XiaoMi MiMo Code v0.1.4 Overhauls Workflow UI, Adds Compose Pipeline

XiaoMi’s MiMo Code framework just shipped version 0.1.4 with a completely redesigned workflow interface and a new deterministic compose pipeline. The GitHub release published June 29th focuses on making multi-agent workflows more observable and adding structured automation for common development tasks.

The biggest change is a full-screen workflow monitoring system that replaces the previous cramped inline panels. Users can now drill down into individual agent conversations while keeping track of overall progress through live status counters.

Full-Screen Workflow Monitoring

The workflow interface now expands to full-screen pages where each agent appears as a message-style card with real-time status indicators. The old inline panel stays but caps at 12 lines to prevent chat flooding, while the full-screen view gives complete visibility into what every agent is doing.

Each workflow run gets its own observability tree that tracks phases, agents, and nested workflows with their parameters and results. Users can drill down into any subagent’s complete conversation history and navigate between different workflow levels while preserving scroll positions.

Deterministic Compose Pipeline

A new built-in compose workflow encodes the standard development cycle as executable code: Brainstorm → Design → Implement → Verify → Review → Merge. Unlike the existing conversational compose agent, this workflow runs non-interactively from start to finish.

The system automatically parallelizes independent tasks into separate Git worktrees and chains structured outputs between phases. Brainstorm results feed into implementation intent, which flows to review verdicts and finally merge decisions. The workflow enforces retry limits and handles the entire pipeline without human intervention.

The team positions this as complementary to the existing compose agent rather than a replacement. The workflow suits well-defined tasks that decompose cleanly into independent subtasks, while the conversational agent remains better for exploratory work where you need to redirect mid-flow or inject judgment between steps.

Provider Support and Windows Fixes

This release adds dedicated system prompts for three Chinese AI providers: DeepSeek, GLM, and MiniMax. The prompts are automatically selected based on the model API identifier, suggesting MiMo Code is expanding beyond its initial OpenAI focus.

Windows users get a fix for CJK character encoding issues that were causing garbled output in shell operations. The system now forces UTF-8 encoding for subprocess operations on non-UTF-8 locales like Chinese GBK/936 code pages, covering both the bash tool and TUI shell mode.

Additional Updates

The release includes several reliability improvements: better detection and recovery from malformed tool calls, keep-alive heartbeats to prevent client timeouts during long operations, and UTF-8 defaults for new installations. The security model also gets an adjustment, removing SSRF protection from user-configured MCP server connections while keeping it for AI-initiated web requests.

Bottom Line

Version 0.1.4 shows MiMo Code maturing from a prototype into a production-ready agent framework. The workflow UI overhaul addresses a real pain point in multi-agent debugging, while the deterministic compose pipeline offers a fire-and-forget option for routine development tasks. The Chinese provider support and Windows encoding fixes suggest the project is building toward broader international adoption beyond its initial developer base.