PDX is a framework for prompt engineering and a dev-ops toolkit.
At the core, it provides a mental-model on how to build and manage agents
. An agent is a collection of prompts and/or prompt templates with information that is used to interact with the Language Models.
Documentation: pdxlabs.io/docs
Website: pdxlabs.io
Quickstart: create-an-agent
pip install pdx
To create your first agent, run the following command:
pdx create my_first_agent
Run and test out the agent by running:
pdx test my_first_agent --verbose
More information here: PDX - Main Concepts
- ๐๏ธ Low dependency footprint -> ease of production deployment and maintainance.
- ๐ Mental model to separate prompt templates from the application code. (Similar to Flask blueprint or FastAPI router).
- ๐ Version control the prompts along with their evaluation metrics.
- ๐ธ Logging and tracing of inputs, prompt render, and model response made easy.
- ๐งฏ Standardize Error handling and logging.
- ๐พ Caching for lowering latency. (Coming soon)
- ๐ Observability out-of-the-box. (Coming soon)
- ๐ฉ Log feedback of the user. (Coming soon)
- ๐๏ธ A/B testing of prompts. (Coming soon)
Check our the demos in the demos repository.
- OpenAI
- Anthropic
- Cohere
- Eleven Labs