What's new in LangBot v4.11.8

GitHub release notes

Share prompts, Skills, MCP servers and plugins to LangBot Space from the WebUI, get Workspace failure alerts, pick a sandbox backend per Workspace, and use a 60% smaller Docker image.

  • NewShare prompts, Skills, remote MCP servers and local plugins to LangBot Space from the WebUI; credentials are blocked or replaced with placeholders before upload.
  • NewThe Pipeline and Agent prompt editors gain a prompt library to search, preview, replace or append community prompts, with favorites synced to LangBot Space.
  • NewWorkspace alerts notify admins by email, LangBot Space or webhook when a bot goes offline or a Pipeline or model keeps failing, and close on recovery.
  • NewEach Workspace can choose Docker, E2B or nsjail as its sandbox backend with resource limits, and files can move between Core, Runners and sandboxes.
  • ImprovedSkills are saved as versioned packages with their files and can be exported and shared.
  • ChangedThe default Docker image shrinks from 768 MB to 308 MB; if you use SeekDB, Milvus or Valkey as the vector database, switch to the matching -seekdb, -milvus or -valkey image or pip extra.
  • ChangedEmail settings move to a top-level email section supporting SMTP and Resend, and old settings migrate on startup; back up data/ before upgrading.
  • SecurityOutbound URL checks now refuse all non-public addresses, and DifyAgent no longer logs form tokens or signed URLs.
  • FixedLocalAgent falls back to a backup model or reports an error on empty model answers, and Qwen3.5/3.6 on vLLM/SGLang no longer fails with Invalid request parameters when a knowledge base is attached.
  • Fixedn8n streaming workflow errors no longer report success, Agent Runners keep token usage when cancelled or disconnected, and LangBot Cloud no longer logs you out on brief server errors.