Skip to content

ChromeClaw v2.0.0 Release Notes

Choose a tag to compare

View all tags
@algopian algopian released this 04 Apr 03:55
v2.0.0
3ce5ab4
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

New Web Providers

  • ChatGPT (chatgpt.com): Full web provider with Sentinel challenge support, model metadata extraction, thinking level control, and automatic 403 "unusual activity" recovery — detects stale sessions after inactivity, refreshes the tab, and retries seamlessly.
  • DeepSeek (chat.deepseek.com): Web provider with Proof-of-Work challenge solving for session authentication.
  • Doubao (www.doubao.com): Web provider for ByteDance's Doubao assistant.
  • Rakuten AI (rakuten.ai): Web provider with WebSocket streaming, HMAC request signing, and token caching.

Web Provider Architecture

  • Plugin system: Web providers are now registered via a plugin/factory pattern (plugin-registry.ts). Each provider is a self-contained plugin file, replacing the previous monolithic switch-case routing in the bridge. Adding a new provider is now a single file + registry entry.
  • Content script extraction: Provider-specific content-fetch handlers have been split out of the 3,300-line content-fetch-main.ts into dedicated files (content-fetch-chatgpt.ts, content-fetch-deepseek.ts, content-fetch-doubao.ts, content-fetch-gemini.ts, content-fetch-glm-intl.ts, content-fetch-rakuten.ts). The main file now contains only shared infrastructure.
  • Bridge improvements: web-llm-bridge.ts now supports providerMetadata for per-provider context and retryAttempt for retry loop prevention.

Existing Web Provider Improvements

  • GLM-CN: Added thinking mode toggle and more robust stream handling.
  • GLM International: Added thinking level support and model rename handling.
  • GLM adapters: Added error handling for stream adapter failures with timeout notices in the UI.
  • Qwen: Fixed thinking_summary phase transition bug.

Agent Backup & Restore

  • ZIP-based backup: Agents can now be exported as ZIP archives containing their full configuration (model, tools, custom tools, compaction config, identity) and all workspace files.
  • Restore in first-run setup: New "Restore" button in the onboarding wizard lets users bootstrap a fresh install from a previous agent backup.
  • Restore in settings: Backup and restore buttons added to the Agents settings panel.

Slash Commands

  • /new: Start a new chat session.
  • /copy and /export now support arguments (e.g., /copy 3 to copy the 3rd-latest assistant message).

UI Improvements

  • Tab indicator: An indigo outline appears on browser tabs while ChromeClaw tools are actively interacting with them, with a brief linger after completion. Reference-counted for overlapping tool calls.
  • Tool call copy button: Copy tool call results directly from the chat UI.
  • Improved /export output: Exports now include tool calls with parameters and results, reasoning blocks in collapsible details, and proper deduplication of tool results.
  • Markdown editor fix: Code blocks are now readable in light mode preview.
  • Removed unused description field from the model configuration form.

Internationalization

  • Localized all previously hardcoded English strings in the onboarding wizard and model configuration UI.
  • Added 32 missing i18n keys across all 10 locales (de, en, es, fr, ja, nl, pt, ru, zh_CN, zh_TW).
  • Translated all firstRun_* strings that were left untranslated.
Assets 4
Loading