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.
| File | Kind | Size |
|---|---|---|
| .npmignore | other | 66 B |
| LICENSE | other | 1.1 KB |
| README.md | markdown | 1.3 KB |
| SKILL.md | skill md | 4.5 KB |
| bin/install.jsexecutable | script | 5.3 KB |
| install.shexecutable | script | 1.1 KB |
| package.json | manifest | 906 B |
| publish.shexecutable | script | 604 B |
| scripts/webhook_server.py | script | 10.3 KB |
| test.shexecutable | script | 488 B |
| tests/test_webhook.py | script | 8.6 KB |
SKILL.md
View on GitHub---
name: didww-to-openai-realtime
description: >
Point 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
1. **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).
2. **OpenAI side:** in platform settings, create a project webhook for
`realtime.call.incoming` pointing at the backend URL; note the
webhook secret. The project ID (`proj_...`) becomes the SIP address.
3. **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 uses
`sip-eu.api.openai.com`). Attach the DID. Set `capacity_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.
4. **Deploy the backend:** `scripts/webhook_server.py` is a stdlib-only
starting point: verifies the webhook signature (it refuses to start
without `OPENAI_WEBHOOK_SECRET` and `OPENAI_API_KEY`), drops duplicate
deliveries, matches the To number exactly against `AGENTS`, accepts
with the model and instructions, and for a DID with `transfer_to`
gives the model a `transfer_call` tool, 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}/refer` with
`{"target_uri": "tel:+..."}`). Adapt, deploy behind HTTPS, set both
variables in its environment.
5. **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](https://www.didww.com/?ref=didww-to-openai) 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: 472k | 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 |