v0.1.25

🚀 Release Highlights — v0.1.25 (2025-12-15)

🤖 UI-TARS Custom Tool

• Vision-Language Model: Custom tool using ByteDance's UI-TARS-1.5-7B for screenshot analysis and action generation via HuggingFace Inference Endpoints
• Flexible Deployment: Browser automation (Playwright) or Docker-based desktop control

🧠 GPT-5.2 Model Support

• New Default Model: OpenAI's GPT-5.2 now default, seamless upgrade from gpt-5.1 with no configuration changes required

📝 Evolving Skill Creator System

• Reusable Workflows: Create SKILL.md workflow plans and scripts/ folder that persist across sessions
• Session Persistence: Load skills from previous sessions via load_previous_session_skills: true

🎨 Textual Terminal Enhancement

• Adaptive UI: Terminal adjusts layouts and panels for different window sizes and agent coordination
• Dark/Light Themes: VS Code-inspired color schemes

📖 Getting Started

• Quick Start Guide: Try the new features today
• Try These Examples:
  • ui_tars_browser_example.yaml - UI-TARS browser automation
  • ui_tars_docker_example.yaml - UI-TARS Docker automation
  • single_agent_textual.yaml - Textual terminal display
  • skills_with_previous_sessions.yaml - Evolving skills with session persistence

What's Changed

• docs: update slides post v0.1.24 by @sonichi in #613
• fix: preserve OpenRouter Gemini thought signatures by @praneeth999 in #625
• feat: Add evolving skill for tasks in code mode by @ncrispino in #629
• feat: Add gpt-5.2 and clarify code exec by @ncrispino in #632
• feat: Backend for ByteDance seed UI-TARS by @franklinnwren in #584
• fix: improve terminal adaptivity and UI layout Improve textual terminal display by @praneeth999 in #589
• docs: Docs for v0.1.25 by @Henry-811 in #637
• feat: v0.1.25 by @Henry-811 in #635

Full Changelog: v0.1.24...v0.1.25