Cost Tracking for Models

**Quick judgment** A focused, practical skill for teams that actively use OpenClaw and want concrete, log-derived cost reporting. It is strongest when used on Unix-like environments with stable access to OpenClaw session logs and jq installed. **What it does well** - Accurately reads `message.usage.cost.total` from OpenClaw JSONL session logs and aggregates by date and model. - Supports common reporting windows (today, yesterday, week, month, custom date range) plus comparisons (e.g., current week vs previous). - Produces Discord-ready formatted reports that are easy to drop into chat for team cost visibility. - CLI-based workflow lends itself to cron jobs or CI automation. **Key risks / limitations** - **Log format dependency:** Heavily tied to OpenClaw’s current log schema and path (`~/.openclaw/agents/*/sessions/*.jsonl`). Any structural changes (e.g., renaming `message.usage.cost.total`) can silently break or skew reports. - **Environment assumptions:** Requires jq and a shell environment; Windows users or non-standard setups may need extra configuration. - **Scalability limits:** On very large log directories, jq-based scans may become slower; no mention of caching or incremental indexing. - **Data exposure risk:** If reports are auto-posted to public or broad-access Discord channels, they may reveal internal modeling patterns, model names, and spending levels. **Recommended scenarios** - Small to medium teams running agents on shared infrastructure who want **simple, automated cost summaries** in Discord or other messaging channels. - Individual power users tracking **daily/weekly/monthly OpenClaw spend** and which models are driving costs. - Engineering / MLOps teams doing **cost governance** and watching for spikes by model or time period. - Internal reporting where **precision from raw logs** is preferred over approximate or manually maintained cost spreadsheets.