github agentscope-ai/QwenPaw v1.1.4

latest release: v1.1.4.post1
15 hours ago

✨ Added

Agent System

  • Memory & Context refactor: Refactor long-term memory module with pluggable backends, automatic memory summarization every N conversation turns, automatic memory retrieval, and a new context management interface (#3548)
  • Plan Mode: Opt-in structured planning mode — use /plan to create step-by-step task plans that the agent follows, with a live plan panel in the console (#3686, #3787)

Providers

  • DeepSeek V4 Models: Added DeepSeek V4 Flash and V4 Pro as built-in models (#3797)

Security

  • Configurable Shell Evasion Checks: Toggle individual shell evasion detection rules on or off from the Security settings page, with per-check risk-type tagging (#3694, #3715, #3781)
  • Auth-Bypass Host Whitelist: Configurable allow_no_auth_hosts list for API access without authentication, with IP validation and security warnings in the console (#3739)

Tools

  • Browser Launch Parameters: Pass custom Chromium flags and a custom browser executable path to browser_use (#3556)
  • Shell Command Timeout: Configurable default timeout for execute_shell_command per agent (#3732)

Console & UI

  • Per-Agent Model Assignment: Assign a specific LLM model to each agent from Settings → Agents (#3599)
  • Session View Button: "View" action on the Sessions table to jump directly into the corresponding chat session (#3615)
  • Session Right-Click Context Menu: Right-click on chat session items for quick open, rename, pin/unpin, and delete actions (#3770)

Channels

  • SIP Voice Channel: New SIP phone channel with dual-mode backends — pyVoIP for development and LiveKit for production (#3449)
  • DingTalk Voice Recognition: Use DingTalk's built-in voice recognition instead of downloading and transcoding audio locally (#3681)
  • Telegram Command Menu: Expose /model and /stop in the Telegram bot's command menu for discoverability (#3714)
  • QQ Quoted Message Support: Parse reply/quote metadata and inject quoted text and attachments into agent requests (#3735)
  • DingTalk Markdown Messages: Send shorter replies as DingTalk Open API markdown for richer message rendering (#3771)
  • DingTalk Cron Message Type: Independent message type for scheduled/cron sends, allowing card-style messages for cron while using markdown for interactive chat (#3778)
  • WeChat Interactive Wizard: Interactive configure_weixin wizard in the CLI channel setup flow (#3699)

🔄 Changed

  • Tool Guard Approval System: Refactored tool-guard with interactive approval cards in the console, cross-session approval routing (approve sub-agent tool calls from the main session), configurable execution levels (Strict/Smart/Auto/Off), REST API and /approval CLI command (#3656, #3737)
  • Dynamic Plugin Registration: Replaced the build-time generated host module registry with runtime import.meta.glob-based registration (#3676, #3692, #3696)
  • Docker Build Improvements: Moved config.json initialization from image build to container entrypoint for volume-mounted deployments, and replaced pip with uv for faster package installs (#3698, #3723, #3769)
  • Agent Language Sync: New agents now inherit the console UI language (#3720)
  • Agent Selection Persistence: Agent selection state persisted in localStorage so it survives browser restarts (#3772)
  • Token Usage Buffering: Asynchronous buffered token usage recording with periodic disk flush (#3766)

🐛 Fixed

Channels

  • Filter Tool Messages: Respect filter_tool_messages setting for in-progress content events (#3660)
  • File URL Encoding: Percent-encode non-ASCII characters in file:// URLs to fix file send failures on Windows and paths with CJK characters (#3625)
  • WeChat Send Response: Fixed WeChat send_text failing on empty or non-dict API responses (#3685)
  • WeChat Icon Mapping: Fixed wechat channel key not mapping to the correct WeChat icon in chat history sidebar (#3691)
  • WeChat QR Login Timeout: Extended WeChat QR-status polling timeout to 60 seconds with retry on read timeout during login (#3700)

Console & UI

  • Agent Stats: Fixed page crash from incorrect chart library imports and stats not refreshing when switching agents (#3724, #3743)

Providers

  • Duplicate Model Warning: Show a localized warning when adding a model ID that already exists (#3773)
  • Anthropic File Block: Convert unsupported file blocks to text placeholders in Anthropic-formatted history to prevent API errors (#3786)

Packaging

  • Discord.py Conda-Unpack: Added discord.py to conda-unpack reinstall list to fix Windows path corruption (#3730)

📝 Documentation

  • Memory-Evolving & Proactive Agent: Added dedicated documentation pages for memory-evolving and proactive agent features (#3755, #3764)
  • Agent Team Practice Guide: Added a comprehensive practice guide for Agent Team collaboration (#3784)
  • ACP Integration: Restructured and polished ACP integration documentation, with in-UI documentation link (#3741)
  • Docker Backup Volume: Added Docker volume mount documentation for backup & restore (#3736)
  • Backup Figures: Updated backup documentation figures (#3678)

🧪 Testing & CI

  • Agent Unit Tests: Added unit tests for hooks, memory, and utils modules (#3630)

New Contributors

Full Changelog: v1.1.3...v1.1.4

Don't miss a new QwenPaw release

NewReleases is sending notifications on new releases.