AI chat

AI chat

storyvox has a per-fiction chat surface that lets you ask questions about what you're reading. Built on a multi-provider LLM router so you can pick whichever backend you trust with your reading.

Providers

Seven providers configurable in Settings → AI:

Provider	Auth	Notes
Claude (direct)	API key	Anthropic API key from console.anthropic.com
Anthropic Teams	OAuth	Browser-based handoff; Teams-plan accounts
OpenAI	API key	GPT-4o, o3, etc.
Vertex	API key	Google's GCP-hosted Anthropic models
Bedrock	AWS keys	Anthropic models via AWS
Foundry	Azure	Azure OpenAI
Ollama	Host URL	Local-first; point at http://your-box:11434

Each provider has model chips in Settings to pick which model to use. The active provider's chip is highlighted.

Grounding controls

What context does the chat see? Toggle independently:

• Include current chapter title — minimal grounding (~5 tokens).
• Include current sentence — the exact sentence you're listening to (~50 tokens).
• Include entire current chapter — full chapter text every turn (~2-5k tokens).
• Include entire book so far — every chapter you've heard so far (can be massive; cost scales).

The defaults are conservative (chapter title only). The grounding stack lets you balance cost / freshness / context as you go.

Quick-action chips

Above the input row, four quick-action chips pre-fill common questions:

• "Recap so far"
• "Who is …?"
• "What just happened?"
• "Summarize this chapter"

Tap a chip to drop the prompt into the input — edit before sending if you want to tweak.

"Who is X?"

Long-press a word in the reader → context menu → "Ask AI about …". Opens chat with the prompt pre-filled. Useful deep into a serial when names blur.

Recap aloud

The chapter recap modal generates a recap via the active LLM, and offers a "Read aloud" button that synthesizes the recap through the storyvox TTS pipeline (same engine, same voice). Hit pause to stop the recap mid-stream.

Sessions

Past chats live in LlmSessionRepository. The Sessions surface (Settings → AI → Sessions) shipped in v0.4.55+ (#218) — review history, delete sessions, jump back into one.

Read-aloud per turn

Every assistant turn in chat has a small Read aloud button next to it (#214). Tapping it routes the turn through the same TTS pipeline as chapter playback — same engine, same selected voice, same per-voice speed/pitch. Useful for the long-recap quick action, or for anything you'd rather hear than read.

Privacy

storyvox doesn't proxy your traffic. API keys are stored locally in DataStore. When you chat with Claude, your turn goes from your phone directly to api.anthropic.com — no relay. The "Forget all AI settings" button in Settings → AI wipes keys and history.