Agentbox v0.3.0 Released - Qwen Support, Multi-Agent Orchestration & More

[scharc]

I'm excited to announce Agentbox v0.3.0!

This release adds a fourth AI agent, multi-agent conversations, and a bunch of reliability improvements.

Highlights

Qwen Code Support
Alibaba's Qwen is now a first-class citizen alongside Claude, Codex, and Gemini:

abox qwen          # Start Qwen agent
abox superqwen     # Auto-approve mode

Multi-Agent Orchestration with AgentPipe
Run conversations between multiple AI agents for code review, brainstorming, or getting different perspectives:

agentpipe run -a claude:Reviewer -a gemini:Architect -p "Review this PR"

Agents can also invoke this via the discuss MCP tool.

Agent Rate Limit Tracking
Automatic fallback when an agent hits rate limits:

• superclaude → supercodex → supergemini → superqwen
• Check status: abox usage status

Remote Q&A via Telegram
Get notified on your phone when agents need input. Answer questions remotely without being at your desk.

Port Conflict Detection
Running multiple containers? Agentbox now detects port conflicts across all your projects.

Other Improvements

• Unified MCP config across all agents (single mcp.json)
• Native Claude Code installer (faster, more reliable)
• GitLab CLI (glab) and FastMCP added to base image
• Faster session discovery with daemon caching
• Consistent CLI patterns and better error messages

Breaking Changes

• Worktree command arguments changed to BRANCH AGENT order
• Removed wt alias (use agentctl worktree)
• Removed deprecated port aliases

Upgrading

# Pull latest and rebuild
git pull
abox rebuild
abox rebase  # Updates running containers

Full Changelog

See CHANGELOG.md for the complete list of changes.