Back to home@GooDAnDReaDY

dsh-messenger-gateway

Telegram messenger bridge for DeepSeek Harness (sessions, steer, homes, TTS voice notes)

Stars
0
Language
JavaScript
Created
Aug 26, 2026
Updated
Aug 26, 2026

Introduction

@goodandready/dsh-messenger-gateway

Telegram messenger bridge for DeepSeek Harness.

Talk to your Harness agent from Telegram: text, voice, photos, documents, inline buttons, named homes, and optional spoken replies.

Features

  • Long-poll or webhook Telegram bot
  • Allowlist + pairing codes
  • Per-user or per-chat sessions (sessionScope)
  • Forum topics as separate sessions
  • Steer: follow-up messages while the agent is busy (instead of aborting)
  • /stop, /new, /model, /status, /voice, /sethome, /home
  • Agent tool messenger_ask (inline keyboard answers return to the agent)
  • Named homes for outbound notify / messenger.send
  • Optional notify bridge: web session events → Telegram home (excludes messenger sessions)
  • Inbound voice → STT (dsh-voice), photos → vision (dsh-vision-bridge)
  • Optional TTS replies (dsh-tts); mp3 is converted to OGG/Opus via ffmpeg for Telegram voice notes

Discord is listed in settings as a placeholder only — the Discord adapter is not implemented yet.

Install

dsh plugin --profile web add @goodandready/dsh-messenger-gateway

Then open Settings → Plugins → Messenger gateway:

  1. Enable Telegram
  2. Paste the BotFather token (write-only; leave blank to keep the current token)
  3. Set allowed Telegram user IDs (or use pairing)

Optional companion plugins (same profile)

PluginRole
@goodandready/dsh-voiceTranscribe inbound voice messages
@goodandready/dsh-ttsSpeak agent replies
@goodandready/dsh-vision-bridgeDescribe inbound photos

They are not npm dependencies of this package — install them separately if you want those features.

Voice notes

Spoken replies use Telegram sendVoice. If TTS returns MP3 (or other non-Opus audio), the gateway runs ffmpeg (libopus) to produce OGG. If ffmpeg is missing or conversion fails, the audio is sent as a regular audio file instead of a voice note.

Commands (Telegram)

CommandDescription
/start /helpHelp
/whoamiYour Telegram user id
/newNew agent session
/stopAbort the current turn
/model /statusModel / gateway status
/voice on|offPer-user spoken replies
/sethome [name]Bind current chat/topic as a named home
/homeList homes

Agent tools & HTTP

  • Tool: messenger_ask — ask the user with inline buttons; choice is fed back into the turn
  • HTTP (when enabled): /messenger/send, /messenger/ask, /messenger/progress
  • Cordis service: ctx.messenger for other plugins

Configuration notes

  • sessionScope: user (default) or chat — how group chats isolate sessions
  • voiceMode: mirror / always / off — when to speak replies (also /voice)
  • tts.enabled / tts.maxChars — TTS gate and length cap
  • notifyBridge — forward non-messenger web session events to a home
  • Bot token is a DSH secret field — never commit it

Requirements

  • DeepSeek Harness web (or compatible) profile
  • Node.js matching your Harness install
  • For voice notes from non-Opus TTS: ffmpeg on the host PATH

License

MIT