Changelog

The final 4.11.x stable release, replacing the withdrawn 4.11.6, fixes concurrent token counting, Windows plugin output, plugin runtime reconnects, first knowledge base ingestion, SQLite writes and Dify/Coze Runners.

• ImprovedToken counting works fully offline, and if counting times out or fails, LocalAgent estimates and keeps the conversation going.
• ImprovedOn adapters such as HTTP Bot and OneBot, Runner interactions like Dify human-input steps fall back to text questions, one form field at a time.
• FixedToken counts are no longer rejected when several conversations use the same model at once; if you installed 4.11.6, upgrade to this release.
• FixedOn Chinese-locale Windows, plugins printing Chinese text or emoji no longer disconnect the plugin runtime, and the first conversation no longer stalls downloading tokenizer files.
• FixedDuring startup or a plugin runtime reconnect, messages wait for plugins to be ready instead of failing or reporting a missing Runner.
• FixedThe first knowledge base ingestion on a slow machine no longer drops the plugin runtime or fails.
• FixedPlugin storage writes no longer fail on pip/uvx installs with SQLAlchemy 2.1, and SQLite database is locked errors are less frequent.
• FixedThe Box sandbox handles very long commands and large attachment imports.
• FixedCoze and Dify Runners no longer treat completed or workflow outputs as empty replies, and LocalAgent no longer errors on an empty reasoning setting.
• FixedThe !func command in the official EssentialCommands plugin works again.

Plugins and Agents can call Workspace APIs, speech models and cross-platform interactive cards arrive, and many platform and runtime issues are fixed.

• NewPlugins and Agents can search and install Marketplace plugins and read the Workspace language through the Workspace API.
• NewTTS and STT speech models with matching pipeline stages, including LangBot Space speech models.
• NewCross-platform interactive cards for QQ, WeCom, Feishu and DingTalk.
• NewA new LINE adapter, Slack thread replies, proactive sends for WeChat Official Accounts, QQ Official emoji parsing, and NVIDIA NIM as a model provider.
• NewNew official MulticaAgent, HandoffBridge and LeadCapture plugins, plus one-step upgrades of Marketplace plugins across all Workspaces.
• ImprovedMonitoring shows plugin-handled messages, the pipeline rule that dropped a message, and model calls plugins make outside a pipeline.
• ImprovedRunners receive images and files in their original order; official Runners such as Dify, Coze, Tbox and ACP are updated.
• FixedDeleted Pipelines and Agents no longer receive messages.
• FixedNon-streaming platforms deliver Runner replies one message at a time, and HTTP Bot /sync returns each reply only once.
• FixedWeCom reconnects after a failed heartbeat, and plugin runtimes start more reliably on Windows.

Official plugins upgrade independently, expired image links refresh automatically, and database migration, plugin runtime and platform issues are fixed.

• NewPlugin packages under development can be uploaded straight to LangBot Space, optionally linked to a GitHub repository.
• ImprovedOfficial plugins such as LocalAgent, LangRAG and DifyAgent ship on their own and can be upgraded without waiting for a LangBot release.
• ImprovedExpired OneBot image links are refreshed automatically in history and model requests.
• ImprovedSQLite databases reclaim space incrementally, using less disk.
• FixedCancelled executions no longer stay stuck “in progress”.
• FixedQQ Official group mentions and reply order, smoother WeCom streaming replies, and conversation context for LINE and HTTP Bot are fixed.
• FixedDatabase migration failures and rollbacks are fixed.

A fix release for dedicated deployments.

• FixedSigning in with LangBot Space on dedicated environments no longer returns 404, and adding the first model no longer fails.
• FixedPlugin installs reject archives with unsafe file names.

Improves monitoring performance, message handling and plugin compatibility.

• ImprovedSession lists in Monitoring load faster.
• ImprovedFailed plugin installs show clearer diagnostics.
• FixedFixes execution end states, message filtering, explicit attachment delivery and message handling on some platforms.
• FixedHistorical images are stored and migrated more reliably.

Bots can react to platform events, Agent runners become plugins, and the WebUI adds a Workspace Assistant.

• NewEvent-based bots: joins, leaves, recalls, reactions, card clicks and more can be routed to a Pipeline, an Agent or a plugin in Event Routing.
• NewEvent debugger: edit event payloads in a form and inspect run logs and execution details.
• NewNew Agent runners including Codex, Claude Code and ACP.
• NewWorkspace Assistant: look up resources and set up Pipelines and knowledge bases by chatting in the WebUI; changes run only after you confirm.
• NewPasskey sign-in and TOTP two-factor authentication.
• NewThe operation log records what each change did.
• NewSlack Socket Mode, DingTalk card interactions, the WeChat typing indicator, OneBot WebSocket and more.
• NewA guided tour and a scenario-based bot setup wizard.
• ImprovedMonitoring adds a unified execution view, tool-call tracing and a live in-progress panel.
• ImprovedMedia files are deduplicated, so the database grows more slowly.
• ChangedRunners such as LocalAgent, Dify, Coze and n8n are now official Marketplace plugins; upgrading installs them automatically and backs up your configuration.
• FixedFixes empty messages and early-ending streaming replies in the WeCom AI Bot.