#voice-eng

Anyone got the OAuth flow for the platform APIs working? I want to script the token.

client_credentials works. Register the app, grab client id and secret, POST to /oauth/token with grant_type=client_credentials. Token lasts about a month.

Perfect, that worked. Thanks.

Reminder that the official docs are the source of truth for customer-facing statements. If something in here contradicts them, raise it.

SFTP export connector won't connect on the acc server. Sprinklr just says there's an issue with the connection details. Host, port, user, directory and key are all correct - I've triple checked and I can sftp in myself from my laptop with the same key.

Docs just say verify the connection details and check the firewall. Firewall is open, I can see the connection arrive.

This is not a credentials problem. Check /var/log/auth.log - I bet you see 'Disconnected ... [preauth]' mid-authentication.

Yes! Exactly that. What is it?

Sprinklr's SFTP connector uses an old SSH client that signs RSA keys with the legacy ssh-rsa signature, which is SHA-1. OpenSSH 8.8 and later disabled SHA-1 RSA signatures by default and only accept rsa-sha2-256 or rsa-sha2-512. The server rejects the signature type and drops the connection. The UI surfaces that as a generic connection error, which is why it looks like bad credentials.

Quick fix for test, scoped to the sftp group only: Match Group sftpusers PubkeyAcceptedAlgorithms +ssh-rsa then systemctl reload sshd. The rest of the box keeps modern defaults.

For production do it the other way round - use an ed25519 or ECDSA key for the destination and avoid SHA-1 entirely. Only re-enable ssh-rsa server-side, narrowly scoped, if you have no choice.

Can we get this into the docs? It cost Jonas two days.

Raised it. Not holding my breath.

Is there a GET callback-status API? I need callback completion out to our system and I can't find an endpoint.

There isn't one. Don't keep looking. It's a push, not a pull.

Proven on prod19 today. Campaign Post Call Workflow journey -> External REST API connector -> your authenticated webhook. Correlation works: Schedule Callback returns a taskId and it comes back on the call as VOICE_CONVERSATION.SCHEDULED_CALLBACK_TASK_ID, so the push can carry it and the receiver matches it up.

Gotchas, each of which cost me real time: - Register an External REST API first (All Settings -> APIs and Integrations). The journey API node has no inline URL, only a connector dropdown. - Map values with the resource PICKER, not raw VOICE_CONVERSATION.x in custom code. In Post Call Workflow context only PROFILE is populated, so raw Groovy returns null. - The OpenAPI spec on a connector is read-only after creation. To rename a body field you recreate the connector. - Callbacks only exist on a Callback dialer profile. Preview dials have no callback task, so callbackTaskId comes back empty.

This should be a docs page on its own.

Heads up - client_credentials is not enabled any more, you get 'Default Admin User Not Setup'. Use the auth-code flow instead: browser, logged in, hit /oauth/authorize with client_id and redirect_uri, submit, pick the client, submit, then exchange the code at /oauth/token. Headers on calls are Authorization: Bearer <token> plus Key: <client_id>.

Ah - that's why my script broke last week. I assumed I'd broken it myself and rewrote the whole thing.

Dialer isn't calling on the Wechsler UAT tenant. Segment says Data Exhausted but Number of Callable Records is non-zero and Attempted Records is 0.

Best guess: Data Exhausted is sticky - updating the leads doesn't reopen a segment that already finished its pass. Probably need a fresh segment on the same filter rather than another upload.

Flagging that as a guess, not a finding. I haven't proven it.

Rechurn is not the lever either - its job is overriding attempt limits for leads that HAVE been attempted. With Attempted Records 0 there's no limit to override.

SEAT Cupra VoiceConnect trunk: outbound is fine, inbound never arrives. No error anywhere. Trunk looks half alive.

What's in the IP Access Control List?

The AudioCodes hostname - qqjjgaaaqdcr.sip1-region1.audiocodes.io. It resolves to the right IP, I checked.

That's it. The VC ACL does not resolve host names, it compares the literal value. A resolving implementation would have matched, so we know it doesn't. Put 20.113.51.15/32 in there.

Inbound working. That would have taken me another day.

Whitelist the media address too - 20.113.4.49. Signalling and media are different hosts and people miss the second one.

Update on the Data Exhausted thing: it was not the segment. The segment went Active on its own and still scheduled zero calls, and brand new campaigns in the same tenant failed identically. It's an environment-level scheduler problem, escalated to Sprinklr support. So please don't go rebuilding segments on my earlier guess - treat Data Exhausted as a thing to clear, not as an explanation.

New here - starting on the DT integration. Where do I even find how any of this works?