rina-bot

RINA AI on the messaging channels you already use.

v0.1 ships Telegram support. More channels (Discord, Slack, WhatsApp) on the roadmap — same BotBrain, swappable ChannelAdapter.

Install

npm install -g @siliconcorerina/rina-bot

You also need a model API key. The bot uses the same backend abstraction as rina-agent:

export DEEPSEEK_API_KEY=sk-...
# or OPENAI_API_KEY / ANTHROPIC_API_KEY / MISTRAL_API_KEY

Quick start (Telegram)

1. Create a bot with @BotFather

1. Open Telegram, search for @BotFather, start a chat.
2. Send /newbot.
3. Pick a name (MyRinaAI) and username (my_rina_ai_bot).
4. Copy the token it gives you (123456:ABC-DEF...).

2. Find your Telegram user id

1. Open Telegram, search for @userinfobot, start a chat.
2. It replies with your numeric Id: 12345678. Copy that.

3. Set env vars and run

export TELEGRAM_BOT_TOKEN=123456:ABC-DEF...
export TELEGRAM_ALLOWED_USER_IDS=12345678         # comma-separated
export DEEPSEEK_API_KEY=sk-...
rina-bot start

On Windows PowerShell:

$env:TELEGRAM_BOT_TOKEN = "123456:ABC-DEF..."
$env:TELEGRAM_ALLOWED_USER_IDS = "12345678"
$env:DEEPSEEK_API_KEY = "sk-..."
rina-bot start

4. DM your bot

Open Telegram, find your bot by its @username, send a message. It replies.

What it does

• Per-chat memory — each Telegram conversation keeps its own history. Up to ~30 turns retained, oldest dropped first.
• Built-in commands:
  • /start, /help — show the help message
  • /reset, /clear — wipe the current chat's history
• Long messages are auto-split across multiple Telegram bubbles (provider limit: 4096 chars/message).
• Network blips are tolerated — the bot retries getUpdates on transient failures instead of crashing.
• Graceful shutdown — Ctrl-C or SIGTERM stops polling cleanly.

Safety

A messaging bot is internet-exposed — anyone who knows your bot's username can DM it. The defaults are tuned for that reality:

• Allowlist: TELEGRAM_ALLOWED_USER_IDS is a comma-separated list of user ids that may use the bot. Anyone else gets a polite rejection without touching your API quota.
• No file writes / no shell by default. The bot is a chat surface, not the autonomous rina-agent surface — the model cannot edit files or run commands from a Telegram message.
• --allow-writes opt-in: if you understand the risk and trust every user on your allowlist, this flag wires in the full rina-agent tool set. Don't use it in a group/channel.

If TELEGRAM_ALLOWED_USER_IDS is empty, the bot prints a loud warning and accepts messages from anyone. That's fine for a quick demo on a private bot, not for anything beyond.

Configuration

| Flag | Default | Meaning | |---------------------|----------------------------|-------------------------------------------| | --backend SPEC | $RINA_BACKEND or deepseek:deepseek-chat | Provider:model | | --lang en\|fr | $RINA_LANG or en | Reply language | | --workdir, -C DIR | cwd | Workdir the agent is scoped to | | --allow-writes | off | Enable write_file / edit_file / shell |

Environment variables

| Variable | Required | Meaning | |------------------------------|----------|------------------------------------------------| | TELEGRAM_BOT_TOKEN | yes | From @BotFather | | TELEGRAM_ALLOWED_USER_IDS | rec. | Comma-separated allowlist of user ids | | DEEPSEEK_API_KEY / OPENAI_API_KEY / ANTHROPIC_API_KEY / MISTRAL_API_KEY | one of | Backend auth | | RINA_BACKEND | no | Default --backend | | RINA_LANG | no | Default --lang |

Roadmap

• v0.2 — Discord adapter
• v0.3 — Slack adapter (Bolt)
• v0.4 — WhatsApp adapter (whatsapp-web.js, with the usual caveats)
• v0.5 — Persistent per-chat sessions across restarts
• v0.6 — Voice on macOS / iOS (TTS + STT)

License

MIT. See LICENSE at the repo root.

Part of the RINA AI project.