The anti-Vapi: a phone stack you can actually own

What it does

Patter is a full voice-AI stack in Python and TypeScript that connects your agent to the phone network. You pick the LLM, speech engine, text-to-speech provider, carrier, and even the voice mode — Realtime, Pipeline, or Hybrid — and Patter runs the agent loop, audio processing, and telephony plumbing. It claims 27+ provider integrations across six layers, from Twilio and Telnyx to OpenAI Realtime and Deepgram.

The interesting bit

The vendor-swapping pitch is unusually concrete: change carrier or model with “one line.” More practically, it ships an automatic LLM fallback chain that fails over mid-call if a provider chokes, plus vendor-neutral OpenTelemetry tracing — the kind of operational hygiene most voice startups charge enterprise rates for. The built-in Cloudflare tunnel (tunnel: true) for local dev is a nice touch; no ngrok account required to test against real phone numbers.

Key highlights

• Python and TypeScript SDKs at “full parity” — same API surface, same hooks
• Three voice architectures: all-in-one realtime, composable pipeline, or hybrid
• Built-in tools, call transfer, and guardrails that behave identically across Twilio, Telnyx, and Plivo
• Anthropic Agent Skills bundle for ~55 coding-agent harnesses (Claude Code, Cursor, etc.)
• Eight self-contained templates including outbound calls with AMD/voicemail drop and a production-ready dashboard

Caveats

• The “4 lines” quickstart is technically true but omits the credential environment-variable dance; real setup is more involved
• Provider parity claims are hard to verify without testing edge cases — some integrations are likely thinner than others
• 564 stars is respectable but early; production battle-testing at scale is unproven

Verdict

Worth a look if you’re building voice agents and Vapi’s pricing or data residency makes you nervous. Skip it if you need a managed SLA or don’t want to operate your own telephony stack — this is still very much “bring your own infrastructure.”