> ## Documentation Index
> Fetch the complete documentation index at: https://docs.agentium.in/llms.txt
> Use this file to discover all available pages before exploring further.

# Ollama on Edge

> Model recommendations, auto-start, and pull helpers for running Ollama on Raspberry Pi.

# Ollama Edge Helpers

Utilities for running Ollama on resource-constrained edge devices — ensure it's running, pull models, and get hardware-appropriate recommendations.

***

## Quick Start

```typescript theme={null}
import { ensureOllama, pullModel, recommendModel } from "@agentium/edge";
import * as os from "node:os";

// Ensure Ollama is running (starts it if needed)
const status = await ensureOllama();
console.log(status);
// { running: true, version: "0.1.0", models: ["tinyllama:1.1b"] }

// Get the best model for available RAM
const ramMb = os.totalmem() / 1024 / 1024;
const rec = recommendModel(ramMb);
console.log(rec);
// { model: "phi3:mini", label: "capable", parameterSize: "3.8B", ramRequired: 4000 }

// Pull the model if not cached
await pullModel(rec.model);
```

***

## Functions

### `ensureOllama(baseUrl?)`

Check if Ollama is running. If not, spawn `ollama serve` and wait up to 10 seconds for it to become available.

```typescript theme={null}
const status = await ensureOllama();
// { running: boolean, version?: string, models?: string[], error?: string }
```

### `checkOllama(baseUrl?)`

Read-only check — does not attempt to start Ollama.

### `pullModel(model, opts?)`

Pull a model from the Ollama registry if not cached locally.

```typescript theme={null}
const result = await pullModel("tinyllama:1.1b");
// { success: true, model: "tinyllama:1.1b" }
```

### `recommendModel(ramMb)`

Pick the largest model that fits comfortably in available RAM (with 30% headroom for OS and Node.js).

| RAM   | Recommended Model | Label    |
| ----- | ----------------- | -------- |
| 1 GB  | tinyllama:1.1b    | fast     |
| 2 GB  | tinyllama:1.1b    | fast     |
| 3 GB  | llama3.2:1b       | balanced |
| 6 GB  | phi3:mini         | capable  |
| 8+ GB | mistral:7b        | powerful |

### `hasModel(model, baseUrl?)`

Check if a specific model is cached locally.

### `listModelTiers()`

Get all known model tiers with their RAM requirements.

***

## Installing Ollama on Raspberry Pi

```bash theme={null}
curl -fsSL https://ollama.com/install.sh | sh
ollama pull tinyllama:1.1b
```

For Pi 4 with 2 GB RAM, `tinyllama:1.1b` is the only viable option. Pi 5 with 8 GB can comfortably run `phi3:mini` (3.8B parameters).