Audio and podcasting decision
Vapi
A small team can build a useful voice-agent workflow (calls→STT→LLM→TTS) and run it for development or low-volume production, but reproducing Vapi's enterprise compliance, SLAs, and high-scale low-latency infrastructure is expensive and slow to match.
Visit website↗No open-source build does this yet
Nothing published replaces this one, so a replacement starts from an empty file. Here is what it would have to cover.
Code Vapi publishes itself
Not a way out of the subscription - these are the vendor’s own repositories. Worth a look for how they build, and for anything you would have to integrate with.
- VapiAI/client-sdk-python★124 GitHub starsMITThis package lets you start Vapi calls directly in your Python application.↗
- VapiAI/server-sdk-typescript★121 GitHub starsThe official TypeScript SDK for accessing Vapi's API↗
- VapiAI/docs★77 GitHub starsMITVapi docs↗
- VapiAI/vapicon-2025-hardware-workshop★68 GitHub stars↗
What a replacement has to do
- Answer inbound phone calls: accept audio, stream to STT, send transcripts to an LLM for dialog logic, synthesize responses via TTS, and route audio back to the caller while logging calls.
What it still won’t have
- Enterprise SLAs and guaranteed uptime
- SOC2/HIPAA/PCI compliance and associated attestations
- High-scale infrastructure optimizations for sub-500ms latency at millions of calls
- Dedicated deployment/support team and managed integrations
- Built-in guardrails and enterprise data residency features
What remains hard
- Compliance and regulation
SOC 2, HIPAA, and PCI compliant
- Infrastructure at scale
Scale up to millions of calls with sub-500ms latency.
First-year cost
Keep paying
Paying is—cheaper in year one.
On cash alone, building overtakes the subscription at 16 seats.
Money you would actually spend
Time you would spend
—
What you would spend
What we assumed
The verdict above measures whether you could build it. This one is only about money.
Runnable build prompt
Build a minimal hosted voice-agent platform in Node.js (Express) + Postgres + Docker that accepts Twilio webhooks, streams audio to a streaming STT (OpenAI/Whisper or commercial streaming STT), forwards transcripts to an LLM (OpenAI/Anthropic) to produce reply text, converts reply text to audio via a TTS API, and returns audio to the caller. In scope: SIP/PSTN integration via Twilio, streaming STT integration, LLM request/response handling, TTS playback, storing call transcripts and metadata in Postgres, a basic web UI to view recent calls and logs, error handling for API failures, retries, rate limits, and automated tests (unit for core modules and end-to-end smoke test against a sandbox Twilio number). Out of scope: SOC2/HIPAA attestation, multi-region scaling, custom voice model training, commercial SLA. Provide Docker Compose for local dev, deployment Dockerfile, CI to run tests, and a README with run and cost-estimate instructions.
How we checked
How the score was reached
- Pay verdict base20
- 2 cited sources+1
- Price verified on pricing page+3
- Hard moats found in the evidence-6
- Evidence score18
The base comes from the verdict. Everything under it is a check that either happened or did not, and each one is a fact frozen in this record rather than a judgement made at render time - so the same evidence always produces the same number.
How scoring works →Cited sources · 2
Every page the run actually retrieved.
- official productVapi - Build Advanced Voice AI Agents
- official pricingVapi Pricing
Integrity checks
What held up, and what did not.
