Ask where an on-chain AI agent can get funded and you usually get a list of every grant program in crypto. That is not a map. Many of those programs will not fund an agent at all, some will only fund one after it has users, and a few are a strong fit that founders overlook because the word “agent” does not appear on the page.
This is the map we give teams who ask us. Every program below is on our Web3 Grants Tracker, re-checked on 27 September 2026, with the ticket it states publicly; where there is no fixed ticket, we say “varies” rather than guess. Terms change, so read each program’s own page before you apply.
Four doors, four different questions
- Ecosystem grants ask whether your agent makes a chain more useful. Foundations fund builders on their own stack, so the chain your agent transacts on matters more than how clever the model is.
- Public-goods rounds ask whether the work is open and has already helped people.
- VC grant programs ask whether the project could become a company worth backing. They can be chain-agnostic, and they read the team as closely as the milestone.
- Accelerators and incubators ask whether structure will make the team faster. Our tracker lists grant programs, not standalone accelerators, so we name none here; if you consider one, check what it takes in equity or tokens first.
Ecosystem grants: start with the chain your agent runs on
A few programs now name AI or agents in their focus:
- Superteam Earn Grants (Solana) lists agentic and AI grants alongside regional Solana Foundation grants: up to $10K per grant in USDG/USDC, micro-grants up to $500, equity-free, aimed at idea and prototype stage.
- Prezenti (Celo) runs a Frontier Pool for AI and agent-economy infrastructure at $25K+, next to an Anchor Pool of roughly $25K–$50K by stage. It wants live apps with usage; the current round is open until 29 December 2026 and pays in USDm.
- Stacks Foundation Grants (Bitcoin L2) lists AI experimentation in its focus. Ticket varies across its Getting Started, Builders and Community grants and community-led DeGrants.
General ecosystem programs also fund agents when the agent is good for the chain. The Solana Foundation (ticket varies) runs milestone-based grants, convertible grants and RFPs for open-source infra and tooling. The Starknet Foundation offers Seed Grants of up to $25K in STRK for an MVP or proof of concept and Growth Grants of $25K–$1M in STRK for production products. Avalanche Builder Hub (ticket varies) lists Team1 Builder Grants, research grants of up to $50K and developer credits. If your agent lives on one protocol, that protocol’s own program can be the sharper ask: the Uniswap Foundation and Aave Grants DAO both list their tickets as varying, and Aave takes applications on a rolling basis.
Two traps. Arbitrum’s general Foundation and DAO grant programs are listed as inactive; what is active is the Audit Program ($10M ARB), ArbiFuel gas sponsorship, Gaming Ventures and Alchemy-Arbitrum infra credits of up to $500K. The audit, gas and credit programs help an agent near mainnet; none is a general first grant. And Sui says it no longer runs open grants; it uses RFPs, a $25K academic research award and direct investment instead.
Verifiable and private agents: the cryptography programs
If the point of your agent is that its actions can be proven or kept private, look at programs that fund cryptography directly. The Ethereum Foundation’s Ecosystem Support Program covers infra, research and cryptography/ZK, with a varying ticket, rolling applications, and wishlist and RFP items. Aleo’s developer grants go up to $100K for privacy-first and zero-knowledge apps and tooling. They are milestone-based, require open source and list initial feedback within one week.
Public-goods rounds: for open agent tooling, after it ships
Public-goods money rarely fits a closed agent product, but it can fit open infrastructure around agents: SDKs, policy engines, datasets, monitoring. The catch is timing. Optimism Retro Funding pays retroactively for impact already delivered on the Superchain, through missions such as Onchain Builders and Dev Tooling, with enrollment via OP Atlas. Octant allocates community-directed ETH to live Ethereum public goods in recurring epochs. Gitcoin runs seasonal quadratic-funding rounds, and Giveth hosts ecosystem-sponsored rounds with their own matching pools. Quadratic funding weighs how many people back you, so it works once real users exist, not before.
VC grant programs: funding the company, not only the deliverable
A VC grant program is paid for by an investor rather than a chain or a DAO, so the question becomes “is this a team we want to keep backing?” Ours is the one on the tracker: ZK Prime Capital Grants, $5K–$75K, no equity, idea or prototype stage, for on-chain AI agents, DeFAI and ZK infrastructure, with a reply within 5 business days. What differs from a foundation grant is what follows: if the milestone ships, an optional $50K–$250K round, incubation and a token launch.
Idea stage vs MVP: where to apply first
- Idea or prototype. Start where early work is explicitly in scope: ZK Prime ($5K–$75K), Superteam Earn (up to $10K; micro-grants up to $500) and Starknet Seed (up to $25K in STRK for an MVP or proof of concept). Any-stage programs such as ESP, the Solana Foundation and Aleo are open to early applications too; name one concrete milestone.
- MVP with users. Traction-gated programs open up: Prezenti’s Frontier Pool (live with usage), Aptos Foundation Grants (live product with early traction; ticket varies, plus up to $25K in audit credits and up to $100K a year in cloud credits, with a Payments track that suits agents that move money), Starknet Growth, Arbitrum’s audit and gas programs, then the retro and quadratic rounds.
Apply to more than one: a milestone funded by one program is evidence for the next.
What reviewers look for in an agent project
Agent applications are judged on three questions an ordinary dApp application never has to answer.
- Who holds the keys. Say where the signing key lives, what it is allowed to spend and how it is revoked. A hot key on a server with unlimited approvals is a red flag. A smart account with session keys, spending limits or allowlisted contracts shows you have planned for the day the model does something wrong.
- What the agent actually executes. The output of an on-chain agent is a transaction, not a paragraph. Name the actions, how they are simulated before sending, and what happens on a revert, a price move or a stuck transaction. If the agent would run the same way on a server with no chain, reviewers will notice.
- How anyone can verify it. Users are trusting software with money. An on-chain audit trail, signed decision logs or a proof that the agent followed its stated policy turns “trust us” into something a reviewer can check.
What we fund and pass on is laid out on our grants for on-chain AI agents page.
Building an on-chain agent? Pick two or three programs from the tracker that match your chain and stage, and send us the same one-pager. Apply with your project ▸