LLM logger is a logger & visualisation tool for LLM-based systems and processes such as Chain-of-though, Tree-of-thought, Multi-Agent Systems, Retrieval Augmented Generation (RAG), etc.
LLM systems (especially agentic systems) can create hard-to-track conversations with hundreds or throusands of entries. If the messages are logged into a text file, it might not be possible to detect deviations in the conversation flow. LLM-logger helps to visualize the conversation flow (e.g. among agents, or within a RAG system) as a scrollable-graph that naturally displays the conversation in chronological order.
LLM-logger is split into 2 tools/sub-packages:
logger tracks messages passed within the system with one or more LLMs and preserves them within a graph-like structure. This graph-structure keeps the sender-receiver information and the content of the passed message in chronological order.
visualizer interprets the graph and interactively displays the logged data in a browser.
The user interface is divided into 2 parts, left-side and right-side:
Visualisation tool is fully responsive (adjusts to vertical format on narrow screens).
And includes light/dark mode for comfortable reading.
