AI Agents in Crypto: What They Are, What They Can Do, How to Judge Them
An AI agent in crypto is a language model connected to tools, such as price data, a wallet or a trading service, so it can take actions as well as answer questions. What matters is what it is allowed to do: which tools it has, who holds the keys, and what limits sit outside the model.
What AI agents in crypto are
An AI agent in crypto is a model, a set of tools, and a set of permissions. The model writes and reads language. The tools fetch prices, query wallets, or submit transactions. The permissions decide which of those tools exist. Where it runs can be your laptop, a hosted app, or a chat product. A scripted bot follows a fixed rule. An agent chooses the next tool call. That choice is the whole difference, and it is why the permissions matter more than the model name.
If the only limit is a sentence in the prompt, you do not have a limit. You have a hope. The rest of this page is about putting the limit somewhere the model cannot edit.
Kinds of crypto AI agents
Trading agents place or prepare orders. Research and alert agents read and message you, and they should not have a signer. Social agents post. Launchpad or agent token projects attach a token to an agent and sell the token. DeFi automation moves funds inside a protocol according to a contract. These are categories, not a ranking, and this page names no agent token project.
Owning a token that is tied to an agent does not usually give you a claim on anything the agent earns, unless that project's documents say so in plain language. A ticker is not a share of the fees. Read the documents or skip the token. This page will not list coins.
What a crypto AI agent can and cannot do
It can read data you give it tools for, summarize what it read, and place a transaction if a tool allows that and the limits allow it. No AI can predict markets. It cannot be trusted with unlimited permissions, because a model will follow a bad instruction, a hidden instruction, or a stale number with the same confidence as a good one.
Worked example: three ways to give an agent access
Example, not a recommendation. The rows are setups people actually use. The worst case column is the realistic mistake, not a horror story invented to sell a product.
| Setup | What the agent holds | Can it send funds out | What limits exist | Worst realistic mistake |
|---|---|---|---|---|
| Raw wallet key | The private key | Yes, anything the key can sign | Only the prompt | It sends the balance where a message told it to |
| Exchange key, withdrawals off | A trade only API key | No, if the setting holds | The exchange's key settings | Bad trades up to the balance that key can trade |
| Moby Agent | Its own wallet and a Moby agent key | No withdraw, transfer or send tool | Your caps, plus a red token block on buys | Buys inside your limits that lose value |
Row one is how agents lose funds people did not mean to send. This page does not restate a February 2026 dollar figure, because the news pages for that incident did not load on November 22, 2026. The controls that matter, including what a missing send tool changes, are on AI agent security. Row three, in words: One agent per user, with its own wallet. Agents trade Solana tokens only. The agent cannot withdraw, transfer or send funds. You set your agent's per-trade and daily spending limits and market-cap range; Moby blocks red-rated tokens. Limits are only as tight as you set them. Sells and exits skip the market-cap and safety checks, and tokens with no safety rating are not blocked. Rate limits protect against runaway loops and scale with the fees your agent pays; you set the spend limits.
How to judge an AI agent before you trust it
Ask who holds the keys, and expect a name, not a vibe. Ask for the tool list, in writing. Ask which limits sit outside the model, and try to break one with a prompt. If a sentence overrides a cap, the cap is fake. Ask whether you can pause it, and whether there is a log of tool calls. Ask who built it and whether you can find them. If a token is attached, ask whether any document gives you a right to fees or to the treasury. If someone promises returns, stop. The CFTC's January 25, 2024 advisory, Release 8854-24, checked November 22, 2026, says claims of high or guaranteed returns are red flags of fraud.
AI agents for crypto trading
Trading is the case where the tool list includes an order. The frame is agentic trading, and the bot comparison is AI trading bot. Supported hosts: Codex, Claude Code (CLI), Gemini CLI (with a paid Gemini API key or enterprise access) and Grok CLI. ChatGPT: works on Business, Enterprise and Edu workspaces that allow custom MCP servers (web). Moby does not run a schedule for your agent; the schedule lives in your AI client. Your resting orders (limit, take profit, stop loss, trailing stop, DCA) keep working on Moby even when your app is closed. Turn on Ask before every trade and your AI asks you first. Moby does not hold or approve trades. A schedule in your chat app is not the same thing as a resting order. The orders keep working when the chat is closed. The model does not. Rule bots that are not agents are on crypto trading bot.
A Solana specific map of frameworks is on Solana AI agent, and the finance wide version of the same question is on AI agents in finance. The protocol notes, if you are comparing tool servers, are on MCP for trading. None of those pages is a list of agent tokens. If a search result is a list of coins with prices, it is answering a different question than this one. Come back to the checklist: keys, tools, limits, pause, log, and a named builder.
How to build a crypto AI agent
Three build paths, depending on how close you want to be to the code. A general agent loop is build an AI agent for trading. A Solana path is build a Solana trading agent. A tested Python backtest, before any agent, is build a trading bot in Python. Build the refusal path before the buy path.
Risks
Prompt injection, a tool that is not the tool you think, a leaked key, a loop that buys until the balance is gone, and data that was faked. Two longer pages carry the detail: AI agent security and AI agent guardrails. Payments are a different product from trading. Moby does not do payments. The contrast is on agentic payments versus agentic trading.
A Moby Agent, briefly
One agent per user, with its own wallet. Agents trade Solana tokens only. The agent cannot withdraw, transfer or send funds. You set your agent's per-trade and daily spending limits and market-cap range; Moby blocks red-rated tokens. Limits are only as tight as you set them. Sells and exits skip the market-cap and safety checks, and tokens with no safety rating are not blocked. A 0.75% swap fee (minimum $0.90 per swap), plus onchain fees. AI Trading and copy trading currently need an access pass in the app. Download Moby. Setup is how to set up agentic trading. Plans are on agent strategies. The wallet page is agent wallet.
Pause is part of the design
A project that cannot tell you how to stop new trades is a project that has not finished the product. Ask for the pause control before you ask for the feature list. Then ask what pause does not do. Open positions usually stay open. A pause that also promised to exit everything at a good price is a second claim, and it needs a test, not a slogan. Revoking a key is the stronger stop, because the model can no longer call the tool. Do that drill once with a tiny balance before the balance is not tiny. Write down the result of the drill so the pause is a memory, not a menu item you have never used.
Frequently asked questions
What are AI agents in crypto?
They are language models connected to tools that can read crypto data or submit transactions. The model chooses a tool. The permissions decide whether that choice can move money. A prompt is not a permission. Read the tool list before you fund anything.
Can an AI agent have its own crypto wallet?
Yes. How much it can do depends on who holds the keys. One agent per user, with its own wallet. Agents trade Solana tokens only. The agent cannot withdraw, transfer or send funds. Other setups hand the agent a raw key, which is a different risk. The comparison is on agent wallet.
Are AI agent tokens the same as AI agents?
No. An agent is software with tools. An agent token is a token someone attached to that story. Owning the token does not usually give you a claim on fees or on a treasury unless the documents say so. This page does not list tokens.
Can an AI agent trade crypto for me?
It can if a tool can place orders and you accept the limits around that tool. No AI can predict markets. On Moby the agent trades Solana tokens only, from its own wallet, and it cannot withdraw. You can require it to ask you first.
Is it safe to give an AI agent my private key?
No. A private key can sign anything the wallet can do, including a transfer. Use a scoped key or a setup with no withdraw tool. If a product asks you to paste a seed phrase into a chat, stop and use a different product.
Can an AI agent send my crypto by mistake?
Yes, if it has a tool that can transfer. A message can tell it to. A Moby Agent has no withdraw, transfer or send tool, so that particular mistake is not available. Buys inside your limits can still lose money.
How do I judge a crypto AI agent project?
Check who holds keys, which tools exist, whether limits live outside the model, whether you can pause it, and whether a log exists. If a token is attached, read whether it grants any right. Promised returns are a reason to walk away, not a feature.