ModelMRI is an MIT-licensed developer tool for inspecting local language models, vision-language models, robot policies, and agents during execution. It exposes model attention, internal features, and agent steps to help AI researchers and developers debug and understand model behavior.
ModelMRI sits in PulseGate's LLM & agent tracing category. It focuses on understanding and debugging the internal behavior of local AI models, vision-language models, and robot policies while they run. It is built as an open-source project for AI researchers and developers working with local models and agents. The project is open source (Open Source). It ships for the command line, and it can be self-hosted.
Muhammad Mahad Azher builds and maintains ModelMRI. Among its 7 catalogued features are attention inspection, feature visualization, and agent step tracing.
Summary written by a language model from the project’s public pages.
What PulseGate has recorded for this listing
Closest matches by what these projects do