@decentchat/decentchat-plugin

OpenClaw channel plugin for DecentChat -- peer-to-peer encrypted chat over WebRTC.

Connects your OpenClaw agent to DecentChat workspaces so it can read and reply to messages in channels, groups, DMs, and threads.

Install

openclaw plugins install @decentchat/decentchat-plugin

Configure

The quickest way to set up is with the interactive wizard:

openclaw configure

Select DecentChat when prompted. The wizard will:

1. Offer to generate a new 12-word seed phrase (or let you paste an existing one)
2. Ask for a display name
3. Ask for an invite URL to join a workspace

You can also set the seed phrase via the DECENTCHAT_SEED_PHRASE environment variable instead of storing it in the config file.

Manual configuration

If you prefer to edit the config directly, add a channels.decentchat block to your OpenClaw config (~/.openclaw/openclaw.json or per-project):

channels:
  decentchat:
    enabled: true
    seedPhrase: "your twelve word BIP39 mnemonic goes here ..."

That's the minimum. The bot will join the DecentChat network using the default signaling server (https://0.peerjs.com/), respond to all messages, and call itself "DecentChat Bot".

Optional settings

| Key | Default | What it does | |-----|---------|-------------| | signalingServer | https://0.peerjs.com/ | PeerJS signaling endpoint | | invites | [] | DecentChat invite URIs to auto-join | | alias | "DecentChat Bot" | Display name for the bot | | dataDir | auto | Override the data directory | | dmPolicy | "open" | DM access: open, pairing, allowlist, or disabled | | streamEnabled | true | Stream responses token-by-token | | replyToMode | "all" | Reply behavior: off, first, or all |

Reply mode per chat type

You can override replyToMode for specific chat types:

channels:
  decentchat:
    enabled: true
    seedPhrase: "..."
    replyToMode: all
    replyToModeByChatType:
      direct: "off"
      group: "all"
      channel: "all"

Thread settings

channels:
  decentchat:
    thread:
      historyScope: "thread"     # "thread" or "channel"
      inheritParent: false
      initialHistoryLimit: 20

• historyScope: thread -- the bot only sees messages in the current thread (not the full channel history)
• inheritParent: false -- don't prepend the parent message to thread context
• initialHistoryLimit -- how many prior thread messages to load for context

Huddle (voice) settings

channels:
  decentchat:
    huddle:
      enabled: true
      autoJoin: false
      sttEngine: "gemini"         # whisper-cpp | whisper-python | openai | groq | gemini
      whisperModel: "gemini-2.5-flash"
      sttLanguage: "en"
      sttApiKey: "${GEMINI_API_KEY}" # used by cloud STT engines (openai/groq/gemini)
      ttsEngine: "gemini"         # elevenlabs | gemini
      ttsModel: "gemini-2.5-flash-preview-tts"
      ttsApiKey: "${GEMINI_API_KEY}" # optional; falls back to sttApiKey or env
      ttsVoice: "Kore"            # Gemini voice name (or ElevenLabs voice id when ttsEngine=elevenlabs)
      vadSilenceMs: 800
      vadThreshold: 0.5

Company simulation

The plugin includes a company simulation subsystem (@decentchat/company-sim) that lets you run multi-agent teams inside DecentChat workspaces. Configure it per-account:

channels:
  decentchat:
    companySim:
      enabled: true
      manifestPath: "./company-manifest.yaml"
      companyId: "acme"
      employeeId: "bot-1"
    companySimBootstrap:
      enabled: true
      mode: "runtime"

Multiple accounts

Run multiple bot identities from one OpenClaw instance:

channels:
  decentchat:
    defaultAccount: "main"
    accounts:
      main:
        seedPhrase: "first mnemonic ..."
        alias: "Bot A"
      secondary:
        seedPhrase: "second mnemonic ..."
        alias: "Bot B"
        dmPolicy: "disabled"

Each account gets its own peer connection and data directory.

Quick safety toggles

If the bot is too chatty:

# Shut off all replies
replyToMode: "off"

# Keep replies but use full channel history (no per-thread split)
thread:
  historyScope: "channel"

# Disable thread context prefill
thread:
  initialHistoryLimit: 0

How it works

The plugin creates a DecentChat peer (using @decentchat/protocol and @decentchat/transport-webrtc) that joins the P2P mesh. Incoming messages are routed through OpenClaw's agent pipeline. Responses are sent back to the originating chat, with thread-aware routing so replies land in the right thread.

All traffic is end-to-end encrypted (ECDH + AES-GCM-256). The bot's identity is derived from the seed phrase, same as any other DecentChat client.

Dependencies

• @decentchat/protocol -- DecentChat SDK
• @decentchat/transport-webrtc -- WebRTC transport layer
• @decentchat/company-sim -- company simulation subsystem
• openclaw -- peer dependency (the host runtime)

License

MIT