AI Agents Don't Have to Break the Bank.

Token costs are real. Typical usage runs $1–150/month per agent. But without visibility, it can spiral. Pinchy gives you the tools to stay in control.

What AI agents actually cost.

💸

Typical: $1–150/month

Most agents running on Claude Sonnet or GPT-4o cost between $1 and $150 per month, depending on usage volume. That's manageable — if you know what you're spending.

🔥

Horror stories: $3,600+

Run expensive models without limits, let agents loop on complex tasks, and you'll see bills that make your CFO cry. It happens. We've heard the stories.

🤷

The real problem: no visibility

OpenClaw doesn't track token usage or costs. You find out what you spent when the API bill arrives. That's not good enough for teams.

Model choice matters more than anything else.

The difference between models isn't incremental — it's an order of magnitude.

Model Relative Cost Best For
Claude Opus / GPT-410xComplex reasoning, critical tasks
Claude Sonnet / GPT-4o3xDaily work, good balance
Claude Haiku / GPT-4o-mini1xSimple tasks, high volume
Ollama (local)Free*Air-gapped, cost-zero, privacy-first

* Local models require your own hardware but have zero API costs.

💡 Pro tip

Assign cheaper models to routine tasks and reserve expensive models for complex reasoning. A triage agent doesn't need Opus. Pinchy lets you set the model per agent — this works today.

Visibility first. Then control.

We're honest about where we are. Some features work today, some are coming soon. Here's the full picture.

🎛️

Model Selection Per Agent

Choose the right model for each agent. Your email triage agent runs on Haiku. Your code review agent runs on Sonnet. Your strategy agent gets Opus. You decide.

Works today
📋

Audit Trail Shows Every Tool Call

Every action an agent takes is logged with HMAC-signed integrity. You can see exactly what happened, when, and how many tool calls were made. Not cost tracking per se — but full transparency into what your agents are doing.

Works today
📊

Usage & Cost Dashboard

A dashboard that shows token consumption, estimated costs, and usage patterns per agent. See which agents are expensive and why. Make informed decisions about model assignments.

Coming next (#35)
🚧

Per-Agent Budget Limits

Set a monthly token budget per agent. When the budget is reached, the agent pauses and notifies you. No more surprise bills. No more runaway loops eating your API credits.

Coming soon (#36)

We're building this. Honestly.

Model selection per agent works today. The audit trail works today. The usage dashboard is next, budget limits follow. We ship fast, and we don't claim features before they exist. Follow our progress on the blog or GitHub.

Want to control your AI agent costs?

Book a call and we'll show you how Pinchy helps you manage spending — today and as we ship more.

Book a Call →

Or email us: hey@clemenshelm.com