Agent Run Cost Estimator
Pick a realistic OS agent task, choose the model, and see the exact cost of one execution plus daily/weekly projections. Built for the way agents actually run in the Web OS.
CLIENT-ONLY • NO DATA SENT
Why a dedicated run estimator for the OS?
Agents in the Web OS execute structured plans that combine long context, multiple tool invocations, and handoffs. A generic chat token estimator underestimates (or overestimates) the cost of real agent work. This tool uses task-specific multipliers derived from the way OS agents actually behave.
5 use cases
- Quickly cost a new research agent before adding it to your daily cron.
- Compare the marginal cost of giving an agent access to the real browser vs pure search.
- Model the economics of a 5-agent collaborative workflow vs a single large agent.
- Decide whether to use a premium model only for the final step of a pipeline.
- Build a simple business case for adding more always-on agents.
Frequently asked questions
They are excellent planning numbers. Real runs can vary; the OS shows exact token usage after execution so you can calibrate the estimator over time.
The multipliers embed typical overhead for browser actions, file reads, search, etc. For extremely tool-heavy agents you can increase the step count.
Use this for per-run precision on OS-native tasks. Use the general calculator when you want to model broader LLM spend or compare against non-OS usage.
The platform already tracks real usage. This estimator is the planning companion so you can make good decisions before you launch or scale agents.
100% client-side. Change parameters all day — nothing leaves your browser.