MiroFlow
MiroFlow is the framework part of the MiroMind Research Agent project, designed for multi-step internet research such as future event prediction. The README describes it as modular, with multi-turn conversation, a tool ecosystem and hierarchical sub-agent orchestration. The wider project also includes MiroThinker, an open-source agent model with tool-assisted reasoning, and MiroVerse, a 147k-sample training dataset.
The README reports results from 2025: v0.3 ranked first on the FutureX leaderboard as of 2025-09-10, and a comparison table lists GAIA validation at 82.4%, HLE 27.2%, BrowserComp-EN 33.2%, BrowserComp-ZH 47.1% and xBench-DeepSearch 72.0%, set against OpenAI Deep Research, Gemini Deep Research, Kimi Researcher and others. These are the authors' own figures from over a year ago, so treat them as a snapshot. It lists GPT, Claude, Gemini and Qwen as supported models and MCP-server tools for audio transcription, Python and file reading.
Quick start needs Python 3.12+ and `uv` on Linux or macOS: clone, `uv sync`, put an OpenRouter API key in `.env`, then run `uv run main.py trace` with a config and a task. The README says the MiroThinker model can run a research service on a single RTX 4090. License is Apache 2.0 and the last push was 2026-07-06. This review is based on the README and repository data; we have not run it ourselves.
Features
Why choose it
Trade-offs
Maintenance check
- Last commit
- 2026-01-30
- Commits, last 90 days
- 0
- Releases, last 90 days
- 0
- Latest release
- —
- Open issues
- 2
- Recent issues with at least one comment
- 13/16
Oldest week on the left; the latest four weeks are orange. Source: GitHub commit activity.
Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.
Compatibility
Quick start
Use cases
Alternatives
Comments
No comments yet. Be the first!