podifyr-ai 2026.7.7

Automated developer onboarding via podcast-style audio walkthroughs of Python codebases.

Podifyr-AI

AI-powered CLI that transforms Python codebases into true multi-speaker podcast walkthroughs using LangGraph agentic pipelines, AST analysis, dependency graph traversal, and neural text-to-speech synthesis

---

What is Podifyr-AI?

Podifyr-AI is a CLI tool that analyzes a Python repository's architecture and generates a conversational, podcast-style audio walkthrough. It's designed to accelerate developer onboarding by letting new team members listen to an AI-generated explanation of the system architecture — like having a senior engineer give them a KT session on day one.

How It Works?

Repository → AST Parsing → Dependency Graph → Host ↔ Expert Dialogue → Multi-Voice TTS → 🎧

1. Parse: Traverses the repo and extracts structural metadata (classes, functions, imports) using Python's AST module
2. Graph: Builds a directed dependency graph to understand module relationships and reading order
3. Script: Runs a multi-agent LangGraph pipeline — the Analyzer produces a technical summary, then the Dialogue Writer rewrites it as a Host/Expert conversation (JSON turns)
4. Audio: Synthesizes each turn with a distinct TTS voice (free Edge TTS by default), then FFmpeg-stitches the full episode

Prefer a single-narrator walkthrough? Pass --style monologue to switch back to the classic single-voice mode.

Quick Start

Installation

pip install podifyr-ai

Podifyr-AI is now more user friendly. Pick a provider (openai, azure, or ollama) and pass the relevant flags.

OpenAI

podifyr-ai generate ./my-project \
  --provider openai \
  --api-key sk-your-key-here

Azure OpenAI

podifyr-ai generate ./my-project \
  --provider azure \
  --api-key <your-azure-key> \
  --azure-endpoint https://your-resource.openai.azure.com \
  --azure-deployment gpt-4o-mini

Ollama (local, no API key)

Make sure Ollama is running locally (ollama serve) and the model is pulled (ollama pull llama3):

podifyr-ai generate ./my-project \
  --provider ollama \
  --model llama3

By default audio is synthesized with the free Edge TTS backend — no API key required.

Dialogue voices (default mode)

Dialogue mode uses two distinct voices. Defaults work for the OpenAI TTS backend; for Edge TTS override them with Microsoft Neural voice ids:

podifyr-ai generate ./my-project \
  --provider ollama --model llama3 \
  --host-voice en-US-AriaNeural \
  --expert-voice en-US-GuyNeural

To get a classic single-voice walkthrough instead:

podifyr-ai generate ./my-project --style monologue --voice nova

CLI Reference

podifyr-ai generate <REPO_PATH>     Generate a podcast walkthrough

Provider options:
  --provider, -p TEXT             LLM provider: openai (default), azure, ollama
  --model, -m TEXT                LLM model name (e.g. gpt-4o-mini, llama3)
  --api-key TEXT                  API key for the LLM provider (openai/azure)
  --azure-endpoint TEXT           Azure OpenAI endpoint URL
  --azure-deployment TEXT         Azure chat model deployment name
  --azure-api-version TEXT        Azure OpenAI API version
  --ollama-base-url TEXT          Ollama server URL [default: http://localhost:11434]

Output and audio options:
  --output, -o PATH               Output directory [default: ./podifyr_output]
  --style TEXT                    Podcast style: 'dialogue' (default, two speakers) or 'monologue'
  --tts-backend TEXT              TTS: 'edge' (free, default), 'openai', 'elevenlabs'
  --voice TEXT                    Voice for monologue style (e.g. alloy, echo, fable, onyx, nova)
  --host-voice TEXT               Voice id for the Host speaker (dialogue mode)
  --expert-voice TEXT             Voice id for the Expert speaker (dialogue mode)
  --tts-api-key TEXT              API key for the TTS backend (falls back to --api-key)
  --skip-audio                    Generate script only, skip audio
  --no-cache                      Disable caching for this run
  --concurrency, -c INT           Max concurrent TTS requests [1-20]
  --graph-details                 Show dependency graph metrics
  --verbose, -V                   Enable debug logging

podifyr-ai cache clear              Clear cached data
podifyr-ai cache stats              Show cache statistics
podifyr-ai --version                Show version

TTS Backends

Backend	Cost	API Key Required	Quality	Setup
Edge (default)	Free	No	Good (Microsoft Neural)	None
OpenAI	~$0.015/1K chars	Yes (--tts-api-key or --api-key)	Good	API key
ElevenLabs	Varies	Yes (--tts-api-key)	Excellent	pip install podifyr-ai[elevenlabs]

Configuration

Podifyr-AI is purely CLI-driven — there is no .env file, no PODIFYR_* env vars, no config sub-command. Every run is fully described by the flags you pass to podifyr-ai generate.

Features

• True two-speaker dialogue — Host ↔ Expert conversation with distinct TTS voices (default), or single-narrator monologue mode (--style monologue)
• AST-based analysis — Extracts architecture without executing code
• Dependency graphing — Cycle-aware topological sorting with NetworkX
• Multi-agent pipeline — LangGraph orchestration with Analyzer + (Scriptwriter │ Dialogue) nodes
• Unified LLM interface — One CLI, three providers: OpenAI, Azure OpenAI, Ollama
• Free TTS included — Edge TTS (Microsoft Neural voices) works out of the box
• Plugin backends — Swappable TTS providers (Edge, OpenAI, ElevenLabs)
• Smart caching — Content-hash invalidation avoids redundant API calls
• Rich CLI — Beautiful progress bars, graph metrics, and colored output
• Docker support — Reproducible builds with multi-stage Dockerfile
• Production-grade — Strict typing, structured logging, comprehensive test suite

Architecture

src/podifyr/
├── core/          # Exceptions, constants, protocols, types
├── config/        # Plain pydantic models for runtime settings (CLI-driven)
├── logging/       # Structured logging with structlog
├── utils/         # Filesystem helpers, retry logic, async patterns
├── cache/         # Disk-based caching with content-hash invalidation
├── parsing/       # AST visitor, models, filtering engine
├── graph/         # NetworkX graph builder and analyzer
├── llm/           # Unified provider factory (OpenAI, Azure, Ollama)
├── agents/        # LangGraph nodes, prompts, orchestrator
├── audio/         # TTS backends (Edge, OpenAI, ElevenLabs, Azure), stitcher
└── cli/           # Typer commands with rich display

Requirements

• Python 3.10+
• FFmpeg (for audio stitching)
• One of: an OpenAI API key, an Azure OpenAI deployment, or a local Ollama server
• TTS: Edge TTS works free with no key; OpenAI/ElevenLabs require API keys

Install FFmpeg

# macOS
brew install ffmpeg

# Ubuntu/Debian
sudo apt-get install ffmpeg

# Windows
winget install ffmpeg

Development

# Clone and install in dev mode
git clone https://github.com/anunayandkumar/podifyr-ai.git
cd podifyr-ai
pip install -e ".[dev,docs,all]"

# Run tests
pytest

# Run linter and type checker
ruff check src/ tests/
mypy src/

License

MIT — see LICENSE for details.