Local-only analysis — zero network requests
Agent Job Waste Audit
TokenSave audits recurring AI agent jobs for avoidable token burn. Drop your OpenClaw export — Import → Diagnose → Evidence → Manual Fix. All analysis stays local.
Sample Finding
D1
Duplicate Jobs
High confidence — Two active jobs run the same task 5 minutes apart
Evidence: same prompt summary, same tool calls, same schedule pattern.
Manual fix: confirm the canonical job, then deactivate the duplicate.
How to get your OpenClaw diagnostic files
BEST
Full local data — strongest audit readiness
~/.openclaw/cron/jobs.json
~/.openclaw/cron/runs/*.jsonl
Unlocks: jobs, schedules, models, tokens, errors, failure loops, cost exposure
OK
jobs.json only
Unlocks: job definitions, schedules, models, duplicate active-job review
Missing: tokens, errors, failure-loop analysis
PARTIAL
Run history (JSONL) only
Unlocks: tokens, errors, failure-loop / burst / zero-token review
Missing: job schedules, models, duplicate-job analysis
What To Do First
Start here — one job at a time
View full report Import summary · Waste findings · Job table
Active Recurring Waste Findings
Ranked by recurring token waste — tokens/day first, then tokens/run, then fallback tokens × error rate
Approx. Cost by Job (Secondary)
Tap a column header to sort