Grant programmes, fellowships and community funding rounds take weeks or months, and the reasons for a no are rarely shown. Quadratic funding rounds and DAO grants are faster but run on popularity, not rules. Instant Grants is a small-grant machine: funders publish plain-language rules, three different AI models judge each application against each rule and show their reasons, and money lands in about a minute. Every decision goes into a public log and every no can be appealed to humans.
The data it runs on
Applications: the applicant's form, budget lines and links (a public code repository, a portfolio, a short video).
Funders' criteria: rules written by the funder in plain language, versioned and hashed when the pool is published, so they cannot quietly change.
Evidence checks: public repository metadata (creation date, commit authors, fork status, license file) read through the host's public API; student documents checked for tampering and duplicates; phone or ID hashed to stop repeat claims.
Limits and gaps: AI cannot verify enrolment with a university directly; vague rules produce split votes; some applicants have no public footprint.
Fallback: any split on a rule the funder marks as "hard", any fraud signal, and any appeal goes to human reviewers chosen by the funder. Pools can require a short video call above a set amount.
1. The mission
Make small money fast and fair. A student with a good idea should not wait three months and fill in twelve pages for $200. Rules in public, reasons in public, money in a minute.
2. Why people can't resist it
Speed as a thrill: watching the clock and the three judges tick through your rules is gripping, and a yes in under a minute feels unreal.
Fairness you can see: reasons per rule turn a cold no into a clear "fix this and reapply".
Funder pride: funders watch their money reach 50 builders in a week and can show the log to anyone.
3. How it works for real
Pool: funder writes rules; a frontier LLM (Claude via OpenRouter) rewrites them into checkable criteria and warns about vague ones. The rule set is hashed and published with the pool. Funds are deposited as USDC on Base into a pool contract; card deposits via an on-ramp.
Apply: applicant signs in with email, a smart wallet is created (Coinbase Smart Wallet, Privy or Dynamic). Gas is sponsored by a paymaster, so they never see crypto.
Evidence: a fast cheap model (Gemini Flash or DeepSeek) fetches and summarises repo metadata, checks the license, and runs document checks with a vision model.
Panel: three different model families (for example Claude Sonnet, GPT class, Gemini Pro via OpenRouter) each judge every rule independently with a structured output: pass or fail, quoted evidence, reason. A rule passes with 2 of 3; hard rules and fraud need 3 of 3.
Payout and log: the pool contract pays the applicant's wallet; the decision record (rule hash, verdicts, model versions, reasons hash) is written to the public log. Cash out to bank or mobile money via an off-ramp partner.
4. Cost per use
About $0.10 per decision. Three judge calls with mid-size frontier models at about $0.02 to $0.03 each, evidence gathering about $0.01, transfer on Base well under a cent and sponsored.
Free for applicants, always.Funders: free for pools under $2,000; above that, 3% of grants paid, which covers decisions, appeals tooling and the log. Human appeal reviewers are the funder's own people or a paid reviewer network.
5. Growth loop
Funded applicants share "Funded in 47 seconds" cards and their project updates, which bring the next applicants. The public log shows funders exactly how money was used, which brings the next funders. Each pool page is a ready template other funders can clone.
6. Risks and guardrails
Fraud: forked repos, fake IDs and sock puppets. Repo history checks, document duplicate detection, one claim per hashed phone per pool, and a payout cap per person per month.
Bias: three model families reduce single-model bias; monthly audits compare approval rates by country and gender where known; funders see this.
Privacy: the public log shows initials and country only. Documents are deleted after the decision.
Legal: grants may be taxable; the app shows this and gives the applicant a record. Sanctions screening through the off-ramp partner.
7. MVP build order
Pool page with published rules and a simple application form.
Three-model panel with per-rule structured verdicts and reasons.
Pool contract on Base paying an email smart wallet, gas sponsored.
Public decision log and an appeal queue for the funder's reviewers.