PromptMask is a local first privacy layer for large language model users. It redacts sensitive data from prompts before they leave the user's machine and restores the original values in the response, so only anonymized text reaches a third party AI service. The masking itself is done by a trusted language model, typically a local one running behind an OpenAI compatible endpoint such as Ollama, llama.cpp or vLLM, though a remote model the user trusts can be configured instead.
It can be used in several ways. Python developers can swap in the OpenAIMasked client as a drop in replacement for the standard OpenAI client, or use the PromptMask class for finer control over masking and unmasking, including async versions and streaming responses that are unmasked as they arrive. Users who do not write Python can run a local web server that exposes an OpenAI compatible gateway endpoint, plus a web interface and web API, and point existing applications at it. Behavior is set through a TOML configuration file, and the project publishes benchmarks to help choose a local model.
The documentation site covers a Python API reference and links to an examples directory in the source repository. PromptMask is distributed as a Python package, with a Docker image, and is released under the MIT license. No paid plan is offered.
Related Tools
Other development & code tools in the same category
