We've shipped AI-powered session summarization and categorization, giving you instant qualitative insights into your Claude development conversations.
> impact
We have integrated a new AI-powered analysis layer directly into the session detail view of the Claude Code Inspector dashboard. When you open any session log, the system now automatically triggers a call to a Large Language Model (LLM) which analyzes the full conversation. This process generates a concise, one-sentence summary of the session's content and classifies its primary intent into categories such as 'Debugging', 'New Feature Generation', or 'Refactoring'.
Previously, the dashboard was limited to quantitative metrics like token counts, cost, and timestamps. While useful, this data lacked context. To understand the actual purpose or outcome of a coding session, developers had to manually read through the entire conversation log. This was a time-consuming process that made it difficult to spot higher-level patterns or quickly recall the context of past work, hindering efficient analysis of LLM usage.
This enhancement transforms the analytics dashboard by adding a powerful layer of qualitative context. Now, you can understand the 'what' and 'why' of any session at a glance, saving significant review time. It enables you to quickly identify trends in your development process—for example, discovering if you spend more time debugging existing code or generating new features—making the inspector a far more insightful tool for optimizing your workflow and interaction with Claude.
> Try this now
try this
# 1. Navigate to your Claude Code Inspector dashboard.
# This is your central hub for all session analytics.
# 2. Click on any existing session in your history to open the detail view.
# At the top of the view, you'll now see two new fields: 'AI Summary' and 'Category'.
# 3. As soon as you open the view, a request is sent to an LLM.
# The prompt is structured to extract a summary and classification:
# --- LLM PROMPT ---
# Given the following developer conversation, provide a one-sentence summary
# and classify its primary intent into one of the following categories:
# 'Debugging', 'New Feature Generation', 'Refactoring', 'Documentation', 'Exploration'.
# Conversation:
# {full_conversation_text}
#
# Respond with JSON: {"summary": "...", "category": "..."}
# --- END PROMPT ---
# 4. The returned summary and category are instantly populated in the UI.
# This gives you immediate context on past work without needing to read the log,
# allowing you to quickly scan and understand your development history.