Agentic system for combustion CFD simulations with DeepFlame + OpenFOAM, delivered through a focused, tool-aware CLI.

Inspired by opencode — this project adapts the opencode architecture to combustion CFD workflows.

FlamePilot (DeepFlame Agentic Runner) is a Python-based, schema-compliant, tool-first agent designed to automate and assist CFD workflows. FlamePilot integrates with Google ADK, streams responses in real time, and provides a clean CLI experience tailored to simulation setup, diagnostics, and iteration.

Features

• Agentic architecture with direct access to CFD tools and file utilities
• Interactive CLI with session management and streaming responses
• Google ADK integration with schema-compliant tool validation
• Modular core/config/types packages and dynamic agent discovery
• Comprehensive test suite across unit, integration, and end-to-end levels

Contents

• Quick Start
• Configuration
• Usage
• Documentation
• Architecture
• Development
• Contributing
• License
• Support

Quick Start

Requirements

• Python 3.8+
• OpenFOAM installation (bin paths available)
• DeepFlame installation (bin paths available)

Install

pip install -e .

Run

df-agent

Configuration

Update config.yaml to point to your local OpenFOAM and DeepFlame installs. A sample is available at config.yaml.example.

model:
  provider: deepseek
  name: deepseek-chat

openfoam:
  bins:
    - /opt/openfoam7/bin
  bashrc_path: /opt/openfoam7/etc/bashrc

deepflame:
  bins:
    - /path/to/deepflame/bin
  bashrc_path: /path/to/deepflame/bashrc

Usage

CLI Commands

• /help: Show available commands
• /use <agent>: Switch agents
• /session <name>: Switch sessions
• exit or quit: Exit the CLI

Default Agent

• deepflame_agent: main coordination agent for CFD workflows

Documentation

• docs/index.md — full documentation portal and navigation
• docs/simulation_setup/installation.md — installation guide
• docs/simulation_setup/configuration_reference.md — configuration reference
• docs/tools_and_utilities/solvers_overview.md — solver overview
• docs/tools_and_utilities/troubleshooting.md — troubleshooting
• AGENTS.md — agent/tool development guidelines

Architecture

src/
├── agents/      # Agent definitions and tool registry
├── cli/         # CLI, session management, and streaming
├── core/        # Messaging, providers, sessions, tools
├── config/      # Configuration loading
└── types/       # Shared type definitions

Development

Tests

pytest

Code Style

• PEP 8 formatting, 4-space indentation, 88 char line length
• Type hints for all functions, Google-style docstrings

Contributing

Open an issue describing the change, then submit a PR. See AGENTS.md for tool, schema, and style guidelines.

License

License information will be added once finalized.

Support

Start with docs/index.md. For questions or bug reports, open an issue in the repository.