How-To Guides

Task-oriented. Each page assumes you already know what you want to do — if you're looking for orientation, start with Tutorials.

TTS

• Switch TTS Backend — Fish ↔ Kokoro
• Clone a Voice — upload a reference clip and speak in that voice
• Run Without the Fish Sidecar — Kokoro-only, single container

LLM

• Use an External LLM — point at a gateway, a remote vLLM, or OpenAI

All-API setup

• Use LocalAI (all-API) — swap STT + TTS + LLM to OpenAI-compat endpoints; run protoVoice on a CPU box

Voice agent behaviour

• Configure Verbosity — tune filler chattiness (silent / brief / narrated / chatty)
• Backchannels — listener-acks ("mm-hmm") during long user turns
• Delivery Policies — now / next_silence / when_asked for async tool results
• Personas & Skills — swap voice + system prompt per skill YAML
• Users & API Keys — configure the auth roster (Infisical or config/users.yaml)
• Audio Handling — echo guard, half-duplex, noise filter, smart-turn

Extending the agent

• Build a Tool — sync vs async patterns, latency tiers, the result_callback gotcha

Fleet integration

• A2A Integration — inbound JSON-RPC + callback webhook + outbound dispatch

Ops

• Benchmarking — measure LLM / TTS / STT / A2A latency with scripts/bench.py