Skip to main content

Ecosystem

Piopiy supports a broad ecosystem across STT, LLM, and TTS. This page helps you choose and operate a stack without vendor ranking claims.

Scope​

Piopiy integrates with:

  • Speech-to-Text (STT) for transcription
  • Large Language Models (LLM) for reasoning
  • Text-to-Speech (TTS) for spoken responses
  • Telephony/WebRTC transport managed by Piopiy

For the full catalog, see Supported Services.

How to Select Providers​

Choose providers using measurable requirements from your product and operations constraints:

  • Latency target: acceptable time to first audio response and full-turn completion
  • Quality requirements: transcription accuracy, reasoning quality, voice naturalness
  • Language coverage: caller languages, accents, multilingual switching needs
  • Reliability: uptime expectations, rate-limit behavior, fallback options
  • Compliance and data residency: regional processing and policy requirements
  • Cost model: expected call volume, peak bursts, and margin targets
  • Operational fit: observability, retries, timeout controls, error handling

Sample Stack Patterns​

These are implementation patterns, not recommendations or rankings.

Pattern A: Cloud-Native Baseline​

  • Use one cloud STT provider
  • Use one cloud LLM provider
  • Use one cloud TTS provider
  • Add Silero VAD for interruption handling
pip install "piopiy-ai[deepgram,openai,cartesia,silero]"

Pattern B: Quality-Focused Cloud​

  • Use provider combination chosen by your quality tests
  • Keep one backup STT/TTS path configured
pip install "piopiy-ai[assemblyai,anthropic,elevenlabs,silero]"

Pattern C: Local/Hybrid Deployment​

  • STT and/or LLM can run locally
  • TTS can be local or self-hosted
  • Keep transport and session handling in Piopiy
pip install "piopiy-ai[whisper,piper,silero]"

For local LLM runtime options, see Ollama (Local).

Runtime Switching and Fallback​

Use ServiceSwitcher for controlled runtime switching and fallback:

  • Fail over to secondary providers on timeout/error thresholds
  • Route traffic by policy (tenant, region, tier, campaign)
  • Switch voice/STT profile for specific call flows

Reference:

Validation Workflow​

Before production rollout:

  1. Define your KPIs (latency, ASR error rate, call success rate, cost/call)
  2. Test at realistic concurrency and call duration
  3. Compare at least two stack patterns using the same test script
  4. Verify failure behavior (timeouts, retries, provider outage)
  5. Promote only after SLOs are met

Implementation References​