AgentKit (Coinbase Developer Platform's AI Agent Crypto Wallet Toolkit)
Every AI agent deserves a wallet — enable onchain interactions, fee-free stablecoin payments, and monetization.
Evidence shows: README mentions CDP Secret API Key and OpenAI API Key, implying sensitive data handling; SECURITY.md and HackerOne reporting channel exist; CodeQL workflow present. However, no explicit least-privilege design, user confirmation mechanism, data flow transparency, dependency security audit, external effect control, or rollback mechanism found. Deductions: these aspects lack concrete implementation or documentation.
Evidence shows: README is well-structured, quickstart examples are consistent; CI workflow (format.yml) and test files (conftest.py) exist. However, no dependency availability guarantees (e.g., lock files, mirrors) or detailed failure message design found. Deductions: dependency availability and failure messages lack evidence.
Evidence shows: README explicitly supports multiple frameworks (Langchain, Vercel AI SDK, MCP, etc.) and multiple wallets (CDP, Privy, Viem), and lists several example scenarios; action-providers and wallet-providers directories indicate clear capability boundaries. However, trigger precision (e.g., action trigger conditions) is not detailed. Deductions: trigger precision lacks documentation.
Evidence shows: README provides installation instructions, repository structure, examples, license (Apache-2.0), security reporting channel; references CONTRIBUTING.md and WISHLIST.md. However, no explicit version changelog or known limitations section found. Deductions: versioning changelog and known limitations missing.
Evidence shows: README provides quickstart examples, output is actionable transaction links; as a framework, it offers multiple integrations, high marginal value. However, cost-benefit is not quantified, and no performance or resource consumption data provided. Deductions: cost-benefit lacks data support.
Evidence shows: README claims (e.g., support for multiple protocols) are supported by corresponding directories and files; test files exist. However, no cross-source corroboration or clear separation of facts and inferences found. Deductions: cross-source corroboration and fact-inference separation insufficient.
- No explicit least-privilege design found; agents may have excessive permissions.
- No user confirmation mechanism found; agents may execute sensitive actions automatically.
- No data flow transparency documentation found; users may not understand how data is used.
- No dependency security audit found; supply chain risks exist.
- No rollback mechanism found; erroneous actions may be hard to undo.
- No version changelog found; users may find it hard to track updates.
- No known limitations section found; users may be unaware of risks.
What does this agent do, and when should you use it?
AgentKit is a toolkit from Coinbase Developer Platform that gives AI agents a crypto wallet and onchain interaction capabilities. It is framework-agnostic and wallet-agnostic, supporting TypeScript and Python. It integrates with AI frameworks like LangChain, Vercel AI SDK, OpenAI Agents SDK, and more, and wallet providers like Coinbase, Privy, and viem. The repository includes core libraries, scaffolding tools (create-onchain-agent), framework extensions, and comprehensive examples, offering 50+ (TypeScript) and 30+ (Python) pre-built action providers covering DeFi, NFT, social, and other scenarios. All agent actions are executed via wallet providers, producing onchain transactions. It runs on networks like Base, Ethereum, and Solana. The project actively welcomes community contributions.
AgentKit provides a set of framework-agnostic libraries that allow AI agents to operate crypto wallets and execute onchain transactions. Key components include: core libraries (@coinbase/agentkit for TypeScript, coinbase-agentkit for Python) that include action-providers (pre-built actions) and wallet-providers (wallet adapters for cdp, privy, viem). Framework extensions support LangChain, Vercel AI SDK, Model Context Protocol, OpenAI Agents SDK, Autogen, Pydantic AI, etc. Example projects (e.g., langchain-cdp-chatbot, openai-agents-sdk-smart-wallet-chatbot) demonstrate usage. The typical flow: a user gives a natural language command (e.g., 'Fund my wallet with some testnet ETH'), the agent uses a wallet provider to create/manage a wallet, performs an action (e.g., request testnet ETH from faucet), and submits a transaction onchain, returning a transaction hash and block explorer link.
- Developers want to add crypto wallet functionality to their LangChain-based AI agents for onchain operations.
- Teams using OpenAI Agents SDK need to integrate Coinbase wallet and stablecoin payments.
- Individual developers want to quickly bootstrap a full-stack onchain agent project using the create-onchain-agent scaffold.
- Developers who want to add social features (e.g., Farcaster, Twitter) or DeFi operations (e.g., Compound, Morpho) to their agents.
- Developers building smart wallet functionality can refer to the smart-wallet example projects.
What are this agent's strengths and limitations?
- Framework-agnostic, supports major AI frameworks like LangChain, Vercel AI SDK, OpenAI Agents SDK
- Wallet-agnostic, supports multiple wallet providers like Coinbase, Privy, viem
- Provides 50+ pre-built action providers covering DeFi, NFT, social, etc.
- Rich set of examples and scaffolding tools lower the entry barrier
- Supports fee-free stablecoin payments and agent monetization
- Heavily depends on Coinbase Developer Platform, requiring CDP API keys
- Some features may only support specific networks (e.g., Base, Ethereum, Solana)
- Experimental software provided 'AS-IS', with legal and privacy risks
- For non-Coinbase ecosystem users, additional adaptation may be needed
- Documentation is developer-oriented, not beginner-friendly
How do you install or deploy this agent?
For Node.js, ensure you have Node.js 22+, a CDP Secret API Key, and an OpenAI API Key. Run npm create onchain-agent@latest to create a new project, navigate to the project directory, configure the .env file with your API keys, then npm install to install dependencies. For Python, require Python 3.10+, Poetry, and the same API keys. Use pipx run create-onchain-agent to create a project, set environment variables, then poetry install to install dependencies.
How do you use this agent?
After installation, run npm run dev (Node.js) or poetry run python chatbot.py (Python) to start the development server or chatbot. Then issue natural language commands to the agent, e.g., 'Fund my wallet with some testnet ETH.' The agent will execute and return a transaction link. See the README quickstart for detailed examples.