---
title: "Ollama"
description: "Configure Ollama as a local AI provider for Nanocoder"
sidebar_order: 1
---

# Ollama

[Ollama](https://ollama.com) is a popular tool for running large language models locally. It provides an OpenAI-compatible API out of the box.

## Configuration

```json
{
	"name": "Ollama",
	"baseUrl": "http://localhost:11434/v1",
	"models": ["your-model-name"]
}
```

No API key is required for local use.

## Setup

1. [Install Ollama](https://ollama.com/download)
2. Pull a model: `ollama pull your-model-name`
3. Ollama starts automatically and serves on port `11434`

## Context Length

Ollama's default context window is small (a few thousand tokens, depending on the Ollama version), which is too small for agentic coding. Set the context length as high as your system's memory can handle. A larger context means the model can track more of the conversation history, tool calls, and file contents.

```bash
OLLAMA_CONTEXT_LENGTH=32768 ollama serve
```

Or set it permanently in your environment, or per model with `PARAMETER num_ctx 32768` in a Modelfile.

### Signs of an Insufficient Context Limit

If your context limit is too low, you may notice:

- **Model refuses or fails to use tools** — The model can't include tool definitions and conversation history in its limited context, making tool calling unreliable or broken.
- **Poor memory and reasoning** — The model loses track of recent conversation, forgets what it was working on, or contradicts earlier decisions.
- **Repetitive or looping responses** — The model repeats the same suggestions or asks the same questions because it can't see what it already said.
- **Responses cut off mid-sentence** — Context overflow can cause incomplete or truncated output.
- **Ignoring system instructions** — Critical system prompt content gets pushed out by conversation history, leading to off-topic or misaligned behavior.

## Fetching Available Models

The `/settings providers` wizard can automatically fetch your installed Ollama models when configuring this provider.