Stop Paying Tokens Per File

To rename 500 files, an agent lists the directory, reasons over every name, and emits a tool call per file. The bill grows with the batch. A Disk Tamer task describes the whole job once: the agent writes it, you review it, and a deterministic engine does the work. Same token cost for 5 files or 5,000.

Read the task schema →

Who it's for

Developers using coding agents

Stop watching Claude Code or Codex spend a context window renaming build artifacts. Have it write one task instead, and keep the context for the work that actually needs reasoning.

Teams on metered AI budgets

Bulk file work is the least valuable thing an agent can spend tokens on, and the easiest to move off the meter. The cost of a task does not grow with the number of files.

Anyone who wants a human in the loop

The agent proposes; you approve. You see exactly what will happen, choose the folder yourself, and click Run. Nothing touches your disk until you say so.

How it works

  1. Your agent writes the taskOne small JSON document naming the operation and the rule that selects files — not a list of filenames. Point your agent at the schema page and it can write one correctly the first time.
  2. You review itOpen the link or paste the task. Disk Tamer shows in plain language what it would do before asking for anything. A task never carries a folder — you always pick that yourself.
  3. Run it, then hand back the resultThe operation runs locally against the folder you chose. Copy the compact result receipt back to your agent so it can verify what happened and carry on.

Common use cases

  • Rename hundreds of build artifacts or exports to a convention without spending a context window
  • Generate one file per row of a CSV — contracts, configs, scaffolds — from a single instruction
  • Clean up temporary or log files by rule instead of listing them all into the agent's context
  • Apply the same content edit across every matching file in a folder
  • Move matched files into an archive folder in one reviewed step
  • Give a non-technical colleague a link that does the bulk task your agent designed

Frequently asked questions

Can the agent run this itself?
No, and that is deliberate. Disk Tamer is agent-authored and human-run: the agent writes the task, but a person always chooses the folder and clicks Run. Nothing executes automatically, including a task where every field is valid. That is what makes it safe to accept a task from a link.
How much does this actually save?
It depends on batch size, and the estimator on this page lets you check with your own assumptions. The structural point matters more than any single number: an agent's cost grows with the number of files, and a task's does not. At 5 files the difference is noise; at 5,000 it is the whole bill.
Does my agent need a plugin or an MCP server?
No. It writes a JSON document and gives it to you — as a link, as pasted text, or as a .disktamer file. There is nothing to install on either side.
Is my data sent anywhere?
No. Disk Tamer works directly against a folder you pick, in your browser, using the File System Access API. Deep links carry the task in the URL fragment, which browsers never send to a server — so even the task itself stays on your machine.
What if my agent writes an invalid task?
Disk Tamer reports every problem at once, each with the exact path that is wrong, such as steps[0].params.findText. One correction fixes everything, so a mistake costs one cheap retry rather than a guessing loop.

What does it actually cost?

Assumptions

A task document is a fixed ~250 tokens regardless of batch size, because it describes the rule rather than the files.

Agent, file by file: 35,000 tokens
One Disk Tamer task: 250 tokens
Roughly 140× fewer.

Estimates, not measurements. Real costs vary by model, filename length, and how often the directory is re-read. Adjust the assumptions above to match what you actually see.