Documentation map

Use these pages for setup and operation, and the repository's own guides for detailed contracts.

Start here in the repository

• PyPI and the 0.1.0 GitHub release: the published package, release assets, and installation requirements.
• The README: install, speech services, where calls run, and setup through an agent.
• The docs index: every guide in the repository's docs/ folder.
• sleeptalk host --help: the recipe an agent follows to set up voice and start a call.

Use Sleeptalk

• Support matrix: what is supported, preview, and out of scope.
• Voice from a key you already have: OpenRouter, Venice, Groq, and Together as speech providers, and key reuse with host ensure.
• Where your keys live: the private store and what each page you type a key into does with it.
• Voices: curated voices per provider, custom voice IDs, and changing the voice during a call.
• Calls over your tailnet: --tailnet, tailscale serve, and the refusal codes.
• Discord setup: the bot token, permissions, intents, and who can talk to the bot.
• Voice modes: Cascade, Local speech, GPT Realtime, and GPT Live.
• Background workers: keep talking while several agent tasks run.
• Local speech and Just a model.
• Deployment: keep a profile running with launchd, systemd, or Docker.
• Multi-participant rooms: Discord rooms with more than one person or bot.
• Voice UX standard and voice speech patterns: conversation rules, wake and stop words, and pronunciation.
• sleeptalk.example.toml: every profile field, including the [policy] settings for turn taking.

Connect an agent

• Bring your own agent: command, HTTP, ACP, bridge, and plugin adapters, checked with sleeptalk adapter test.
• Agent Bridge protocol and the custom agent walkthrough.
• Agent adapter contract, agent conformance, and adding a built-in agent.
• Component contracts: register transports, agents, STT, and TTS.
• Examples: runnable adapter and session examples.

Operate, secure, and release

• Browser-link service operations: running the hosted relay Worker.
• Security policy and the caller authentication threat model.
• Release checklist and changelog.
• Contributing guide: setup, tests, and the release proof.

Decisions and research

• Architecture decision records.
• OpenAI Realtime and GPT Live browser preview: separate voice integrations, with different surface and concurrency limits.

The support matrix defines the supported product; preview features are labelled on their guides. Research notes and decision records explain implementation choices and do not expand that support promise.