Discord

Discord works well for community-facing use cases and casual personal use. The setup requires creating a bot application in Discord's Developer Portal — slightly more involved than Telegram, but well-documented. Discord runs on any supported OpenClaw host, and costs only your AI provider's API fees — there's no Discord subscription or per-server charge.

New to OpenClaw? The installation guide covers getting the gateway running first. If you want to validate the setup with a simpler channel, try Telegram — it takes about 2 minutes — then come back here.

export DISCORD_BOT_TOKEN="your-bot-token-here"

{
  "channels": {
    "discord": {
      "enabled": true,
      "dmPolicy": "pairing",
      "groupPolicy": "mention",
      "requireMention": true,
      "botToken": "${DISCORD_BOT_TOKEN}",
      "allowedGuilds": [
        "123456789012345678"
      ],
      "allowedChannels": [
        "987654321098765432"
      ]
    }
  }
}

openclaw gateway restart
openclaw logs --filter discord

@YourBot are you there?

Auto Threads

When threadReplies: true is enabled, OpenClaw auto-creates threads for new conversations. By default, thread names are based on the triggering message content.

You can set autoThreadName: "generated" to have threads renamed asynchronously with concise LLM-generated titles:

{
  "channels": {
    "discord": {
      "threadReplies": true,
      "autoThreadName": "generated"
    }
  }
}

When set to "generated", the thread is created immediately (so replies aren't delayed) and then renamed in the background once the LLM produces a short title. The default message-based naming remains if you omit this setting.

Stability Considerations

Discord's WebSocket gateway can experience crashes, zombie connections, and rate limiting during high traffic. A zombie connection is when the bot appears online but isn't receiving events — the worst kind of failure because it's silent.

OpenClaw centralizes Discord gateway error handling behind a lifetime-owned supervisor, so early, active, and late-teardown errors are classified consistently and no longer surface as process-killing crashes.

Enable reconnection logic in your config:

{
  "reconnect": {
    "enabled": true,
    "maxRetries": 10,
    "retryDelay": 5000,
    "heartbeatInterval": 30000
  }
}

This automatically detects and recovers from dropped connections.

Timeouts (v2026.4.12)

If the inbound Discord worker times out before a final reply starts, OpenClaw now sends a visible timeout reply to the user — including in auto-created threads and queued runs. Previously, timeouts could result in silent failures with no feedback in the channel.

Media Handling (v2026.4.12)

• Media cap raised to 100MB (previously 25MB), bringing Discord closer to Telegram's limits for large video and audio attachments
• Generated images are now properly tracked with real MEDIA: paths instead of missing local files
• Component-only media sends are preserved without extra plain-output requeueing
• Text-plus-video deliveries are intelligently split into a text reply followed by a media-only send, ensuring video attachments render properly

Tool-Call Text Handling (v2026.4.15)

Voice Sessions (v2026.5.20)

Voice user-following is opt-in per Discord account. Once configured, the bot joins the voice channel a tracked user enters (within the allowedChannels list), hands off cleanly between users, and recovers DAVE end-to-end encryption state across reconnects.

To disable profile context in voice instructions — useful for a privacy-conscious voice agent, or to reduce voice latency on slow networks — set:

{
  "voice": {
    "realtime": {
      "bootstrapContextFiles": []
    }
  }
}

The default includes IDENTITY.md, USER.md, and SOUL.md (bounded to keep voice latency low). Empty array opts out entirely.

Discord voice is also the first live source for transcript capture — your agent can auto-capture conversations from voice sessions and surface them via openclaw transcripts. This started as the Meeting Notes plugin in v2026.5.22 and became OpenClaw's core transcripts tool in v2026.5.26. See the CLI reference for details.

Component Callback Lifetime (v2026.5.22)

{
  "channels": {
    "discord": {
      "agentComponents": {
        "ttlMs": 3600000
      }
    }
  }
}

The default is short-lived (suitable for ephemeral confirmations). Raise it when your bot posts components that users may not interact with immediately — approval workflows, polls, long-running task UIs.

Model Picker for Large Catalogs (v2026.5.26)

If you run Discord with a wide model catalog (for example a provider/* wildcard), the in-channel model picker no longer makes you paginate through prev/next pages. When a provider list — or a single provider's model list — exceeds 25 items, the picker shows an alpha-bucket select so the right model is one click away:

A–G (12) · H–N (18) · O–Z (5)

It falls back to numeric chunks when every item shares the same first letter. Live Discord runs also surface useful in-progress context in progress drafts as of v2026.5.28, so you can see what the agent is doing mid-run instead of waiting for the final reply.

Slash Command Access Policy (v2026.4.23)

Recent Fixes (v2026.4.12)

• Media cap raised to 100MB — supports larger generated videos and audio files
• Generated images properly tracked with real file paths
• Component-only media sends no longer produce duplicate plain text
• Text-plus-video deliveries split correctly so both arrive
• REST, webhook, and monitor traffic stays on the configured proxy
• @everyone and @here mention gates are honored
• ACK reactions stay on the active account
• Voice connect/playback timeouts are split so auto-join is more reliable
• Reply tags no longer leak control metadata like [[reply_to_current]] into preview text
• Explicit reply-tag threading is honored during final delivery
• Reply-to-batched delivery uses correct replyToMode: "batched" instead of legacy flags
• Visible timeout replies sent when worker times out before final reply starts
• Numeric Discord server and channel IDs in config are now auto-coerced to strings — no more startup failures if you paste a raw number into allowedGuilds or allowedChannels
• Auto-thread names now work correctly with reasoning models (enough output budget for thinking + title text)

Related channels

• Telegram — simpler setup, better for personal 1:1 use
• Slack — better for work teams with existing Slack workspaces

Evaluating OpenClaw against other AI tools? See OpenClaw vs ChatGPT — Discord integration is one of the things ChatGPT simply doesn't offer.

Security Recommendations

Never grant Administrator permissions. Your bot only needs to read and send messages in specific channels.

Restrict to allowedChannels. Don't let the bot operate in every channel — create a dedicated #assistant channel.

Use threads. Configure threadReplies: true to keep conversations organized and reduce rate-limiting risk.

Avoid public servers. If your bot is in a public server, use allowedGuilds to restrict it to servers you control.

Rate Limiting

Discord enforces rate limits on bots. If you're sending many messages quickly (e.g., from an automation), you may hit these limits.

To avoid rate limiting:

• Use threadReplies: true — threads have more permissive limits than main channels
• Add response delays for automated messages
• Don't use Discord for high-frequency automations — use Slack or iMessage instead