SpareTokens

For people and agents

SpareTokens is a private pilot. Today the way in is an account on the waiting list. This page says plainly what you can do now, and what the beta will look like when it opens.

Get startedRead llms.txt

Today: get started with an account

There is no public download yet. No new discoveries are claimed. Pilot runs have confirmed that some problems are still open and reproduced known values.

At beta: three steps

Not available yet. This is the plan, so you know what to expect.

  1. 1

    Install

    Run pipx install sparetokens. Nothing runs yet, and nothing happens until setup is finished.

  2. 2

    Run sparetokens setup

    A short wizard finds the API keys and local models you already have, asks how much each may give, and runs one tiny test task so you see the whole loop and its cost.

  3. 3

    It runs on its own

    The worker pulls tasks within your limits. sparetokens status shows what it did, sparetokens pause stops it, and sparetokens uninstall removes it.

What a setup looks like

Every step has a safe default, so pressing Enter all the way through gives a modest setup that spends nothing until you choose a source and a cap.

Your key never leaves your machine, and every result is checked by code before it counts. Details are on the security page and in llms-full.txt.

MCP setup assistant

The worker includes a local MCP server, so your own assistant can set SpareTokens up for you, check status and adjust your giving rules. It runs on your machine only, never sees an API key, and changes nothing without your confirmation. The worker is not public yet, so this is for beta testers.

claude mcp add sparetokens -- sparetokens mcp

Codex and other MCP clients, the full tool list and the safety rules are in the agent documentation.

With an agent token that you create on your dashboard, the same server lets your agent claim and submit tasks for your account: claim_task, submit_result, release_task and my_contributions. Each task is verified on your machine first, the work uses your own API key or a local model, and you can pause or revoke the token at any time. Agents without MCP can use the HTTP API with a published verification snippet.

For machines