Guides

Messaging & Voice

Messaging connects your chat app to SlyCode so you can drive any card's AI session from your phone: send prompts, get replies, switch cards, and talk by voice. Telegram is the supported channel today; the channel layer is an abstraction, so more channels are planned.

What messaging gives you

Once connected, your chat is a remote control for every session SlyCode manages. A message you send is routed to the AI session for whatever card or project you are pointed at, and replies come back to the same chat. You can:

Send prompts to any card's session and read the replies.
Navigate between projects and cards without opening the dashboard.
Dictate prompts as voice messages and get spoken replies back.
Send photos into a session and receive files out of it.

The bot is locked to a single authorized user: your Telegram user id. Messages from anyone else are ignored.

Telegram setup

Create a bot. Message BotFather on Telegram and follow the prompts. It gives you a bot token.
Add the token and your user id to the .env file in your SlyCode folder. The setup wizard asks for these during install; if you skipped that step, add them now.
Restart SlyCode with slycode restart.

.env

# Telegram channel
TELEGRAM_BOT_TOKEN=YOUR_BOT_TOKEN
TELEGRAM_USER_ID=YOUR_TELEGRAM_USER_ID

# Voice (optional)
STT_BACKEND=openai
OPENAI_API_KEY=YOUR_OPENAI_API_KEY
ELEVENLABS_API_KEY=YOUR_ELEVENLABS_API_KEY

Finding your way around

Every reply carries a breadcrumb showing where you are: the project and card context plus the current git branch and uncommitted-change count. You always know which session you are talking to before you send the next prompt.

Command	What it does
`/switch`	Navigate projects, then stages, then cards. Includes a cross-project Recent jump.
`/search`	With text: search for cards. Empty: quick-access view with active sessions first, then recent.
`/status`	Full state: current target, provider and model, git branch, and the card's status line.
`/provider`	Change the AI provider (Claude, Codex, or Gemini) for the current card or terminal, then pick a model.
`/global`	Jump to the global terminal.
`/project`	Jump to the current project's terminal.

When a reply originates from a session other than the one your chat is pointed at, it is labeled with its origin and comes with a "Switch to Card" button, so cross-session messages never masquerade as answers from the current card.

Voice input

Send a voice message and SlyCode transcribes it to text before routing it to the session, tagged so the AI knows it arrived by voice. The speech-to-text backend is selectable via STT_BACKEND:

Backend	Value	Needs
OpenAI Whisper API	`openai`	`OPENAI_API_KEY`
Local whisper.cpp	`local`	`WHISPER_CLI_PATH` and `WHISPER_MODEL_PATH`
AWS Transcribe	`aws-transcribe`	`AWS_TRANSCRIBE_S3_BUCKET`; region and language are optional. Uses the instance IAM role, no API key.

Voice output

Replies can come back as voice messages, rendered with ElevenLabs. Set ELEVENLABS_API_KEY in .env to enable it, then control delivery from the chat:

/mode picks text, voice, or both.
/voice searches and swaps ElevenLabs voices.
/tone sets the speaking tone from presets or free text.

Voice, mode, and tone can be set per project, falling back to your global choices. A voice reply from a project's session speaks in that project's voice even if your chat last navigated somewhere else.

Quick-launch shortcuts

/start with a token deep-links straight to a target instead of showing help: the global terminal, a project's terminal, a card by number, or a saved shortcut. Saved shortcuts are managed per project from the web UI and can carry an auto-fired prompt, a forced provider, and a prefer-existing-session flag, so one tap from your phone can land on a card and start work immediately.

Files and photos

Media moves in both directions. Send a photo (or several) and it is delivered into the current session as a screenshot the AI can look at. Going the other way, a session can ship files back to your chat: documents, audio, or video, with sensible size limits and a deny-list that refuses sensitive paths like .env files and SSH keys.

Next steps

Automations: schedule recurring prompts and let cards report back over messaging.
Configuration: the full .env and workspace settings reference.