Privacy-Preserving Dashboard

AI Token Burn Dashboard

A privacy-preserving view of human-directed AI usage: model routes, daily token totals, cost estimates, and aggregate operating signals. It deliberately excludes prompts, responses, session names, file paths, private work details, heartbeat polls, cron jobs, and subagent bootstrap traffic.

Source: human-directed OpenClaw usage logsTimezone: America/New_YorkUpdated: 9/1/2026, 6:20:51 AM

Total tokens

1.5B

13,136 human-directed calls

Active days

77

18.9M avg tokens/day

Model mix

3

distinct model routes

Estimated cost

$0.881

from local usage metadata

Daily Human-Directed Token Burn

GitHub-style activity map. Click a day to inspect model-level usage.

May 31, 2026 - Sep 1, 2026
LessMore

Jul 29, 2026

57.5M tokens

Input2,068,255
Output83,006
Cache read55,336,192
Cache write0

Calls

455

Cost

$0.00

Model Breakdown

openai/gpt-5.5
455 calls ยท $0.00
57.5M

All-Time Model Mix

openai/gpt-5.5100%
nvidia/nemotron-3-ultra-550b-a55b:free0%
anthropic/claude-sonnet-4.60%

Operational Signals

Output share

0%

Cache share

95%

Peak day

Jul 29, 2026

Top route

openai/gpt-5.5

This is intentionally a usage telemetry view, not a conversation archive. It shows where human-directed compute is going without exposing the work itself.

Accounting Scope

Counted records

13,136

Excluded records

1,273

Timestamp basis

OpenClaw local record timestamp

Human-directed turns only.

Privacy Boundary

The published data file contains only daily aggregate human-directed usage by model: input tokens, output tokens, cache tokens, call counts, and cost estimates when available. Automated heartbeat polls, cron jobs, and subagent bootstrap sessions are excluded. Raw conversations and local session identifiers stay on the OpenClaw workstation.