Claude Code session inspector
The Inspector is a panel inside Canopy that shows what every Claude Code session is doing: status, cost, token usage, context window, and tool calls. It works with both Claude Code and Gemini CLI.
Claude Code cost tracking and token usage
Each Claude Code session tracks input tokens, output tokens, cache reads, and a running cost for the active model. The Inspector sums Claude Code token usage per session and per worktree, so you can see which branch burned the most context today.
- Input and output token counters per session
- Cache read and write counts where the model reports them
- Running cost based on the active model's published rate
context window visibility
A bar shows how much of the context window a session has consumed. When the window starts filling up, you can decide to compact, restart, or keep going before the session hits its limit.
- Percent of context window used
- Model name and its context length
- Warning when the window approaches the limit
tool call history
Every tool call the session makes, from file edits to bash commands, is logged in a scrollable list. You can open a call to see its input and output, which helps when a session made a change you did not expect.
- Chronological list of tool calls
- Input and output for each call
- Filter by tool name
questions about the inspector
- What does the Canopy Inspector show?
- Session status, cost, input and output tokens, context window usage, and the full tool call history for Claude Code and Gemini CLI sessions running inside Canopy.
- Can I track costs per session?
- Yes. Each session has its own counters, and Canopy also aggregates costs per worktree so you can compare branches.
- Does the Inspector work with existing Claude Code installs?
- Canopy launches Claude Code for you and reads its session output, so you do not configure anything extra. You still use your own Claude Code authentication.