> For the complete documentation index, see [llms.txt](https://docs.k3-labs.com/introduction/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.k3-labs.com/introduction/ai-automation-suite/ai-agent.md).
# AI Agent
**AI Agent** is the central intelligence block in K3 Labs workflows, allowing you to connect your favorite AI model, set a custom prompt, and have the AI analyze, summarize, or take actions based on your blockchain data. You can use AI Agent for both simple and advanced automation tasks, making it a flexible tool for all types of users.
***
### How AI Agent Works
**1. Choose Your AI Provider**
You can select from the following AI providers:
* **Gemini (K3 Free Credits)** — Use Gemini for free, sponsored by K3 Labs. You will have access to Gemini 2.0 Flash and Gemini 2.0 Flash Light models.
* **Gemini** — Use your own Gemini API key.
* **OpenAI** — Connect your OpenAI account.
* **Claude** — Connect your Claude (Anthropic) account.
> **Tip:** To connect external providers (OpenAI, Claude, Gemini), go to the Integrations page, select your provider, and securely add your API key as shown below:
All API keys are fully encrypted, and only used for generating responses within your workflows.
**2. Select Your AI Model**
Once you pick a provider, you’ll see all the latest available models (e.g., Gemini 2.0 Flash, GPT-4o, Claude 3 Opus, etc). K3 automatically keeps the list up to date as new models are released.
> **Choosing the Right Model**\
> Each AI model has its strengths and limitations. For example, basic models (like Gemini 2.0 Flash or ChatGPT 4.0) are great for simple, direct tasks—they’ll try a single operation and return a result. More advanced models (like Claude Sonnet 4.0, Gemini 2.5 Pro, or OpenAI GPT-4o) are capable of multi-step reasoning: if they encounter errors, they can adjust, retry queries, and refine their approach until they succeed. This means your workflow results can vary depending on which model you use.
**3. Write Your AI Prompt**
Your prompt tells the AI what to do with your data. Here are two typical ways to use AI Agent:
***
#### **A. Using AI Agent Without Tools (Prompt-Only Mode)**
\|
|
In this mode, the AI only works with the data you pass it (no external tools used).
Example 1: Wallet Balance Summary
Use the “AI Wallet Balance Summary” template to generate a daily summary of your wallet token holdings.
The workflow fetches wallet balances (using the Read Wallet Balances function), sends them to the AI Agent, and emails you a short, human-friendly report.
Example prompt:
You are my financial assistant preparing a daily wallet report.
\[Prompt continues...]
Other use cases:
Transaction screening (“Summarize suspicious transactions from last 24 hours”)
Monitoring NFT activity
Generating portfolio insights
Formatting data for compliance reports
|
\| --------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
***
#### **B. Using AI Agent With Tools (Tool-Enabled Mode)**
This is where AI Agent gets really powerful. When you enable tools, the AI can call external services to fetch, analyze, or act on blockchain and off-chain data—making your workflows much more dynamic.
**How It Works:**
* Choose an AI tool from the dropdown after writing your prompt.
* Supported tools (so far):
* **Web3 Data Fetcher** (Uniblock): Lets the AI fetch on-chain data, such as historical prices, wallet transactions, and more.
* **MCP Server**: Lets the AI call any compatible tool or service configured as an MCP (using SSE endpoint + authentication). Example: connect to The Graph, or any third-party API set up as MCP.
**Example 2: Fetching Blockchain Data**
* Prompt:\
`Show me the historical price of WBTC for May 26th 2024 on Ethereum.`
* Tool:\
`Web3 Data Fetcher`
* Result:\
The AI fetches and summarizes the price, so you can use it further in your workflow.
**More tool use cases:**
* Pulling on-chain analytics or liquidity data
* Running cross-chain comparisons
* Querying specialized services (with MCP)
* Triggering actions based on real-time blockchain events
***
### Want to learn more?
[Learn more about Web3 Data Fetcher.](/introduction/ai-automation-suite/ai-agent/web3-data-fetcher-tool-powered-by-uniblock.md)
[Learn more about configuring MCP tools.](/introduction/ai-automation-suite/mcp-server.md)
***
#### Connecting New Tools
With MCP support, you can connect virtually any external tool or service that supports MCP SSE integration and authentication - opening the door to custom analytics, custom APIs, and more. Just provide your endpoint and credentials, and the AI Agent can use it in your automations.
***
#### What’s Next?
We’re rapidly expanding the list of supported AI tools. You’ll see new pre-configured tools rolling out all the time, and you can already use MCP to connect your own. Stay tuned for updates!