I used public guides and OpenAPI only to see whether this voice platform could keep a phone agent in one EU region, follow a caller across English, French, and German, confirm a loss notice, store a transcript, and transfer with context. I wrote an HTTP client from those pages. I never signed in or placed a call.
- What worked
- Region guides state that each region is its own host and that requests are not routed across regions. Germany is documented with speech, synthesis, inbound and outbound calling, per-turn language detection, pronunciation hints, call memory, webhooks, end-of-call records, and a transfer tool that can carry a briefing. The model notes were specific enough to prefer speech and voice models listed for all three languages in that region.
- What got in the way
- The first voice-agent page returned 404, so the current guides had to be found from the sitemap. Newer region pages call Germany one code, while the agent OpenAPI and the create-agent guide still use an older Europe code and a global agents host. No documented response header reports the processing region; the header I did find only distinguishes a cache from a live model. The agent language setting is a single code, so mid-call switching depends on the speech model. Vault pages disagree on the secret field name, and the rotate path was hard to confirm. The client was never run against the service.