Zurück zum Blog
Ratgeber

OpenClaw 2026.2.21: Gemini 3.1, Discord Voice, and 100+ Fixes

Aktualisiert 21. Februar 2026 3 Min. Lesezeit
OpenClaw 2026.2.21

OpenClaw 2026.2.21 just dropped, and it's a big one. This release packs Google Gemini 3.1 support, Discord voice channel integration, major security improvements, and over 100 bug fixes. Let's break down what's new.

Headline Features

Gemini 3.1 Support

Google's latest model is now fully supported. You can use google/gemini-3.1-pro-preview as your primary model, giving you another powerful option alongside Claude and GPT-4.

Discord Voice Channels

This is a big one for Discord users. OpenClaw can now:

  • Join and leave voice channels via /vc commands
  • Auto-join configured channels for realtime voice conversations
  • Stream responses directly in voice
  • If you've wanted to talk to your AI assistant in Discord voice chat, now you can.

    Thread-Bound Subagent Sessions

    Discord threads just got smarter. Spawned helper agents can now be bound to specific threads with per-thread focus and list controls. This means you can have dedicated AI agents working in different threads without them stepping on each other.

    Discord Streaming

    Live draft replies are now available on Discord with partial/block streaming options and configurable chunking. The same streaming experience Telegram users have enjoyed is now on Discord.

    Security Hardening

    This release includes significant security improvements:

  • SHA-256 everywhere: Gateway locks and tool-call IDs now use SHA-256 instead of SHA-1
  • Owner ID obfuscation: Now uses a dedicated HMAC secret, decoupled from gateway tokens
  • Retry loop protection: Capped internal retry cycles to prevent unbounded loops (GHSA-76m6-pj3w-v7mf)
  • Duplicate bot detection: Telegram now detects duplicate bot tokens at startup, preventing endless polling conflicts
  • Security Advisory

    The retry loop fix addresses a security advisory (GHSA-76m6-pj3w-v7mf). If you're running an older version, update as soon as possible.

    iOS and Watch Improvements

    The mobile experience got some love:

  • Cleaner chat UI: Stripped metadata noise, formatted tool outputs into concise summaries
  • Watch integration: Quick-reply actions from Apple Watch notifications now bridge properly to iOS
  • Gateway stability: Background wake and reconnect behavior is more reliable, with better handling of BGAppRefresh and APNs
  • Prompt Caching Tweaks

    Cache hit percentages are now properly included in /status output, and the caching behavior has been refined across providers.

    Other Notable Changes

    New Providers

  • Volcano Engine (Doubao) and BytePlus support, including coding variants
  • Channel Improvements

  • Per-channel model overrides via channels.modelByChannel
  • Simplified Telegram streaming config (just channels.telegram.streaming: true)
  • Configurable lifecycle status reactions for queued/thinking/tool/done/error phases
  • Ephemeral defaults for Discord slash commands
  • Subagent Defaults

  • Default spawn depth is now 2, allowing depth-1 orchestrator spawning by default
  • 100+ Bug Fixes

    The changelog lists dozens of fixes, but here are some highlights:

  • Fixed memory search failures that could crash tools on missing files
  • Fixed Telegram streaming cleanup when runs fail
  • Fixed TUI backspace issues in SSH sessions (no more double-deletes)
  • Fixed Discord mention spam in block-streamed replies
  • Fixed pairing flow for fresh installs
  • Fixed heartbeat suppression when HEARTBEAT.md is missing
  • Added HTTP 503 as a failover-eligible error for LLM providers
  • How to Update

    If you're running OpenClaw locally:

    Code Bash
    npm install -g openclaw@latest
    openclaw doctor
    openclaw gateway restart

    If you're on ClawHosted, you can update with one click from your dashboard.

    ---

    Links

  • Full Changelog on GitHub
  • OpenClaw Documentation
  • Community Discord
  • Von

    Renaud

    Clearly not an OpenClaw bot!

    Ähnliche Beiträge