This is not legal advice, and I am not qualified to give it. What follows is what the product does, what the product cannot decide for you, and the specific questions worth putting to somebody who is qualified. If you take one thing from it, take the list of questions.
Two different obligations, routinely confused
Recording. Whether you may capture and keep the audio of a call. In the United States this is state law, several states require every party to consent, and the rule that applies is generally the stricter of the two ends of the call — which for an inbound line you do not control means assuming the strictest.
Disclosure that the caller is talking to a machine. A separate and newer question, moving faster than the recording rules, and increasingly answered by statute in specific states and sectors rather than by general principle.
These get merged in most vendor copy, and merging them produces the wrong conclusion in a common case — which is ours.
Storing nothing does not settle it
Osokoro Phone keeps no recordings, no transcripts, and no caller numbers. What survives a call is a content-free journal row: status, duration, turn count, search count, pricing version, cost estimate. Deleted after thirty days. Abandoned calls are marked failed after twenty minutes.
It is tempting to conclude that recording law therefore does not apply. Do not conclude that from a blog post.
The reasonable-sounding argument is that consent rules attach to capturing and retaining audio, and we do neither. The reason not to rely on it: audio is transmitted to and processed by a third-party model provider in order to be understood at all, and whether that constitutes interception under a given state's statute is a legal question about that statute, not an engineering question about our retention policy. I do not know the answer for your state. Neither does any vendor telling you confidently that you are fine.
What storing nothing genuinely does buy you is narrower and still worth having: there is no archive to breach, no archive to disclose, and no archive to produce.
What the product gives you
A recording disclosure toggle, configured per business, and the agent speaks whatever you write in it. There is also first words — a greeting spoken before the agent thinks, which exists so callers do not hear a pause while a model reasons. That greeting is the natural place for a disclosure, because it is the only text guaranteed to be said before anything else.
Both fields are yours. We do not write them, do not review them, and do not check them against any jurisdiction. That is the honest boundary: the toggle is a mechanism, not compliance.
The questions to actually ask
Take these to a lawyer in your state. They are cheap to answer and expensive to guess at.
- Does our state's consent statute reach audio that is transmitted to and processed by a third party but never stored? This is the central one.
- Whose law applies on an inbound call from out of state, and should we simply assume the strictest?
- Is there a duty to disclose that the caller is speaking to an automated system, distinct from any recording duty, in our state or our sector?
- Does our sector add anything on top? Healthcare, finance, debt collection and legal services all do.
- Does the disclosure have to be at the start, and does the caller have to be given a way to reach a person instead? This one has a product consequence: our agent transfers to a single configured number, so if the answer is yes, that number needs to be staffed during the hours the line is on.
The one I can answer
Do not use this for regulated data. Business knowledge lives in collections hosted by xAI, whose model runs the call, and zero data retention is arranged per xAI team rather than per API key. If you are in healthcare, finance, or anywhere with a statutory duty over what callers tell you, the answer today is no, and no configuration changes that.
I would rather write that here than have you find it in an appendix.
Writing the greeting
Practical notes, not legal ones. Whatever your lawyer says has to go in it, these constraints apply:
- The greeting is spoken before the agent reasons. It is the one line whose timing you can rely on. Put anything mandatory there rather than in the instructions.
- Keep it short. Callers hang up during long preambles, and every second of it is billed at the same per-minute rate as the useful part of the call — with a one-minute minimum, a twenty-second disclosure on a thirty-second call is most of what you paid for.
- Say the machine part plainly. "You're speaking to an automated assistant" is clearer than anything softer, and softer phrasings are what get argued about later.
- Say what happens next. If a person is available, say how to reach them. It reduces the number of callers who spend two minutes discovering it.
Status
Osokoro Phone is a private beta, approved by a person, one business at a time, inbound only. Numbers are provisioned by an operator rather than self-serve. The disclosure toggle and the greeting field exist and work; deciding what belongs in them is yours, and worth an hour of a lawyer's time before the line goes live rather than after.