CostClaw — Token Usage Optimizer

One command. Zero setup. Real dollar amounts.

Most OpenClaw users overpay by 50-90%. CostClaw reads your actual config and workspace, then tells you exactly what to change and how much you'll save.

Why CostClaw vs Alternatives?

| Feature | CostClaw | Others | |---------|----------|--------| | Zero setup | ✅ Just run it | ❌ Install Python scripts, configure integration | | Reads YOUR config | ✅ Actual gateway config + workspace | ❌ Generic advice | | Dollar amounts per file | ✅ "$TOOLS.md costs $4.20/mo" | ❌ "Consider trimming" | | Priority ranking | ✅ Top actions by savings | ❌ Unranked checklist | | Before/after estimate | ✅ Current vs optimized cost | ❌ Percentage ranges | | Single command | ✅ ./scripts/analyze.sh | ❌ 4+ scripts to learn |

Usage

As an Agent (Recommended)

When the user asks about costs, the agent should:

Run the analyzer script:
```
bash scripts/analyze.sh [workspace_path]
```
Default workspace: current directory. Pass explicit path if needed.

For JSON output (machine-readable):

bash scripts/analyze.sh [workspace_path] json

Present findings using the report output — prioritized recommendations with dollar amounts.

Trigger Phrases

Activate this skill when users say anything like:

"How much am I spending on tokens?"
"Why is my API bill so high?"
"Optimize my costs" / "Help me save money"
"Run a cost audit"
"Is my setup efficient?"

What the Analyzer Checks

Workspace files — Size of every .md file injected per turn, with per-file monthly cost
Installed skills — Count and estimated token overhead from skill descriptions
Model pricing — Current cost based on detected/configured default model
Heartbeat impact — Cost of heartbeat polling at current interval
Context injection total — Sum of all per-turn token overhead

Reading the Report

The report has 5 sections:

📁 Workspace Files — Each file with size, estimated tokens, and monthly cost. Flagged: ⚠ LARGE (>10K tokens), ⚡ MEDIUM (>5K tokens), ✓ OK.
🔧 Skills — Count of installed skills and their aggregate token cost.
⚙️ Model & Config — Detected default model and its pricing tier.
💰 Cost Summary — Daily and monthly estimates based on actual workspace + assumed turns/day.
📋 Ranked Recommendations — Prioritized by monthly savings. Each recommendation includes:
- What to change
- Why it saves money
- Estimated monthly savings in dollars
- How to implement it

Agent Actions After Analysis

Based on findings, the agent should:

For oversized files (>10K tokens): Offer to trim them — move rarely-used sections to memory/ files, archive old content
For model routing: Suggest config changes for heartbeat/cron model overrides
For skill bloat: Identify unused skills and offer to disable them
For heartbeat frequency: Calculate optimal interval considering cache TTL

Pricing Reference (March 2026)

The analyzer uses these rates. Update scripts/pricing.env to override.

| Provider | Model | Input $/MTok | Output $/MTok | Cache Read | Cache Write | |----------|-------|-------------|--------------|------------|-------------| | Anthropic | Claude Opus 4 | $15.00 | $75.00 | $1.875 | $18.75 | | Anthropic | Claude Sonnet 4.5 | $3.00 | $15.00 | $0.30 | $3.75 | | Anthropic | Claude Haiku 4 | $0.80 | $4.00 | $0.08 | $1.00 | | OpenAI | GPT-4.1 | $2.00 | $8.00 | $0.50 | — | | OpenAI | GPT-4.1 mini | $0.40 | $1.60 | $0.10 | — | | OpenAI | o3 | $2.00 | $8.00 | — | — | | OpenAI | o4-mini | $1.10 | $4.40 | — | — | | Google | Gemini 2.5 Pro | $1.25 | $10.00 | — | — | | Google | Gemini 2.5 Flash | $0.15 | $0.60 | — | — |

Key Optimization Patterns

1. Workspace File Trimming (Biggest Win)

Most agents load 30-80K tokens of workspace files every turn. Trimming to essentials saves 50-80% of input costs.

2. Model Routing for Background Tasks

Heartbeats and cron jobs don't need Opus. Route to Haiku/Flash for 10-20x savings on background work.

3. Heartbeat Cache Alignment

Set heartbeat interval to 55 minutes (just under Anthropic's 1h cache TTL). Keeps cache warm = cache-read rates instead of cache-write rates.

4. Lazy Skill Loading

Each skill description adds ~200-500 tokens to system prompt. 20 skills = 4-10K extra tokens/turn. Consider a skill index that loads on-demand.

5. Context Pruning

Enable compaction.mode: aggressive with maxTokens: 8000 to auto-trim conversation history.

Privacy

All analysis runs locally. No data leaves your machine. No API calls for the audit.

🦞 CostClaw Pro — Deep Optimization Package

The free analyzer gives you a ranked report. CostClaw Pro gives you the tools to automate savings:

Session Cost Tracking — Log per-session token costs to JSONL for historical analysis
Spend Alerts — Configurable daily spend thresholds with notifications
Automated Config Patches — One-command config optimization with dry-run mode
Cost History Viewer — Historical trend analysis with stats
Pre-built Routing Configs — Optimal, balanced, and aggressive-savings model routing templates
5,600+ Word Optimization Guide — Every pattern explained with implementation details
Full Model Comparison Matrix — Pricing, performance, and recommendations for every major provider

Get CostClaw Pro: https://buy.polar.sh/polar_cl_qKdQKVcwHd4TUo8jXWrDoZyjHZm3rSciMyVCG00EJlO

Storefront: https://polar.sh/morpheis

Author: ClawdActual (@clawdactual) License: MIT