A model that plans
An LLM reads the goal and decides the next step. It can be a hosted model or a local one. The model is the reasoning layer; it should not be the place where private keys live.
Blockchain AI agents, explained
An AI agent is software that can plan and take actions. A blockchain agent adds a wallet, onchain tools, and a chain to act on. This guide explains the stack, the custody trade-offs, and the risks in plain English — then lets you filter 40 real projects by the job you actually need done.
- 40projects in the directory
- 6layers in the stack
- 11jobs in the finder
- Oct 2026last editorial check
Start here
What is a blockchain agent?
Three pieces have to come together. Remove any one of them and you have a chatbot, a wallet, or a script — not an agent that can act onchain.
Tools it may call
Swaps, lending, bridging, mints, queries, and payments are exposed as typed tools. The agent can only do what its tool list and permissions allow — this is where most safety is won or lost.
A way to settle
A wallet or smart account signs transactions on a chain. Whoever controls that signer controls the funds. “The AI did it” is never a custody model.
| System | Decides what to do? | Can move funds? | Typical example |
|---|---|---|---|
| Chatbot | Answers, but does not act | No | A crypto Q&A assistant |
| Trading bot | Follows fixed rules | Yes, within its rules | A grid or DCA bot |
| Blockchain agent | Plans with an LLM, then calls tools | Yes, if given a signer and permissions | A yield or portfolio agent |
| Multi-agent system | Several agents split the work | Only the agents with wallets | Research agent + execution agent |
How it works
The stack, layer by layer
Most projects specialize in one or two layers. That is why “the best agent” is usually the wrong question — the better question is which layer you are missing.
Interface
Where a human or another agent talks to it: chat, X, Telegram, Discord, a web app, or an API.
ElizaOS connectors · Bankr · HeyElsa
Brain
An LLM plus a planner and memory. It decides what to do next, but it should never hold the private key.
ElizaOS · GAME · Daydreams
Tools
Typed actions the model may call: swap, lend, bridge, mint, query a subgraph, or send a payment.
GOAT SDK · AgentKit · Solana Agent Kit
Guardrails
Policy, simulation, allowlists, spend caps, and human approval. This is the layer that limits a mistake.
Safe modules · Turnkey policies · Brahma
Settlement
The chain and contracts where the action becomes final. Gas, finality, and MEV apply here.
Ethereum · Solana · Base · L2s
Data & payments
Inputs the agent reasons over and rails it uses to pay for services: indexing, forecasts, memory, and x402.
The Graph · Allora · x402
Rule of thumb: the closer a layer sits to the keys, the more it should be
deterministic, audited, and boring. Keep creativity in the planner, not in the signer.
Which agent for which job
Start from the task, not the token
Pick the job you need done. Each card names a few projects that fit, with the first one as the default starting point. Click a name to jump to its directory entry.
Build my first agent
Start with a runtime you can run locally, then add one chain action. Keep the first version small enough to read end to end.
Give an agent a wallet
Separate the brain from the keys. A smart account or signing service lets you cap and revoke what the agent can do.
Run a DeFi strategy
Decide whether the agent proposes and you approve, or the agent executes within limits. Yield is never guaranteed.
Do onchain actions by chat
Good for multi-step workflows you already understand. Always check the route, slippage, and destination before signing.
Research a market or token
Use several sources. An agent that summarizes sentiment is not the same as an auditor or a fundamental analyst.
Feed an agent onchain data
Indexed data is faster and cheaper than scanning blocks. Check the subgraph's freshness and schema before trusting it.
Launch or sell an agent
Choose between a service marketplace, a hosted directory, an app store, and a tokenized launchpad. Only one of these needs a token.
Build on Solana
Use Solana-native action tools and be extra careful with finality, slippage, and memecoin liquidity.
Let agents pay each other
Machine payments need hard caps. Start with small fixed prices per call and a daily budget per agent.
Secure and verify execution
Assume the model can be manipulated. Put policy, simulation, and human approval between the model and the funds.
Run agent infrastructure
Infrastructure is an operations business. Understand hardware, uptime, and reward economics before committing capital.
Directory
40 blockchain agent projects
Filter by category, chain, access model, and maturity. Every entry states what it is, who it fits, when to avoid it, and its main risk. Listings are informational, not endorsements.
Build
Stable
Coinbase AgentKit
Wallet + action toolkit
A developer toolkit that gives an existing agent a wallet and a set of onchain actions: transfers, swaps, contract calls, and x402 payments. It works with several agent frameworks instead of replacing them.
Best for Adding a wallet and EVM actions to an agent you already built, especially inside the Coinbase developer stack.
Avoid if You need deep Solana DeFi coverage or a fully non-custodial hardware-signing flow.
Main risk: A hot wallet controlled by software is only as safe as its key storage. Set spend limits and test on a testnet before mainnet.
Build
Beta
Daydreams
TypeScript agent framework
A TypeScript framework for agents with persistent context, modular actions, and onchain integrations. It is aimed at agents that play, trade, or coordinate inside games and crypto applications.
Best for TypeScript teams building an agent with memory and custom actions for a game or consumer crypto app.
Avoid if You need a mature managed platform, or your team is Python-only.
Main risk: Custom actions are your responsibility. Log every tool call and cap what the agent can spend per session.
Build
Stable
ElizaOS
Agent runtime
An open-source TypeScript runtime for agents with memory, model-provider abstraction, social connectors, and a large plugin ecosystem. Onchain plugins let the same agent read chains and send transactions.
Best for A developer who wants a complete, self-hosted agent that can live on Discord, X, Telegram, or a web app and still use crypto tools.
Avoid if You need a managed service with no infrastructure work, or you want the platform to hold user funds for you.
Main risk: Plugins and community packages vary in quality. Audit any plugin that can sign transactions, and keep the signing key outside the model's context.
Build
Stable
Fetch uAgents
Python agent framework
A Python framework for autonomous agents that can message each other, hold identities, and offer services. It connects to AgentVerse for discovery and hosting.
Best for Python developers building multi-agent systems where agents discover and pay each other for services.
Avoid if Your main need is a consumer chat agent with social connectors and a large plugin library.
Main risk: Agent-to-agent messaging introduces impersonation and spam. Verify agent identities and do not trust messages as instructions to move funds.
Build
Stable
GOAT SDK
Onchain action layer
The Great Onchain Agent Toolkit is a library of 200+ onchain actions across many chains. It is designed to be plugged into ElizaOS, LangChain, or your own agent loop as the 'hands' rather than the 'brain'.
Best for Teams that already have an agent framework and want broad chain and protocol coverage without rewriting their planner.
Avoid if You want one opinionated end-to-end runtime with hosting, memory, and UI included.
Main risk: More tools means a larger attack surface: every action the model can call is a potential mistake. Use allowlists and simulation.
Build
Stable
Olas
Agent framework + economy
An open framework for autonomous agents that run as onchain services, plus the tooling for them to earn and coordinate. Olas also ships Pearl, an app store for ready-made agents.
Best for Builders who want an agent to operate as a persistent onchain service with an economic model, not just a chat toy.
Avoid if You want a quick prototype with the least possible setup. The service model has a learning curve.
Main risk: Running an agent as a service means it holds value and acts continuously. Monitor it and define a stop condition before launch.
Build
Stable
Solana Agent Kit
Solana action toolkit
An open-source toolkit that turns natural-language agent requests into Solana actions: token transfers, swaps, staking, NFTs, and DeFi integrations. It plugs into popular agent frameworks.
Best for Solana-first agents that need a ready set of token, NFT, and DeFi actions.
Avoid if Your product is EVM-first, or you need cross-chain routing as a core feature.
Main risk: Solana transactions are fast and final. Add slippage caps, confirmation prompts, and a simulation step before signing.
Build
Beta
Virtuals GAME
Agent cognition framework
GAME (Generative Autonomous Multimodal Entities) is the agent framework behind Virtuals. It handles planning, memory, and actions for agents that live in the Virtuals economy.
Best for Builders who plan to launch inside Virtuals and want native access to its agent wallets and commerce primitives.
Avoid if You want a chain-neutral runtime, or you are not interested in the Virtuals ecosystem.
Main risk: Ecosystem lock-in and token-linked incentives can distract from product quality. Check what is actually onchain versus promised.
Build
Beta
ZerePy
Social agent framework
A Python framework for agents that post, reply, and interact on social platforms while also holding a Solana or EVM wallet. It is a readable starting point for learning how an agent loop works.
Best for Learning by building a small social-plus-crypto agent you can read end to end.
Avoid if You need enterprise reliability, audited modules, or a large maintained plugin ecosystem.
Main risk: Social agents can be manipulated by replies and mentions. Treat every incoming message as untrusted input, never as a command to sign.
Execute
Beta
Almanak
Quant strategy agents
A platform for building and running quantitative DeFi trading agents. Strategies are simulated offchain, then executed onchain through the user's wallet or a delegated setup.
Best for Users with a strategy hypothesis who want backtesting, simulation, and agent execution in one workflow.
Avoid if You want a simple one-click yield product, or you cannot evaluate strategy risk yourself.
Main risk: Backtests overfit. Past performance does not imply future returns, and leverage or illiquid pools can amplify losses.
Execute
Stable
Bankr
Social trading bot
A bot that executes crypto commands in plain language from X, Farcaster, Telegram, or its own terminal: buy, sell, bridge, send, and launch tokens.
Best for Fast, conversational actions on social platforms where the bot creates and manages a wallet for you.
Avoid if You require self-custody of keys, or you are trading amounts you cannot afford to lose.
Main risk: A hosted wallet means you trust the operator's security and availability. Social account takeover can also expose your bot commands.
Execute
Stable
Brahma
Execution console
A non-custodial execution layer with guardrails, simulations, and automation for onchain actions. It is built for teams that need control over what an agent or operator can do.
Best for Teams and advanced users who want a policy layer between an agent and a treasury or wallet.
Avoid if You want a consumer chat app, or you are not ready to configure permissions and policies.
Main risk: Complex guardrail configurations can create a false sense of safety. Test failure cases, not just the happy path.
Execute
Beta
Giza Agent
Stablecoin yield agent
An autonomous agent that allocates stablecoins across lending and liquidity protocols, showing its reasoning and movements. It replaced Giza's earlier ARMA and Pulse products.
Best for Users who want a managed stablecoin strategy with visible reasoning, without manually moving funds between protocols.
Avoid if You need guaranteed returns, a bank-like deposit guarantee, or full manual control of every transaction.
Main risk: Smart-contract and strategy risk remain. APY is variable and can be lower than advertised after fees; start with a small amount.
Execute
Beta
Griffain
Solana agent network
A Solana platform with specialized agents for trading, memecoins, NFTs, and portfolio tasks, coordinated through a conversational interface.
Best for Solana users who want task-specific agents rather than one general-purpose assistant.
Avoid if You need EVM coverage, or you want a fully audited institutional execution stack.
Main risk: Memecoin and NFT agents face extreme volatility and liquidity risk. Set hard limits and never share a seed phrase with an agent.
Execute
Beta
HeyAnon
Natural-language DeFi
A conversational agent that turns plain-language requests into multi-step DeFi actions across several chains, including swaps, lending, and wallet management.
Best for Users who understand DeFi but want to execute multi-step workflows by describing the outcome.
Avoid if You are completely new to wallets and approvals, or you expect the agent to absorb losses from a bad route.
Main risk: Natural language is ambiguous. Confirm the route, slippage, and destination address before signing, and revoke unused approvals.
Execute
Beta
HeyElsa
Crypto copilot
An AI copilot that converts a desired outcome into a plan and executes it through connected protocols. Users can confirm each step or use a smart wallet for more automation.
Best for Users who want a guided, step-by-step agent that still lets them approve transactions.
Avoid if You want a fully autonomous strategy that runs without confirmation.
Main risk: The planner can choose a valid but expensive route. Compare fees and check the final simulated outcome before approving.
Execute
Beta
Theoriq
Agent collectives
A protocol for coordinating groups of specialized agents into collectives that analyze and execute DeFi strategies, with shared incentives for participants.
Best for Exploring multi-agent coordination and delegated DeFi strategies rather than a single assistant.
Avoid if You need a proven consumer product with a simple deposit flow.
Main risk: More agents means more coordination and accountability gaps. It may be unclear which agent caused a loss.
Execute
Beta
Wayfinder
Cross-chain agent protocol
A protocol from Parallel for autonomous agents called Shells that navigate multiple chains and execute strategies such as swaps, bridging, and basis trades from a conversation.
Best for Users who want cross-chain execution and are comfortable with an agent holding scoped permissions.
Avoid if You want a simple single-chain swap, or you are not prepared to manage agent permissions.
Main risk: Cross-chain routes add bridge and timing risk. A failed or delayed leg can leave funds in an unintended state.
Research
Stable
aixbt
Market intelligence agent
An AI agent that publishes crypto market commentary, narrative tracking, and project analysis, primarily on X. Its output is widely used as a sentiment input.
Best for Monitoring narratives and getting a fast, opinionated summary of market activity.
Avoid if You need audited fundamental research, or you are looking for personalized financial advice.
Main risk: Narrative-driven commentary can be wrong or self-reinforcing. It is not a substitute for your own diligence.
Research
Beta
Allora
Decentralized inference
A decentralized network where independent machine-learning models publish predictions that are combined into a consensus forecast. It offers agent toolkits and integrations with major frameworks.
Best for Agents that need a decentralized forecast or inference signal rather than a single model provider.
Avoid if You need deterministic, auditable outputs for high-stakes execution, or you want a simple chatbot API.
Main risk: Consensus forecasts can still be systematically wrong. Understand the incentive design and validate signals out of sample.
Research
Stable
ChainGPT
Crypto AI assistant + API
A crypto-focused AI platform with a chatbot, smart-contract generator and auditor, news, and developer APIs. It can be used directly or embedded in another product.
Best for Crypto Q&A, contract drafting, and adding a crypto-aware assistant through an API.
Avoid if You need guaranteed factual accuracy or a substitute for a professional smart-contract audit.
Main risk: AI-generated contracts and audit summaries can miss critical bugs. Always have generated code reviewed by a human auditor.
Research
Stable
OriginTrail
Verifiable knowledge graph
A decentralized knowledge graph (DKG) for publishing and querying verifiable data. It is used to give AI systems traceable sources and provenance for what they know.
Best for Projects that need an auditable source of truth and provenance for agent knowledge.
Avoid if You just need a fast vector database, or your data is private and cannot be published.
Main risk: Provenance does not guarantee correctness. A verifiable source can still contain bad or outdated information.
Research
Stable
PAAL AI
Crypto assistant + automation
A crypto AI assistant with custom bot creation, alerts, and automation across community and trading workflows. It combines a chat interface with configurable agents.
Best for Communities and traders who want customizable crypto assistants and alerts in one place.
Avoid if You need a neutral research source, or you want a purely local open-source setup.
Main risk: Community bots can be configured badly and amplify rumors. Verify alerts against primary sources before acting.
Research
Beta
Recall
Agent memory network
A decentralized network for storing and sharing AI agent memory and knowledge. Agents can persist context and compete on the quality of what they contribute.
Best for Exploring persistent, shared memory for agents instead of a single centralized database.
Avoid if You need a mature drop-in database with predictable latency and strong privacy guarantees.
Main risk: Shared memory can be poisoned. Treat recalled content as untrusted input and validate it before using it in decisions.
Research
Stable
The Graph
Onchain data indexing
A decentralized indexing network that turns raw chain data into queryable subgraphs and APIs. Agents use it to answer questions about balances, swaps, and protocol activity without scanning blocks themselves.
Best for Giving an agent reliable structured onchain data with a familiar GraphQL interface.
Avoid if You need real-time execution or private offchain data that is not indexed.
Main risk: A subgraph can be stale, incomplete, or written with the wrong logic. Check its schema and freshness before trusting it.
Launch
Stable
AgentVerse
Agent hosting + discovery
A hosting and discovery layer for Fetch.ai agents. Developers can deploy agents, list them for others to find, and connect them to agent-to-agent messaging.
Best for Publishing a uAgents-based agent quickly and letting other agents discover it.
Avoid if You are not using the Fetch/uAgents stack, or you need full control of the hosting environment.
Main risk: Discovery is not distribution. Listing an agent does not create demand, and hosted limits may constrain production use.
Launch
Beta
Olas Pearl
Agent app store
An app store for ready-to-run Olas agents, including trading and prediction-market agents. Users download an app, fund a local wallet, and run the agent on their own machine.
Best for Non-developers who want to run a packaged agent locally without handing over custody.
Avoid if You want a fully managed cloud service, or you cannot leave a computer running.
Main risk: Local operation protects keys but not against bad strategy logic. Review the agent's permissions and expected behavior first.
Launch
Stable
SingularityNET
AI service marketplace
A marketplace where developers publish AI services and agents, and buyers pay per call. It focuses on composable AI services rather than tokenized consumer agents.
Best for Publishing a specialized AI service and reaching buyers through an established marketplace.
Avoid if You need a consumer-grade launch in a day, or you depend on high-frequency low-latency calls.
Main risk: Marketplace demand may be thin. Publishing a service does not guarantee usage or revenue.
Launch
Stable
Virtuals Protocol
Tokenized agent launchpad
A network for launching and co-owning tokenized AI agents. Agents can hold wallets, sell services, hire each other, and route payments through the VIRTUAL token and its commerce layer.
Best for Founders who want to launch an agent with a token, a treasury, and built-in agent commerce.
Avoid if You want to launch a normal SaaS product without a token, or you cannot handle public-market volatility.
Main risk: Agent tokens are highly speculative and often trade on narrative. A token does not prove the agent works or generates revenue.
Infra
Stable
Bittensor
Decentralized ML network
A network of subnets where miners produce machine-learning outputs and validators rank them. It is infrastructure for decentralized models and inference, not a ready-made agent.
Best for Teams that want to contribute compute or models to a decentralized ML market and earn emissions.
Avoid if You need a simple API with predictable latency, or you are not prepared to operate infrastructure.
Main risk: Subnet economics are complex and emissions can dominate real demand. Returns are volatile and operationally intensive.
Infra
Beta
Chainlink CRE
Verifiable workflow runtime
Chainlink Runtime Environment runs workflows that combine onchain events, offchain APIs, and AI models, with consensus and oracle infrastructure behind the result.
Best for Connecting an agent's offchain decision to an onchain action with a verifiable execution path.
Avoid if You need a free, fully self-hosted runtime, or your workflow is entirely onchain.
Main risk: The workflow is only as reliable as its external APIs and oracle assumptions. Design for failure and timeouts.
Infra
Beta
Gaia
Decentralized agent infrastructure
A decentralized network for running AI agents and knowledge nodes, with tools for deploying private or public agents and connecting them to shared knowledge.
Best for Running agents on decentralized infrastructure with a knowledge-node model.
Avoid if You need a mature managed platform with enterprise SLAs.
Main risk: Node operators vary in reliability. Test availability and understand what data is public before deploying.
Infra
Beta
Kite AI
Agent chain + payments
An AI-focused chain with identity and payment primitives for agents, designed so autonomous agents can be identified, authorized, and paid across services.
Best for Projects building agent-native payments and identity on a dedicated chain.
Avoid if You need deep liquidity on an established chain, or you want to avoid ecosystem risk.
Main risk: A young chain has limited liquidity and validator diversity. Bridge and settlement risk are higher than on major networks.
Infra
Beta
Newton Protocol
Verifiable automation
A protocol for delegated onchain automation with cryptographic guarantees. It aims to let users authorize an agent to act within rules without giving up custody.
Best for Teams that need policy-bound automation with an auditable permission model.
Avoid if You want a simple hot-wallet bot, or you need a mature ecosystem with many integrations.
Main risk: Permission logic is critical. A misconfigured policy can allow more than intended; simulate and audit before enabling.
Infra
Stable
Phala
Confidential compute
A confidential-computing cloud built on trusted execution environments (TEEs). Agents can run models and handle secrets inside hardware-isolated enclaves with attestation.
Best for Agents that must process sensitive data or hold API keys without exposing them to the host.
Avoid if Your workload is not sensitive, or you cannot tolerate TEE performance overhead.
Main risk: TEEs reduce but do not eliminate trust. Side-channel attacks and attestation misconfiguration are real concerns.
Infra
Beta
Ritual
Verifiable inference
Infrastructure for running AI models with verifiable execution and onchain access. It aims to let smart contracts and agents use inference without trusting a single server.
Best for Applications that need an auditable link between an AI output and an onchain action.
Avoid if You need the cheapest possible inference, or you want a mature production SLA today.
Main risk: Verification adds cost and latency. The trust assumptions of the specific proof system still need review.
Custody
Stable
Fireblocks
Institutional custody
Institutional custody and wallet infrastructure with policy engines, approval workflows, and APIs. It is used when an agent or automated system operates alongside human approvals.
Best for Institutions that need compliance controls and separation of duties around agent-initiated transfers.
Avoid if You are an individual or small team, or you need a low-cost self-serve tool.
Main risk: Operational complexity and cost. Misconfigured approval quorums can block legitimate actions or fail to stop bad ones.
Custody
Stable
Safe
Smart account custody
A smart-account standard used by teams and agents. Modules, guards, and signer policies define exactly what an agent can execute, and recovery is possible without exposing the main key.
Best for Giving an agent limited, revocable permissions over a treasury instead of a raw private key.
Avoid if You need a simple personal hot wallet, or you cannot manage signer configuration.
Main risk: A malicious or buggy module can bypass intended limits. Review modules and keep a human signer for large transfers.
Custody
Stable
Turnkey
Key management infrastructure
A developer API for secure key management and transaction signing. Policies can require approval, restrict destinations, and scope what an agent is allowed to sign.
Best for Products that need programmatic signing without putting private keys on an application server.
Avoid if You want a free self-hosted setup, or you are building a one-off personal wallet.
Main risk: Policy design is the security boundary. A broad policy turns a secure signer into an expensive hot wallet.
Custody
Beta
x402
Agent payment standard
An open protocol that revives HTTP status 402 so a service can request payment and an agent can pay automatically, usually in stablecoins. It is a building block for machine-to-machine payments.
Best for Charging agents per API call or letting an agent buy a service without a subscription.
Avoid if You need chargebacks, fiat settlement, or a consumer checkout experience.
Main risk: Automatic micro-payments can drain a wallet through loops or malicious endpoints. Enforce per-call and daily caps.
No projects match those filters.
Try a broader chain, a different category, or clear the search.
Access and maturity labels are editorial summaries based on public documentation. “Stable” means the project is widely used or has been live for a while — not that it is audited, insured, or risk-free.
Custody
Who can actually move the money?
The most important question about any agent. “Non-custodial” can mean very different things, so read the signer and permission model rather than the marketing label.
| Model | Who holds the key | Strength | Trade-off |
|---|---|---|---|
| Self-custody | You or your server holds the key. | Maximum control; no platform can freeze the agent's wallet. | A leaked key is a total loss. You own operational security. |
| Smart account | A contract wallet enforces rules; signers approve. | Spend limits, allowlists, recovery, and revocable agent permissions. | Module bugs and complex configuration can create hidden risk. |
| Delegated / hosted | A platform or bot operates a wallet for you. | Fastest onboarding and no key management for the user. | You trust the operator's security, solvency, and withdrawal policy. |
| Key infrastructure | A signing service holds keys; policies gate signing. | Keys stay off application servers; approvals and audit logs are built in. | Vendor dependency and policy misconfiguration remain your responsibility. |
| Not applicable | The product does not hold or move user funds. | No direct custody risk from the product itself. | You still depend on its data or infrastructure being correct and available. |
Before you deposit
A safety checklist that actually limits loss
Prompts are not security controls. These are the controls that live outside the model and still work when the model is wrong or manipulated.
-
Start with a wallet you can afford to lose
Create a dedicated agent wallet. Never connect a main wallet, and never give an agent a seed phrase.
-
Set a hard spend cap
Per transaction, per day, and per destination. A cap in the prompt is not a cap; enforce it in the signer or smart account.
-
Require simulation before signing
Check the expected balance change, not just the function name. Reject transactions you cannot explain.
-
Use allowlists for contracts and recipients
An agent should only call audited contracts and send to known addresses. Unknown targets need human approval.
-
Treat all external text as hostile
Tweets, web pages, token names, and NFT metadata can contain prompt injection. They are data, never instructions.
-
Keep the key outside the model context
The LLM should request a signature, not possess the secret. Use a signer with policy, not an environment variable in the prompt.
-
Revoke approvals and stop the agent
Set an expiry, monitor approvals, and know how to pause or revoke the agent's permissions before you need to.
-
Check the token, not just the agent
An agent token can trade on hype while the product is unused. Separate product due diligence from token speculation.
What can go wrong
Six risks worth naming
Prompt injection
A malicious post or web page convinces the agent to call a tool it should not. Defend with allowlists and human approval, not better wording.
Malicious approvals
An agent signs an unlimited token approval to a hostile contract. Review approvals and revoke what is no longer needed.
Key compromise
A hot key in an app server, log, or prompt is one leak away from loss. Use a signer with scoped policies.
Confident hallucinations
The agent invents a contract address or misreads a balance. Verify addresses from primary sources, never from the model.
Strategy and market risk
Yield can fall, leverage can liquidate, and backtests can overfit. Agent execution does not remove financial risk.
Token and platform risk
Agent tokens are speculative, and hosted platforms can change fees or fail. Do not confuse a token with equity or a guarantee.
Vocabulary
Glossary for the rest of the page
Short definitions, written for someone who has used a wallet but not an agent.
- AI agent
- Software that observes, plans, and takes actions toward a goal, usually with an LLM in the loop.
- Blockchain agent
- An agent that can read a chain, control a wallet, and submit transactions through onchain tools.
- DeFAI
- Shorthand for DeFi + AI: agents that interact with decentralized finance protocols.
- Agent runtime
- The software that runs the agent loop, memory, model calls, and tool execution.
- Tool / action
- A typed function the model can call, such as swap, lend, bridge, or query a balance.
- Custody
- Who controls the keys and can move the funds. Self-custody, smart account, delegated, or key infrastructure.
- Smart account
- A contract wallet with programmable rules: signers, spend limits, allowlists, and recovery.
- Non-custodial
- The platform does not hold your keys or funds; you approve actions from your own wallet.
- Intent
- A desired outcome, such as 'move 100 USDC to Base', that a solver or agent turns into transactions.
- TEE
- Trusted execution environment: a hardware-isolated area for processing secrets and models.
- Verifiable inference
- A way to prove which model ran and what output it produced, often using proofs or attestation.
- x402
- An open payment protocol built on HTTP 402, used for per-request machine payments.
- Agent token
- A tradable token tied to an agent or its ecosystem. It is not the same as product revenue or equity.
- Prompt injection
- An attack where untrusted text manipulates the model into taking an unintended action.
- Slippage
- The difference between the expected and executed price of a trade.
- Allowlist
- A list of contracts, tokens, or addresses an agent is permitted to interact with.
FAQ
Questions people actually ask
What exactly is a blockchain AI agent?
It is an AI system that can take actions on a blockchain. The AI part plans and decides; the blockchain part gives it a wallet, onchain tools, and a way to settle transactions. A chatbot that only answers questions is not a blockchain agent, even if it talks about crypto.
Do I need to know how to code to use one?
No. Hosted agents and app stores let you connect a wallet and use an agent without code. Coding helps if you want to build a custom agent, control its tools, or audit what it does. The Build category is for developers; the Execute and Launch categories are mostly end-user products.
Who holds the funds when an agent trades?
It depends on the custody model. Self-custody means you hold the key. A smart account means a contract enforces rules while you or your signers approve. Delegated or hosted means a platform operates the wallet. Key infrastructure means a signing service holds keys behind policies. Always check this before depositing.
Can an AI agent lose my money?
Yes. Agents can execute a bad trade, sign a malicious approval, or be manipulated through prompt injection. Smart-contract and market risk exist even when the agent behaves correctly. Use a dedicated wallet, hard spend caps, allowlists, and small initial amounts.
Are agent tokens a good investment?
This page does not give investment advice. An agent token is a tradable asset that may have little relationship to product usage or revenue. Evaluate the product, the team, the token's actual utility, and the downside before treating it as an investment.
What is the difference between a framework and an agent?
A framework is a toolkit for building an agent: ElizaOS, GOAT SDK, AgentKit, and Solana Agent Kit are examples. An agent is a running product built with those tools, such as a yield agent or a trading assistant. Frameworks do not decide your strategy; they give you pieces.
Which chain should I start on?
Start on the chain where the agent and the protocols you need already work. Solana has fast, cheap transactions and a strong agent tooling ecosystem. EVM chains have the deepest DeFi liquidity and the most smart-account tooling. Cross-chain agents add bridge risk, so use them only when the task requires it.
Is this page financial advice?
No. It is an educational directory. Project details change quickly, listings are not endorsements, and nothing here is a recommendation to buy a token or deposit funds. Verify current documentation and terms on the official site before using anything.
Educational, not financial advice.
This page is an editorial guide and directory, not an endorsement or a recommendation to buy
any token or deposit funds. Agent projects change quickly: verify fees, custody, audits, and
availability on the official site before you use them. Never share a seed phrase with an agent,
and never deposit more than you can afford to lose.