Agents
An agent is an intelligent assistant that understands user requests, communicates using LLMs, and performs actions by invoking appropriate tools (exposed as code).
With agents, you can:
- Orchestrate multi-step reasoning and tool use (e.g., answering questions, looking up data, making reservations, sending emails, and more)
- Automatically select and execute the right tools or actions based on user intent
- Maintain conversation context to support dynamic, interactive problem solving
- Adapt behavior, persona, or expertise by simply changing configuration—without changing your Java code
Agents help by acting as an intelligent bridge between users, LLMs, and your application's capabilities. They can automate tasks, provide personalized assistance, and extend what LLMs can do by calling your Java methods or integrating with external systems.
With Ollama4j, creating an agent is as simple as describing its purpose, available tools, behavior, and preferred language model—all defined in a single YAML file.
Why consider building agents using Ollama4j?
- Seamless Customization: Effortlessly fine-tune your agent's personality, expertise, or workflow by editing the YAML—no need to recompile or modify your Java code.
- Plug-and-Play Extensibility: Add new tools or swap out existing logic classes without wrestling with framework internals or glue code.
- Rapid Iteration: Experiment freely. Try different models, instructions, and toolsets to try new behaviors or orchestrations in minutes.
- Clear Separation of Concerns: Keep your core business logic (Java) and conversational configuration (YAML) distinct, promoting clarity, maintainability, and collaboration.
Define an Agent in YAML​
Specify everything about your agent—what LLM it uses, its “personality,” and all callable tools—in a single YAML file.
Agent configuration parameters:
| Field | Description |
|---|---|
name | Name of your agent. |
host | The base URL for your Ollama server (e.g., http://localhost:11434). |
model | The LLM backing your agent (e.g., llama3, gemma, mistral, etc). |
customPrompt | (optional) System prompt—instructions or persona for your agent. |
tools | List of tools the agent can use. Each tool entry describes the name, function, and parameters. |
toolFunctionFQCN | Fully qualified Java class name implementing the tool logic. Must be present on classpath. |
requestTimeoutSeconds | (optional) How long (seconds) to wait for agent replies. |
YAML makes it effortless to configure and tweak your agent’s powers and behavior—no code changes needed!
Example agent YAML:
Implement Tool Functions​
Your agent calls out to Java classes (Tool Functions). Put these implementations on your classpath, register them in YAML.
Instantiating and Running Agents​
Once your agent is described in YAML, bringing it to life in Java takes only a couple of lines:
The API takes care of wiring up LLMs, tool invocation, and instruction handling.
Here's a sample interaction:
Here's another one: