Ollama (Local Models)

Run LLMs locally with complete privacy - no API keys, no internet, no costs.

Overview

┌─────────────────────────────────────────────────────────────────────┐
│                         Ollama Local Setup                          │
├─────────────────────────────────────────────────────────────────────┤
│                                                                     │
│   Your Machine                                                      │
│   ┌─────────────────────────────────────────────────────────────┐  │
│   │                                                             │  │
│   │   ┌──────────────┐      ┌──────────────┐                   │  │
│   │   │  ADK-Rust    │ ───▶ │   Ollama     │                   │  │
│   │   │  Agent       │      │   Server     │                   │  │
│   │   └──────────────┘      └──────┬───────┘                   │  │
│   │                                │                            │  │
│   │                         ┌──────▼───────┐                   │  │
│   │                         │  Local LLM   │                   │  │
│   │                         │  (llama3.2)  │                   │  │
│   │                         └──────────────┘                   │  │
│   │                                                             │  │
│   └─────────────────────────────────────────────────────────────┘  │
│                                                                     │
│   🔒 100% Private - Data never leaves your machine                 │
│                                                                     │
└─────────────────────────────────────────────────────────────────────┘

Why Ollama?

---

Step 1: Install Ollama

macOS

brew install ollama

Linux

curl -fsSL https://ollama.com/install.sh | sh

Windows

Download from ollama.com

---

Step 2: Start the Server

ollama serve

You should see:

Couldn't find '/Users/you/.ollama/id_ed25519'. Generating new private key.
Your new public key is: ssh-ed25519 AAAA...
time=2024-01-05T12:00:00.000Z level=INFO source=server.go msg="Listening on 127.0.0.1:11434"

---

Step 3: Pull a Model

In a new terminal:

# Recommended starter model (3B parameters, fast)
ollama pull llama3.2

# Other popular models
ollama pull qwen2.5:7b    # Excellent tool calling
ollama pull mistral       # Good for code
ollama pull codellama     # Code generation
ollama pull gemma2        # Google's efficient model

---

Step 4: Add to Your Project

[dependencies]
adk-model = { version = "0.2", features = ["ollama"] }

---

Step 5: Use in Code

use adk_model::ollama::{OllamaModel, OllamaConfig};
use adk_agent::LlmAgentBuilder;
use std::sync::Arc;

#[tokio::main]
async fn main() -> anyhow::Result<()> {
    // No API key needed!
    let model = OllamaModel::new(OllamaConfig::new("llama3.2"))?;

    let agent = LlmAgentBuilder::new("local_assistant")
        .instruction("You are a helpful assistant running locally.")
        .model(Arc::new(model))
        .build()?;

    // Use the agent...
    Ok(())
}

---

Complete Working Example

use adk_rust::prelude::*;
use adk_rust::Launcher;
use std::sync::Arc;

#[tokio::main]
async fn main() -> anyhow::Result<()> {
    dotenvy::dotenv().ok();
    
    // No API key needed!
    let model = OllamaModel::new(OllamaConfig::new("llama3.2"))?;

    let agent = LlmAgentBuilder::new("ollama_assistant")
        .description("Ollama-powered local assistant")
        .instruction("You are a helpful assistant running locally via Ollama. Be concise.")
        .model(Arc::new(model))
        .build()?;

    // Run interactive session
    Launcher::new(Arc::new(agent)).run().await?;
    
    Ok(())
}

Cargo.toml

[dependencies]
adk-rust = { version = "0.2", features = ["cli", "ollama"] }
tokio = { version = "1", features = ["full"] }
dotenvy = "0.15"
anyhow = "1.0"

---

Configuration Options

use adk_model::ollama::{OllamaModel, OllamaConfig};

let config = OllamaConfig::new("llama3.2")
    .with_base_url("http://localhost:11434")  // Custom server URL
    .with_temperature(0.7)                     // Creativity (0.0-1.0)
    .with_max_tokens(2048);                    // Max response length

let model = OllamaModel::new(config)?;

---

Recommended Models

Choosing a Model

• Limited RAM (8GB)? → llama3.2 (3B)
• Need tool calling? → qwen2.5:7b
• Writing code? → codellama or mistral
• Best quality? → llama3.1:70b (needs 48GB+ RAM)

---

Tool Calling with Ollama

Ollama supports function calling with compatible models:

use adk_model::ollama::{OllamaModel, OllamaConfig};
use adk_agent::LlmAgentBuilder;
use adk_tool::FunctionTool;
use std::sync::Arc;

// qwen2.5 has excellent tool calling support
let model = OllamaModel::new(OllamaConfig::new("qwen2.5:7b"))?;

let weather_tool = Arc::new(FunctionTool::new(
    "get_weather",
    "Get weather for a location",
    |_ctx, args| async move {
        let location = args.get("location").and_then(|v| v.as_str()).unwrap_or("unknown");
        Ok(serde_json::json!({
            "location": location,
            "temperature": "72°F",
            "condition": "Sunny"
        }))
    },
));

let agent = LlmAgentBuilder::new("weather_assistant")
    .instruction("Help users check the weather.")
    .model(Arc::new(model))
    .tool(weather_tool)
    .build()?;

Note: Tool calling uses non-streaming mode for reliability with local models.

---

Example Output

👤 User: Hello! What can you do?

🤖 Ollama (llama3.2): Hello! I'm a local AI assistant running on your 
machine. I can help with:
- Answering questions
- Writing and editing text
- Explaining concepts
- Basic coding help

All completely private - nothing leaves your computer!

---

Troubleshooting

"Connection refused"

# Make sure Ollama is running
ollama serve

"Model not found"

# Pull the model first
ollama pull llama3.2

Slow responses

• Use a smaller model (llama3.2 instead of llama3.1:70b)
• Close other applications to free RAM
• Consider GPU acceleration if available

Check available models

ollama list

---

Running Examples

# From the official_docs_examples folder
cd official_docs_examples/models/providers_test
cargo run --bin ollama_example

---

Related

• Model Providers - Cloud-based LLM providers
• Local Models (mistral.rs) - Native Rust inference

---

Previous: ← Model Providers | Next: Local Models (mistral.rs) →