Skip to content

v0.1.0

Latest
Latest

Choose a tag to compare

View all tags
@xieyxclack xieyxclack released this 19 Mar 14:38
· 1 commit to main since this release
v0.1.0
b9380fe
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

✨ Added

Architecture

  • Multi-Agent / Multi-Workspace Architecture: Support running multiple agents simultaneously, each with its own isolated workspace containing independent config, memory, skills, and tools. Includes a console agent selector for easy switching between agents (#1375, #1625, #1670, #1714, #1664, #1614)
  • Context Management: Added token counting, /dump_history and /load_history commands, configurable history length limits, and progress indicators during memory compaction (#1628, #1754, #1689)

Security

  • Skill Security Scanner: Skills are now scanned for security risks (prompt injection, command injection, hardcoded secrets, data exfiltration) before installation using static analysis (#564)
  • Destructive Shell Command Detection: Added security rules to detect dangerous shell commands such as disk formatting, fork bombs, reverse shells, and privilege escalation attempts (#1484, #1529)
  • Web Authentication: Added optional web authentication with single-user registration, token-based login, localhost bypass, and CLI command (#1329, #1666, #1719)

Channels

  • WeCom Channel: Added WeCom as a messaging channel with media support, QR code access, console configuration UI (#1407, #1706, #1725, #1728, #1681, #1747, #1766, #1618, #1631, #1860, #1856)
  • XiaoYi Channel: Added XiaoYi as a messaging channel (#1213)
  • DingTalk AI Card Reply: DingTalk now supports AI Card-based replies with incremental streaming, falling back to webhook/markdown when unavailable (#1118)

Skills & Tools

  • LobeHub Skill Import: Skills can now be imported directly from LobeHub (#1350)
  • ModelScope Skill Hub: Skills can now be imported from ModelScope Skill Hub (#1673)
  • Version-Aware Builtin Skill Sync: Built-in skills now track versions and automatically sync updates while preserving user customizations (#1674, #1716, #1749)
  • Guidance Skill: Added a built-in skill that answers CoPaw installation and configuration questions using local docs and the official website (#1522)
  • Zip Archive Skill Import: Skills can now be imported by uploading zip archives directly from the console (#1840)
  • Built-in Tools: Added glob_search and grep_search as built-in tools for file discovery and content search within the agent workspace (#1852)
  • MCP Header Environment Variables: Support MCP HTTP client headers expanding ${ENV_VAR} placeholders (#1629)

Multimodal

  • Console Multimodal Chat: The console chat now supports sending images and files alongside text messages (#1772)
  • View Image Tool: Added a view_image tool that lets the LLM analyze local images for multimodal conversations (#1526)
  • Non-Multimodal LLM Media Fallback: When a non-multimodal LLM receives image/audio/video content, the system now automatically strips media blocks and retries the request (#1676)
  • Audio Transcription: Added voice message transcription via Whisper API or local Whisper model, with audio format conversion and a Voice Transcription settings page in the console (#1476, #1726)

Providers

  • Gemini Provider: Added Google Gemini as a built-in LLM provider (#1507)
  • DeepSeek Provider: Added DeepSeek as a built-in LLM provider (#1498)
  • MiniMax Provider: Added MiniMax as a built-in LLM provider with separate International and China endpoints (#1376, #1735)
  • Kimi Provider: Added Kimi as a built-in LLM provider with separate China and International endpoints (#1832)

CLI & Deployment

  • copaw update Command: Added copaw update and copaw shutdown CLI commands. copaw update automatically detects the environment and upgrades CoPaw from PyPI (#1278)
  • Docker Compose: Docker deployment now uses docker-compose.yml for easier setup and volume management (#1320)
  • Docker Image: Included additional channel dependencies in the Docker image (#1761)

Console & UI

  • Console Dark Mode: Added full dark mode support to the console with a theme toggle (light / dark / follow system) covering all pages (#1566, #1637, #1662)
  • Console Chat Streaming: Agent responses in the console are now streamed via SSE, with support for reconnection and stopping mid-response (#1571, #1672)

CLI, Deployment & Others

  • Timezone Configuration: Added a timezone selector in the console; timezone is used across system prompts, cron scheduling, and heartbeat. Supports auto-detection on multiple platforms (#1535, #1746)
  • OS Information in System Prompt: The system prompt now includes OS information for more platform-aware responses (#1660)

🔄 Changed

Core & Lifecycle

  • Graceful Lifecycle Management: Desktop subprocess, agent reload, and shutdown are now handled gracefully — subprocesses terminate cleanly on close, agents reload with zero downtime, and active tasks are awaited before shutdown (#1646, #1664, #1714)
  • Custom Working Directory: Replaced hardcoded ~/.copaw paths throughout the codebase so custom working directories (COPAW_WORKING_DIR) work correctly (#1652)
  • Dynamic Token Counting: Token counting now uses agent-specific configuration instead of a hardcoded global tokenizer, enabling per-agent token counting settings (#1819)
  • Config Loading Protection: Config loading is now wrapped with error handling so path normalization or heartbeat config failures do not crash the application (#1865)

Providers & Models

  • Provider / Model Consistency: The console now re-fetches active models on navigation so the displayed provider and model stay in sync (#1420)
  • Ollama Default Address: Changed Ollama default address from localhost to 127.0.0.1 for more reliable connections (#1480)
  • Provider Request Headers: Improved provider-specific HTTP header handling with exact URL matching for DashScope (#1757)

Console, UI & Platform

  • Console Internationalization: Improved multi-language support in the console, including synced locale settings, localized document navigation, and localized chat prompts / greeting text (#1409, #1686, #1707, #1858)
  • Workspace Path: Workspace path is now read-only when editing an existing agent to prevent accidental changes (#1624, #1764)
  • Windows Desktop Startup: Pre-compiled Python bytecode for faster Windows desktop startup (#1639)
  • QQ Channel Reply Logic: Improved reply logic with DM support and better handling of blocked URLs (#1650)

🐛 Fixed

Channels

  • Telegram: Fixed thread replies, media handling, and error reporting. Polling now auto-reconnects with exponential backoff on network failure (#1210, #1475)
  • Discord: Fixed messages from different channels being incorrectly merged; normalized Discord IDs for cron dispatch. The debounce fix was also generalized to all channels (#300, #1002, #1583)
  • DingTalk: Fixed incorrect message skipping when sender info is partially missing; skipped empty text blocks in rich text parsing (#851, #1554)
  • Feishu: Fixed card tables exceeding the per-card element limit; added voice message support; fixed channel reload failure caused by WebSocket event loop conflicts (#1627, #1726, #1817)

Providers & Models

  • Ollama / LM Studio: Fixed context length settings not taking effect; connection and model-fetch errors now show the actual error message (#1427, #1745)
  • Streaming Response: Fixed a crash when streaming response chunks have no choices (#1524)
  • Reasoning Content: Fixed reasoning content being lost when the message formatter changes the assistant message count (#1557)
  • Tool Choice: Fixed tool_choice being incorrectly forced to auto; it is now passed through as-is (#1570)

Skills

  • Cron Jobs: Fixed weekday name mismatch causing cron jobs to fire on wrong days; added UTC time context to prevent timezone drift; custom cron expressions now survive editing in the UI; invalid cron jobs are gracefully skipped; agents can now create cron jobs in the correct workspace (#1269, #1432, #1257, #1734, #1768, #1839)
  • Skills: Fixed import failures when skill names contain /; removed duplicate built-in skill listings; fixed incomplete file imports from bundles; restored skill enable/disable after sync refactor; added cancel option and timeout reminder for slow imports; skill descriptions now display correctly in the console; added user-friendly error messages with security scan details when import fails (#1369, #1396, #1576, #1716, #1720, #1626, #1862)

Console, UI & Platform

  • Memory Compaction: Fixed a crash when the system prompt is empty during memory compaction (#1608)
  • Console Chat: Fixed unwanted page redirect on refresh; copy now extracts text content instead of raw JSON; session list refreshes correctly when switching agents; chat sessions now persist across page navigation (#1373, #1471, #1662, #1722)
  • Console UI: Fixed static resource loading failures, workspace list refresh, workspace path display, version notification filtering, and miscellaneous TypeScript errors (#1402, #1737, #1764, #1769, #1822)
  • Windows Compatibility: Fixed cross-disk file moves, suppressed spurious AutoRun stderr in shell commands, and fixed emoji/Unicode logging on GBK consoles (#1483, #1556, #1601, #1495)

📚 Documentation

  • Ollama Context Length: Added a warning about Ollama and LM Studio context length configuration (#1433)
  • Chinese Docs: Fixed formatting and consistency issues across Chinese documentation (#1300)
  • Docker QuickStart: Aligned Docker secrets documentation on the QuickStart page (#1584)
  • Cron Skill: Updated --channel options in cron skill docs to list all supported channels (#1541)
  • Channel Allowlist: Added documentation for channel allowlist configuration, mentions, and ID hints (#1760)
  • Developer Community: Added developer community group information (#1678)
  • Console: Updated the release note link in the console UI (#1622)
  • WeCom SDK: Added instructions for using the official WeCom AI Bot SDK for channel setup (#1843)

New Contributors

Full Changelog: v0.0.7...v0.1.0

Contributors

mvanhorn, Justin-lu, and 23 other contributors
Assets 2
Loading