Discord setup

Run a Discord bot that joins one voice channel in your server and talks only to the people you allow.

How it fits

A Discord bot is its own Sleeptalk profile. It reuses the speech key of your browser profile, so you only add a bot token. By default the bot is closed: it is locked to one server, one voice channel, and your Discord user ID.

You need a Discord server where you can add apps, a standard voice channel (stage channels are not offered), and the Discord extra installed in the Python that runs Sleeptalk:

python -m pip install 'sleeptalk[discord]'

Voice also needs ffmpeg and libopus on the host. Sleeptalk checks for them, and for the optional davey package that adds Discord's DAVE voice encryption.

Set up from the settings page

Open the settings page with sleeptalk settings --open and go to Discord setup. The page walks through four steps.

If this profile is already a Discord bot that is not locked yet, the same page offers to finish steps 2 to 4 with the saved token. An unlocked closed bot will not start.

Other ways to set it up

• From a phone link. An agent that already talks to you in the browser can send a Discord setup link: sleeptalk host input link --request-id <id> --profile discord --surface discord --speech-from-profile default --workspace <dir> --agent <yours>. It opens the settings page on just the Discord steps: the bot token, adding the bot to your server, picking the channel, and locking it to your user ID. The bot's name, your servers, channels, and user ID travel encrypted; the hosted relay sees only ciphertext. The agent never sees the token. When you finish, the page stays open on the new profile's settings.
• From the terminal. Run sleeptalk --discord and follow the prompts. It asks for the bot token, then who can talk to the bot.

Invite link and permissions

Add the bot with the invite link Sleeptalk gives you. It requests the scopes bot applications.commands and exactly these permissions:

If you build an invite in the Developer Portal's URL generator instead, it omits applications.commands unless you tick it. The bot then logs Missing Access and no slash commands appear. Re-invite it with Sleeptalk's link; re-inviting a bot already in the server updates its permissions in place.

Channel overrides can deny permissions granted at the server level. Discord documents permission precedence.

Privileged intents

Voice needs no privileged Gateway intents. Leave Presence, Server Members, and Message Content off unless you turn on one of these features:

Sleeptalk requests Message Content only when one of those is set. If Discord refuses the connection, sleeptalk doctor names the toggle to enable.

Who can talk to the bot

Setup writes a closed access list into the profile:

[transport]
type = "discord"
authorization_required = true
guild_id = "YOUR_SERVER_ID"
allowed_channel_ids = "VOICE_CHANNEL_ID"
operator_ids = "YOUR_USER_ID"

A profile that does not set authorization_required = false refuses to start until guild_id, allowed_channel_ids, and operator_ids are set. With the list in force:

• Slash commands and text summons from anyone else are quietly rejected.
• Speech from an unlisted user is dropped before it reaches speech-to-text.
• The bot plays audio only while everyone in the channel is on the list. If someone else joins, playback stops for the rest of that session.
• Only an operator can answer the agent's tool permission requests.

Add people with comma-separated user IDs:

A speaker can ask the agent for anything its permission mode allows without a prompt. Stop the running bot before editing its profile, then start it again.

authorization_required = false lets anyone in the server summon the bot and talk to your agent, with whatever workspace access the agent has. The bot prints a warning at startup. Use it only in a server you fully control.

Join a call

Slash commands are registered only in the locked server. Use them in the chat of the allowed voice channel. Replace <profile> with the bot's profile name.

1. Join the voice channel in Discord.
2. In its chat, type /<profile> join. The bot joins and greets you.
3. Ask the agent which workspace it is in, and confirm you hear the reply.
4. Type /<profile> leave to disconnect.

Other commands include status, settings (narration, response length, and speech speed), ask for a text reply, and link for a private browser call link. A voice_only profile exposes only settings; you summon it by mentioning the bot while you are in the voice channel. Automatic joining when someone enters the channel is off by default.

Keep it running

sleeptalk run --config /path/to/discord.toml

To keep it running across logins, install it as a user service and start it with the command it prints. The settings page's Keep it running button does the same.

sleeptalk service install discord --yes

Use one bot per runtime. Do not run the same bot token from two places. See deployment for service details.

Common setup failures

• No slash commands, or Missing Access in the log: the bot was added without applications.commands. Re-invite it with Sleeptalk's link.
• Gateway close 4014: the profile turns on text summons or room reading. Enable Message Content in the Developer Portal, or turn that feature off.
• Bot refuses to start: the profile is closed but missing guild_id, allowed_channel_ids, or operator_ids. Finish the lock step on the settings page.
• Bot joins but does not hear you: your user ID is not in operator_ids or speaker_ids, or you are in a channel that is not allowed.
• Bot goes silent when someone joins: that person is not listed. Add them to listener_ids and rejoin.
• Opus errors on start: install libopus (brew install opus or sudo apt install libopus0), or set opus_library in the profile.

A passing setup check does not prove you can hear the bot. Always test a spoken exchange.