The SaaS lock-in in voice AI just broke — and it's being led by people who've already cashed out once.
The Summary
- Dograh is an open-source voice agent platform that runs on your own infrastructure, positioning itself as the self-hostable alternative to closed SaaS platforms like Vapi and Retell
- Built by YC alumni and founders with exits, it's fully open source under BSD 2-Clause license with drag-and-drop workflow builder
- Core value prop: zero vendor lock-in, full data residency control, and the ability to swap any LLM, STT, or TTS provider without platform restrictions
The Signal
Voice agents are having their WordPress moment. Dograh arrived on GitHub trending this week as the first credible open-source challenge to the SaaS duopoly controlling production voice AI. The platform runs entirely on your infrastructure with a single Docker command, putting voice agent infrastructure in the same category as databases and web servers: commoditized, self-hostable, yours.
The timing matters. Vapi and Retell have been charging per-minute SaaS fees while locking teams into their proprietary stacks. That worked when voice AI was experimental and companies would pay for speed. But as voice agents move from demos to production workloads handling customer service, sales calls, and internal operations, the cost structure breaks. A voice agent running 8 hours a day, 5 days a week racks up 10,400 minutes monthly. At typical SaaS pricing, that's thousands in recurring fees for what amounts to orchestration software.
"Your infra, your rules" isn't marketing copy — it's the architectural shift that lets voice agents scale past proof-of-concept.
Dograh's comparison table tells the strategic story. The platform offers full source-level customization, meaning you can fork and modify every component. You control data residency completely, critical for healthcare, finance, or any regulated industry where call recordings can't touch third-party clouds. Most importantly, you can bring your own LLM, speech-to-text, and text-to-speech providers — or switch them mid-flight without platform migration.
The maintainer profile adds signal. This isn't a side project from developers who might ghost in six months. YC alumni and exit founders have already proven they can build, ship, and sustain products. They're choosing to keep voice AI infrastructure open rather than build another SaaS moat. That suggests they see the real value higher in the stack, in vertical applications or managed services, not in controlling the pipes.
The Implication
If you're building voice agents today, the build-versus-buy calculus just shifted. SaaS platforms still win on speed for throwaway prototypes, but anything heading to production should evaluate self-hosted. The economic math is simple: one-time infrastructure cost versus perpetual per-minute fees.
Watch what happens to Vapi and Retell's pricing over the next six months. Open-source alternatives have a habit of forcing proprietary platforms to either race to the bottom on price or race to the top on features they can't easily replicate. Voice agent infrastructure just became a commodity. The value moved to what you build on top of it.