Skip to content

Reports and Visuals ​

TokenPilot provides built-in reporting and a visual inspector so you can see exactly what it's doing.

Host Entry Points ​

HostReport or statusVisual Inspector
OpenClawlightrsi openclaw reportlightrsi openclaw visual
Codexlightrsi codex reportlightrsi codex visual
Claude Codelightrsi claude-code reportlightrsi claude-code visual
DeepSeek Harness/tokenpilot-status inside a Harness sessionShared Visual interface does not apply
pilightrsi pi reportlightrsi pi visual
OpenCodelightrsi opencode reportlightrsi opencode visual

DeepSeek Harness reports estimator activity, scheduling, application, and deferrals through its native read-only status command. See status fields. The shared report metrics and Visual walkthrough below apply to OpenClaw, Codex, Claude Code, pi, and OpenCode.

The Report Command ​

The report command shows a summary of token usage, cache efficiency, and cost savings for the current session:

bash
# Standalone CLI
lightrsi report

# Per-host
lightrsi openclaw report
lightrsi codex report
lightrsi claude-code report
lightrsi pi report
lightrsi opencode report

# Inside OpenClaw
/lightrsi report

What Each Metric Means ​

MetricDescription
Input tokensTotal tokens sent to the model (lower is better)
Cache readTokens served from the model API cache (higher is better)
Cache missTokens that had to be recomputed (lower is better)
Output tokensTokens generated by the model
CostEstimated API cost in USD
Cache hit ratePercentage of input tokens served from cache

Visual Inspector ​

The visual inspector is a browser-based dashboard showing real-time TokenPilot behavior:

bash
lightrsi visual

This opens your default browser with three views: stabilizer, reduction, and eviction.

Context Cleaner Results ​

User-approved Cleaner results are tracked by plan receipt, separately from the general session report:

bash
lightrsi <host> clean --status <plan-id>

Use openclaw, codex, or claude-code for <host>. Distinguish estimated and scheduled savings from applied savings; character-based counts are not token counts or provider billing savings. See Context Cleaner. For DeepSeek Harness, query the native /tokenpilot-clean --status <plan-id> command instead; the shared CLI command above is not available.

Next ​

Released under the MIT License.