Stop wasting money on agent runs

Vibe Billing scans Claude Code, OpenClaw, and OpenAI-compatible agent logs, shows exactly where money is being burned, then fixes it in one command.

189 developers · $8,566.96 saved · 858 loops killed

$ npx @shinertx/vibebilling scan

View on GitHub

Find wasted spend in under 30 seconds. No signup.

Live Activity

$8,566.96 saved

claude-sonnet-4-20250514

1k tokens

1.4s

Upstream 401

claude-sonnet-4-20250514

1k tokens

515ms

Upstream 401

claude-sonnet-4-20250514

1k tokens

317ms

Upstream 401

claude-sonnet-4-20250514

1k tokens

109ms

Upstream 401

claude-sonnet-4-20250514

1k tokens

126ms

Upstream 401

claude-sonnet-4-20250514

1k tokens

205ms

Upstream 401

claude-sonnet-4-20250514

1k tokens

150ms

Upstream 401

claude-sonnet-4-20250514

1k tokens

110ms

Upstream 401

claude-sonnet-4-20250514

1k tokens

151ms

Upstream 401

claude-sonnet-4-20250514

1k tokens

122ms

Upstream 401

claude-sonnet-4-20250514

1k tokens

347ms

Upstream 401

claude-sonnet-4-20250514

1k tokens

232ms

Upstream 401

claude-sonnet-4-20250514

1k tokens

108ms

Upstream 401

claude-sonnet-4-20250514

1k tokens

108ms

Upstream 401

From scan to safe mode in 60 seconds

Find the waste first. Route traffic second. Keep the agent under control after that.

Scan

Analyze local transcripts and see retry loops, context re-sends, and overkill model usage before you install anything.

npx @shinertx/vibebilling scan

Setup

Run one command to patch configs, verify the connection, and route agent traffic through the firewall.

npx @shinertx/vibebilling setup

Stay in control

Prompt caching, loop detection, budget caps, and smarter routing kick in once traffic is flowing.

Agent → Firewall → LLM

Example Scan Output

Run the scan first. If the waste is real, route traffic through the firewall after that.

Example Scan Output

$ npx @shinertx/vibebilling scan

Analyzing your agent usage for waste patterns and savings opportunities...

Agent Waste Report

Runs analyzed:184

Retry loops:6

Context re-sends:34

Overkill model usage:51

Total agent spend:$381.42

Estimated wasted spend:$312.76

Fix with:

$ npx @shinertx/vibebilling setup

Reads local Claude Code and OpenClaw logs first. No signup required to see the waste.

What a 2-hour Claude Code session actually costs

Real numbers from a production coding session with Opus.

Without Firewall

Prompt tokens (repeated context)2.4M

Duplicate full-codebase reads12x

Stuck retry loops3 loops

Cache hit rate0%

Total cost $47.20

With Firewall

Prompt tokens (cached)2.4M

Cache hit rate (auto-injected)90%

Loops killed before waste3 killed

Effective cost after caching$7.80

Total cost $7.80

$39.40 saved per session

That's $591 / month for a developer running 3 sessions per day.