GooDAnDReaDY/dsh-im-hub-media

README 摘要

dsh-im-hub-media Telegram-first fork of dsh-im-hub for DeepSeek Harness (dsh) that adds media support to the multi-platform IM gateway. Talk to your dsh agent from Telegram (voice, photos, documents, video, replies), with Feishu/Lark and WeCom kept from the original. One agent per chat. Multi-turn context. Whitelist access control. Idle reaping. All the dsh-im-hub adapters, plus a media pipeline on Telegram. What's added vs. dsh-im-hub Area Feature Media Incoming photos / voice / documents / video are downloaded to a local cache; supported docs are injected inline or kept for the agent. Voice → text (STT) Voice messages transcribed automatically: Deepgram primary ( api.deepgram.com ), HuggingFace Whisper ( openai/whisper-large-v3 ) fallback. On-device vision Photos without a caption are routed to a vision model (see dsh-vision-bridge ) instead of failing a text-only turn. Outbound media Agents can emit MEDIA: markers and the bot sends the file back into the chat. Reply handling Telegram reply to message is passed through so the agent can answer in-context. Supported inbound document types follow the Hermes allowlist: PDF, Markdown, plain text, CSV, logs, JSON/XML/YAML/TOML/INI/CFG,…

分类