Provider optionality is the moat
Being able to mix and match LLM, TTS and transcription providers without re-architecting is the right developer experience. We swapped LLM vendors mid project with minimal code change.

Voice agent platform for developers with broad LLM and TTS provider choice.

Vapi is the closest competitor to Retell in the developer voice infrastructure tier. Its differentiator is provider optionality: bring your own LLM, swap TTS engines, mix-and-match transcription providers. The trade-off is configuration surface — reviewers report that getting Vapi tuned to a specific use case takes more iteration than competitors with more opinionated defaults.
The strongest reviewer signal is from engineering teams that have explicit model or vendor preferences (a regulated workload that requires a specific LLM, an existing TTS contract, a multilingual flow). Teams shopping for the fastest path to a working agent generally land on a higher-level platform. Compliance and SOC 2 posture have improved through 2025 and now meet expectations for most B2B buyers.
9 of 41 verified submissions shown below.
Being able to mix and match LLM, TTS and transcription providers without re-architecting is the right developer experience. We swapped LLM vendors mid project with minimal code change.
Vapi will let you do almost anything. The cost is that you have to make alot of decisions before you have a working agent. Once we settled on our defaults iteration was fast.
The Discord is genuinely helpful and the team ships fixes quickly. We solved a barge in edge case in 48 hours that would of taken weeks elsewhere.
Your draft agent doesn’t hold the prompt. You make one setting, but then when you switch to another setting, the first setting goes off. It is impossible to pick right Voices for specific language.
Our use case requires three TTS engines depending on language. Vapi was the only platform that supported this cleanly out of the box. Compliance posture is meaningfully better then it was a year ago.
We underestimated the tuning required to get a polished agent. Once tuned its solid. If you want opinionated defaults, look at a higher level platform.
Other voice AI platforms reviewed for engineering workloads.