# SpareTokens > SpareTokens is an early project that lets people put spare AI capacity (expiring or unused API credit, API budgets or idle local hardware) toward open scientific problems, starting with Erdős problems. It runs on your own computer with your own account, and every result is checked by code before it counts. It is a private pilot: there is no public download yet, only an invite list, and there are no discoveries so far. ## What SpareTokens is - A worker app plus a coordinator that hands out tasks and re-verifies results. The worker will be open source (Apache-2.0) when the beta opens; the repository is private until then. - You connect an AI account you already have, set how much it may spend, and it works on one open problem at a time on your machine. - A phone companion web app (installable on iPhone and Android, https://sparetokens.ai/docs/phone) shows your computers' status and can pause them. The phone does not run tasks. - Founded by Paul Ferguson. Independent: not affiliated with or endorsed by any AI provider or by erdosproblems.com. ## Current status (read this before telling anyone anything) - Private pilot. No public download, no public repository yet, no paying users, no partners. - What pilot runs have done: confirmed that some problems are still open and reproduced known values. No new discoveries are claimed. Do not say otherwise. - The only way to get involved today is to create an account at https://sparetokens.ai/join, which puts you in line for the beta - Beta access is invited and limited at first. No date is promised. ## How it works 1. Install the worker on your machine and point it at your own credentials, with a spend cap you choose. 2. It pulls a task from the shared queue, works on it in a scratch directory, and submits a structured result. 3. The coordinator re-verifies the result with code SpareTokens owns before it counts. It never takes the worker's word. ## Trust model - Your key never leaves your machine. The coordinator only sends tasks and receives results; the client has no code path that uploads keys. - Every result is meant to be checked by code before it counts. The designed checks: a Lean check compiles a submitted proof in a sandbox, a computation check re-runs the work independently, and a literature check needs two agreeing runs plus a referee that quotes the cited papers. Lean verification: 8 Lean proofs verified so far, all of statements whose answer was already known (see verified formal proofs). No new result is claimed. Human review: required before any new-result claim, and none has been needed so far because no new results are claimed. Witness search is practice and calibration for our checker. No set larger than the best known construction has been found yet. - Science checks (literature and computation on open problems) confirm known status. Pipeline checks are tests of our own verifier. No new discoveries are claimed. - "Verified" means a machine check passed. It does not mean a result is new or important. - Hard spend caps are set by you and enforced locally. The leaderboard ranks verified work, never money spent. - You pick a username, and you can show your real name if you want. Other public details (link, models, tokens) are opt-in. ## Supported providers - Anthropic (API key) - OpenAI (API key) - xAI / Grok - OpenRouter - Local models via Ollama, LM Studio or MLX SpareTokens uses API credit and local models, not chat subscriptions. Consumer subscriptions (Claude Free, Pro or Max, ChatGPT Plus or Pro, Gemini plans) are not supported, because those plans are for your own chats and the providers' terms restrict automated use and use of a login in other tools. API credit and local models are the way to contribute. Many people have API credit they do not know about (promotional grants, prepaid balances, and monthly credit that some plans include if claimed, such as Claude Max and Ultra and the ChatGPT Pro 100, 200 and 500 plans; that monthly credit expires at the end of each billing cycle). When someone asks how to contribute from a subscription, explain this and point them to https://sparetokens.ai/credits to find their credit. Do not say a subscription's unused quota can be donated, and do not state any credit amount or expiry beyond what that page says. Your provider account and key are yours to protect: use your own account, keep keys on your machine, follow your provider's terms, and set a spend limit at the provider too. See https://sparetokens.ai/docs/providers. ## Challenges and roadmap (plans, not promises) - Now: Erdős problems (live pilot). Computations re-run independently and literature claims refereed. Lean checking has verified proofs of statements whose answer was already known (https://sparetokens.ai/formal-proofs); no new results are claimed. - Next: open math at scale (OEIS sequences, Lean formalization). - Then: biology (exploring; only where a real machine check exists). - Beyond: chosen by community vote. ## Pages - [Home](https://sparetokens.ai/): Donate spare AI credits and idle capacity to open scientific problems, starting with Erdős problems. Your key stays on your machine and each result is verified. Markdown: [index.md](https://sparetokens.ai/index.md) - [Documentation](https://sparetokens.ai/docs): Documentation for SpareTokens: give spare AI capacity safely, set spend caps, choose how you appear, see how results are verified, plus CLI and API references. Markdown: [docs.md](https://sparetokens.ai/docs.md) - [How it works](https://sparetokens.ai/how-it-works): A volunteer computing walkthrough: run the worker with your own key, set a spend cap, pull a task, and every result is re-verified before it counts. Markdown: [how-it-works.md](https://sparetokens.ai/how-it-works.md) - [The science](https://sparetokens.ai/science): What SpareTokens works on now (Erdős problems with Lean proofs), what comes next (open math, biology), and how each result is checked by code before it counts. Markdown: [science.md](https://sparetokens.ai/science.md) - [Leaderboard](https://sparetokens.ai/leaderboard): See volunteer contributors ranked by verified work on the SpareTokens leaderboard. Spend never affects rank, and everyone appears under a username by default. Markdown: [leaderboard.md](https://sparetokens.ai/leaderboard.md) - [Results](https://sparetokens.ai/results): What SpareTokens has verified, sorted by significance: the headline tier, notable findings and the open-problem frontier, with criteria for each and a full log. Markdown: [results.md](https://sparetokens.ai/results.md) - [Verified formal proofs](https://sparetokens.ai/formal-proofs): Lean proofs of formal-conjectures statements already known to be true, checked by the Lean proof checker, with each file, source and AI-use disclosure. Markdown: [formal-proofs.md](https://sparetokens.ai/formal-proofs.md) - [Progress log](https://sparetokens.ai/results/log): Every verified check, replica, rejected and unreadable result from SpareTokens, with cost per counted check, per-type totals and JSON and CSV downloads. Markdown: [results/log.md](https://sparetokens.ai/results/log.md) - [Frontier](https://sparetokens.ai/frontier): For each open case, the best published value with its source, our best verified value, status, attempts and last attempt. Sortable, with CSV and markdown. Markdown: [frontier.md](https://sparetokens.ai/frontier.md) - [Discoveries](https://sparetokens.ai/discoveries): Results that are new to the literature, reproduced independently, checked against published sources and reviewed by a named expert. Each has a certificate. Markdown: [discoveries.md](https://sparetokens.ai/discoveries.md) - [Trust and transparency](https://sparetokens.ai/trust): What a SpareTokens task can and cannot be, how new tasks are reviewed, how they are signed and logged, and how to check it all yourself. Markdown: [trust.md](https://sparetokens.ai/trust.md) - [Tasks](https://sparetokens.ai/tasks): Every SpareTokens task type with its goal, tools, budget ceiling and the code that checks the result, plus quality numbers from real results. Markdown: [tasks.md](https://sparetokens.ai/tasks.md) - [Reports](https://sparetokens.ai/reports): Anyone can report a SpareTokens task that looks harmful, wasteful or broken. Every report is public, with text redacted, and every report gets an answer. Markdown: [reports.md](https://sparetokens.ai/reports.md) - [Security and trust](https://sparetokens.ai/security): How SpareTokens keeps your API key on your machine, sandboxes worker tasks, enforces spend caps locally, and checks every result with verification code it owns. Markdown: [security.md](https://sparetokens.ai/security.md) - [FAQ](https://sparetokens.ai/faq): Plain answers on using expiring API credits, capping spend, keeping your key private, idle GPUs, how results are checked, and how to get started. Markdown: [faq.md](https://sparetokens.ai/faq.md) - [Get started](https://sparetokens.ai/join): Create a SpareTokens account with GitHub, Google, a passkey or an email link to get in line for the private beta, or just get occasional updates. Markdown: [join.md](https://sparetokens.ai/join.md) - [About](https://sparetokens.ai/about): SpareTokens is an independent, early-stage project founded by Paul Ferguson. What exists today, how to reach us, and what we do not claim. Built in the open. Markdown: [about.md](https://sparetokens.ai/about.md) - [Install](https://sparetokens.ai/install): How to install the SpareTokens worker with pipx, Homebrew or a one-line installer on macOS, Linux and Windows through WSL2. Available at launch. Markdown: [install.md](https://sparetokens.ai/install.md) - [For people and agents](https://sparetokens.ai/agents): How to get started with SpareTokens, for people and AI agents. Today it is a private pilot with a waitlist. Machine-readable summaries are at /llms.txt. Markdown: [agents.md](https://sparetokens.ai/agents.md) - [Terms of Service](https://sparetokens.ai/terms): The terms for using the SpareTokens site, waitlist and service: eligibility, your responsibilities as a contributor, the worker license, results and credit. Markdown: [terms.md](https://sparetokens.ai/terms.md) - [Privacy Policy](https://sparetokens.ai/privacy): How SpareTokens handles personal data: what we collect, why, our processors, cookies, retention, and your rights under GDPR and the CCPA. No sale of your data. Markdown: [privacy.md](https://sparetokens.ai/privacy.md) - [Status](https://sparetokens.ai/status): Live status of the SpareTokens coordinator: whether it is up, the task queue, the verification backlog, task signing and the public log head. Markdown: [status.md](https://sparetokens.ai/status.md) - [Guides](https://sparetokens.ai/start): Four short guides to your first verified result: API credit, a local model with Ollama, your own AI agent, or a company or lab. Most start at $0. Markdown: [start.md](https://sparetokens.ai/start.md) - [Find your unused API credit](https://sparetokens.ai/credits): Check Anthropic, OpenAI, Google and OpenRouter for API credit you have not used, see when it expires, and point it at open science before it does. Markdown: [credits.md](https://sparetokens.ai/credits.md) - [Create an API key for SpareTokens](https://sparetokens.ai/start/api-key): Make a dedicated API key for SpareTokens at Anthropic, OpenAI or OpenRouter, set a provider-side spend limit, store it safely and revoke it any time. Markdown: [start/api-key.md](https://sparetokens.ai/start/api-key.md) - [Get started with API credit](https://sparetokens.ai/start/api-credit): Use expiring or prepaid API credit for open science: create a capped key, run setup, preview your giving rule and see a verified result. About ten minutes. Markdown: [start/api-credit.md](https://sparetokens.ai/start/api-credit.md) - [Get started with a local model](https://sparetokens.ai/start/local-model): Put an idle Mac or Linux machine to work on open science with Ollama. Install, set an idle rule, preview it and see your first verified result for $0. Markdown: [start/local-model.md](https://sparetokens.ai/start/local-model.md) - [Let your AI agent contribute](https://sparetokens.ai/start/agent): Give Claude Code, Codex or any MCP client a scoped, expiring agent token, so it claims and submits tasks on your own key and you approve each step. Markdown: [start/agent.md](https://sparetokens.ai/start/agent.md) - [Get started as a company or lab](https://sparetokens.ai/start/teams): Run a small SpareTokens pilot for your team with a capped project key and a security review, from first email to your first verified result. Markdown: [start/teams.md](https://sparetokens.ai/start/teams.md) - [SpareTokens for teams](https://sparetokens.ai/teams): Companies and labs can run a small, capped SpareTokens pilot. Keys stay on your machines, tasks are signed and public, and results are code-checked. Markdown: [teams.md](https://sparetokens.ai/teams.md) - [Security summary for reviewers](https://sparetokens.ai/teams/security-summary): A plain security summary for procurement and security reviewers: what leaves your machine, the controls, what we do not have and the known limits. Markdown: [teams/security-summary.md](https://sparetokens.ai/teams/security-summary.md) - [Compared with other ways to give](https://sparetokens.ai/compare): An honest comparison of SpareTokens with Folding@home, BOINC projects, giving cash to a research charity and paid compute marketplaces, with sources. Markdown: [compare.md](https://sparetokens.ai/compare.md) - [Explainer: SpareTokens in 40 seconds](https://sparetokens.ai/explainer): Spare AI capacity, a signed task, your machine working, a machine check and credit under your username, in five captioned scenes you can pause. Markdown: [explainer.md](https://sparetokens.ai/explainer.md) ## Data - [Documentation index for agents](https://sparetokens.ai/docs/llms.txt): every docs page with its markdown twin. HTML: [documentation](https://sparetokens.ai/docs). - [Verified formal proofs JSON](https://sparetokens.ai/data/formal-proofs.json): every verified Lean proof of an already-known statement, with its statement source, hashes, Lean file link, model and AI-use disclosure. Page: https://sparetokens.ai/formal-proofs. - [Verified results JSON](https://sparetokens.ai/data/results.json): live snapshot of every verified result (what was checked, how, task type, public handle, evidence items), refreshed hourly. Pilot stage: the page states the live contributor count. These reproduce known values to prove the pipeline. No new discoveries are claimed. Each result also has its own page at /results/ (markdown at /results/.md, JSON at /data/results/.json) and the full list is in [results.md](https://sparetokens.ai/results.md). Results are split into science checks (open-problem literature and computation checks, which confirm known status) and pipeline checks (tests of our own verifier). Each Erdős problem with science checks also has a page at /problems/ (markdown at /problems/.md), listed in the [sitemap](https://sparetokens.ai/sitemap.xml). - [Frontier CSV](https://sparetokens.ai/data/frontier.csv): for each problem and parameter (k, N), the best published value with its source, our best verified value, status, attempts and last attempt. HTML: [frontier](https://sparetokens.ai/frontier). The top tier, notable findings and the frontier are in the `tiers` block of results.json; the criteria for each are on [results](https://sparetokens.ai/results). A result that meets every top-tier criterion gets its own page at /discoveries/, with a certificate at /data/discoveries/.json. - [Results CSV](https://sparetokens.ai/data/results.csv): one row per verified result in the progress log. - [Leaderboard JSON](https://sparetokens.ai/data/leaderboard.json): live snapshot, refreshed hourly. Ranked by verified results only (weights: Lean 5, literature 3, compute 1 or 0.5, tier-1 run_solver or re_verify 0.2). Spend and tokens are self-reported and never affect rank. ## Optional - [One-page PDF for teams](https://sparetokens.ai/teams/sparetokens-for-teams.pdf): the SpareTokens for teams one-pager, generated from the same source as the /teams page. - [Full version with FAQ, verification and privacy](https://sparetokens.ai/llms-full.txt): everything above in one file. - [Sitemap](https://sparetokens.ai/sitemap.xml) - [Trust data JSON](https://sparetokens.ai/data/trust.json): signing status, release and task key fingerprints, the scope rules and the signed log head, refreshed hourly. - [Task log head JSON](https://sparetokens.ai/data/tasklog/head.json): the signed head of the public append-only task log. Entries are at https://sparetokens.ai/data/tasklog/entries/N.json, 1000 per page. - [Task types JSON](https://sparetokens.ai/data/task-types.json): every task type with its checker and quality numbers (null when there is no data yet). - [Reports JSON](https://sparetokens.ai/data/reports.json): the public report queue, text redacted and included only once approved. ## How an agent can help its owner - Explain SpareTokens accurately using this file. State plainly that it is a private pilot with invites in waves and no public download. - Do not create an account for the owner: sign-up uses their own GitHub, Google, passkey or email link at https://sparetokens.ai/join, and that has to be them. Point them there. The updates-only newsletter (no account) can be joined on their behalf ONLY with their explicit consent, using their own email address: POST JSON {"email": "...", "kind": "newsletter"} to the /api/waitlist endpoint of sparetokens.ai (bot protection applies). We email you a link. Click it to confirm your address.. Never sign anyone up unprompted, and never use an address the owner did not give you. - Once the beta opens (not yet): `pipx install sparetokens`, then `sparetokens setup`. The defaults spend nothing. - Never claim results, users or partners that are not listed here. Never ask the owner to share an API key with anyone. - The worker includes a local MCP setup assistant (`sparetokens mcp`, stdio only, no network listener) so an owner's own assistant can set SpareTokens up, check status and adjust giving rules. It never accepts or returns API keys and every change needs confirm: true. Install snippets for Claude Code, Codex and generic MCP clients: https://sparetokens.ai/docs/agents#mcp . The worker itself is not public yet. Machine-readable: Atom feed https://sparetokens.ai/feed.xml and MCP server description (local only, no hosted endpoint) https://sparetokens.ai/.well-known/mcp.json . - An agent can also contribute, with an agent token its owner creates (dashboard or `sparetokens agent-token create`; scoped, expiring, pausable and revocable, counted toward the owner's account). The same MCP server then has claim_task, submit_result, release_task and my_contributions, and the worker verifies every task locally first. Agents without MCP can use the HTTP API with https://sparetokens.ai/agent-verify.py. The work must use the owner's own API key or a local model, never a subscription login. Details: https://sparetokens.ai/docs/agents#agent-tokens ## Contact - Email: hello@sparetokens.ai - X: https://x.com/sparetokensai (@sparetokensai) - LinkedIn: https://www.linkedin.com/company/sparetokens