Fluxyr

A Python framework for agents that expand their own capabilities and repair their skills through a self-healing workflow. Describe what you need in the chat: Fluxyr can research an integration, generate Python actions, test them and save them as reusable skills.

Create an application with a Flask-style API; Fluxyr handles tools, memory, conversations, streaming, background executions and scheduled routines.

Install

pip install fluxyr

Requires Python 3.11+. The package includes the web UI; Node is only needed when contributing to the frontend.

Your first application

Create a .env file:

FLUXYR_MODEL=minimax/minimax-m3
OPENROUTER_API_KEY=your-key

Create app.py:

from fluxyr import Fluxyr

app = Fluxyr(__name__)

@app.tool()
def greet(name: str) -> dict:
    """Greet someone by name."""
    return {"message": f"Hello, {name}!"}

@app.get("/api/hello")
def hello():
    return greet("Fluxyr")

if __name__ == "__main__":
    app.run()

Run python app.py and open http://localhost:5050. The agent is ready for you to configure its skills, credentials and routines and start a conversation.

Build skills through conversation

Ask Fluxyr to create an integration in the chat, for example:

Create a weather skill for my provider's API, with a forecast action accepting a city. Ask me for the documentation URL and help me configure any credentials in Vault. Test it for São Paulo, then create a routine that retrieves the forecast every day at 08:00 in America/Sao_Paulo.

Fluxyr turns the request into skill instructions and a technical specification. A dedicated builder generates the Python actions and declares their dependencies and Vault credentials. The agent inspects the generated code, runs tests, checks the results and activates a tested version for future conversations and routines. You can follow the build and execution progress in the UI.

The same loop supports self-healing: when an action fails, the agent can inspect the execution, diagnose the problem, revise the specification and rebuild the affected action. It tests the replacement before activation and can improve the skill's instructions from what it learns. Each code change creates a new version, preserving the previous versions and execution history. Missing credentials or required user choices are handled through embedded dialogs.

Configuration

Fluxyr reads .env from the directory where you start the application. Environment variables take precedence. Only the model and its provider's API key are required.

With another provider, supply its key: OPENAI_API_KEY, ANTHROPIC_API_KEY or CUSTOM_API_KEY. See the configuration reference for provider endpoints and other options.

All local instance storage lives under .fluxyr/. Set FLUXYR_ROOT to mount it as one volume. Existing installations use the explicit storage migration.

Skills and integrations

Point FLUXYR_SKILLS_DIR at a folder to load its .md files as agent instructions. Use @app.tool() to expose Python functions and standard Flask routes for your API. The agent can also build and test Python actions, manage credentials in Vault, request human input and schedule routines.

• Minimal application
• Framework API and configuration
• Custom tools (@app.tool) and human-in-the-loop workflows
• Native tools reference
• Headless browser, private input, OTP and passkeys
• Reactive routines and webhooks
• Continuous worker routines
• Python actions and credentials
• Human interaction
• Approval API and optional Bearer key
• Docker and execution protection

Development

git clone https://github.com/shpyrd-io/fluxyr.git
cd fluxyr
pip install -e '.[dev]'
pnpm --dir frontend install --frozen-lockfile
pnpm --dir frontend build
fluxyr --reload

See CONTRIBUTING.md for local development and test suites, and the release guide for versioning and publication.

Licensed under MPL 2.0. Third-party notices are in NOTICE.