ImFeH2/flowent

Flowent is a serious full-stack attempt at a local workflow orchestration platform for multi-agent collaboration. The architecture is much more complete than the short README suggests: a FastAPI backend, React/Vite frontend, CLI entry point, Docker support, npm and PyPI packaging, provider configuration, MCP support, permission flows, workflow routes, state persistence, sandbox/tool execution modules, and streaming agent responses. The codebase also shows strong engineering discipline. The root package.json has clear scripts for dev, build, lint/type checks, tests, packaging, and release; the backend pyproject.toml uses modern Python packaging with uv, Ruff, mypy, pytest, and Python 3.11+; and the repo includes both frontend and backend tests plus regression tests for conversation lifecycle, workflow saving, MCP shutdown, context windows, permissions, patching, usage, logging, and startup requirements. I particularly liked the FLOWENT_AGENT_SYSTEM_PROMPT and streaming agent loop in backend/src/flowent/agent.py, because it shows the project is thinking carefully about how agents should continue after tool results rather than treating tool calls as final answers. The biggest improvement area is user-facing onboarding. For a tool this broad, the README is quite short; screenshots, an architecture diagram, provider setup examples, a first workflow walkthrough, and a clearer explanation of the permission/sandbox model would help users trust and understand it. Since it requires Bubblewrap and ripgrep, platform limitations should also be stated more explicitly. Overall, Flowent feels like an alpha-stage but well-engineered developer platform with real testing discipline and a clear direction.