---
title: Local Ollama
description: Running NIKI 100% offline with local GPU inference.
sidebar:
  order: 4
  label: Local Ollama
---

## Local Offline Inference with Ollama

Run NIKI completely offline without sending any code or prompts to external cloud providers.

---

## 1. Start Ollama

Ensure Ollama is running locally:

```bash
ollama serve
ollama pull qwen2.5-coder:32b
```

---

## 2. Configure `niki.toml`

Point provider configuration to your local Ollama endpoint:

```toml
[providers.ollama]
base_url      = "http://localhost:11434/v1"
api_key       = "ollama"  # placeholder string required by client
default_model = "qwen2.5-coder:32b"

[agents.planner]
provider = "ollama"
model    = "qwen2.5-coder:32b"

[agents.coder]
provider = "ollama"
model    = "qwen2.5-coder:32b"

[agents.tester]
provider = "ollama"
model    = "qwen2.5-coder:32b"

[agents.reviewer]
provider = "ollama"
model    = "qwen2.5-coder:32b"
```