langchain-interpreter

AFM LangChain Interpreter

A LangChain-based reference implementation of an interpreter for Agent-Flavored Markdown (AFM) files.

Features

• Support for all interface types:
  • Console chat (interactive CLI)
  • Web chat (HTTP API + optional UI)
  • Webhook (WebSub-based event handling)
• Multi-interface agents - run multiple interfaces simultaneously
• MCP support for tools (Model Context Protocol)
• Validation - dry-run mode to validate AFM definitions

Prerequisites

• Python 3.12 or later.
• uv for dependency management.

Installation

Using pipx (Recommended)

To run afm as a standalone command-line tool:

pipx install afm-cli

Using pip

pip install afm-cli

Quick Start

# Set your API Key
export OPENAI_API_KEY="your-api-key-here"

# Run with an AFM file
afm path/to/agent.afm.md

Configuration

Configuration via environment variables or CLI options:

• OPENAI_API_KEY, ANTHROPIC_API_KEY, etc. (Required based on provider)
• HTTP port can be set via -p or --port (default: 8000)

Running with Docker

# Build the image
docker build -t afm-langchain-interpreter .

# Run with an AFM file mounted and API key
docker run -v $(pwd)/path/to/agent.afm.md:/app/agent.afm.md \
  -e OPENAI_API_KEY=$OPENAI_API_KEY \
  -p 8000:8000 \
  afm-langchain-interpreter afm /app/agent.afm.md

Testing

uv run pytest

Project Structure

langchain-interpreter/
├── src/afm_cli/
│   ├── interfaces/        # Interface implementations (console, web, webhook)
│   ├── tools/             # Tool support (MCP server)
│   ├── resources/         # Static assets (web UI)
│   ├── agent.py           # Core agent logic
│   ├── cli.py             # CLI entry point
│   ├── parser.py          # AFM file parsing
│   ├── models.py          # Model configuration
│   ├── providers.py       # LLM provider handling
│   └── templates.py       # Prompt templates
├── tests/                 # Unit and integration tests
├── afm-samples/           # Example AFM definitions
├── Dockerfile             # Container build
├── pyproject.toml         # Python project configuration
└── uv.lock                # Dependency lock file

License

Apache-2.0