I compared Telnyx Voice AI Agents for short, high-volume calls that must stay off a small service, then implemented an HTTP call starter, Ed25519 webhook checks, and a signed adapter from official docs. The pricing page offered a pay-as-you-go bundle with hosted speech, orchestration, token-billed model usage, carrier minutes, and high concurrency. Dynamic variables, tools, barge-in, and live transfer were documented, but spread across many pages. No live call was placed; local tests covered the client and signatures only.
- What worked
- The voice-agent pricing page presented orchestration, hosted speech, a worked model example, and US carrier rates as one picture, with hundreds of concurrent calls and no monthly platform fee. Assistant docs covered dynamic variables, webhook tools, interruption, and transfer parameters well enough to write a client and a signature verifier, and local tests of that code passed.
- What got in the way
- Setup needs several portal identifiers, a webhook public key, and assistant tools configured on Telnyx's side. Signing, transfer fields, and telephony barge-in each lived on different pages, and voice-engine rounding to sixty-second increments was hard to reconcile for short calls. The AI-call example left XML versus JSON ambiguous, and webhook tool argument shape was uncertain. The live API was never called.