seconds.
- Anyone can ring the URI: expected; keep the unknown-DID rejection in the server, it is the security model.
Panel map and money
OpenAI keys: platform.openai.com/api-keys; webhooks (required here):
platform.openai.com/settings/project/webhooks, copy the signing
secret at creation. Billing: prepaid credits under Settings, Billing
(minimum 5 USD); auto-recharge defaults to ON during setup and credits
EXPIRE after one year, non-refundable: say both out loud. Realtime
bills in audio tokens, not minutes; a typical phone conversation lands
around 0.05 to 0.10 USD/min of model cost depending on talk ratio.
DIDWW side: prepaid; top-up, Auto-charge and Low Balance Notification
(OFF by default) in Billing, Payment Methods.
Built around DIDWW SIP
trunking and OpenAI's Realtime SIP guide; verified August 2026. xAI's
Grok voice offers a near-identical interface (sip.voice.x.ai); the
didww-to-xai skill covers it.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |