Skip to content

Ranger's AI Provider (or model engine)

Ranger’s ability to understand natural language and respond intelligently is powered by it's AI Provider — it's LLM backend.

By default, Ranger uses Ollama for this role — an open-source model server that can run large language models (LLMs) locally. But the AI Provider is not tied to Ollama: it’s designed to support any compatible LLM backend, including OpenAI, Claude, open-weight models via vLLM, or even internal fine-tuned APIs.

Why Use Ollama?

Ollama was chosen as the default for several reasons:

  • Local-first: It runs entirely on-prem without internet.
  • Supports many models: LLaMA, Mistral, Phi, and others.
  • Hardware optimized: Runs efficiently on CPUs, Apple Silicon, or GPUs.
  • Simple interface: Exposes a REST API (default: port 11434) that Ranger can talk to.
  • Air-gapped friendly: No reliance on cloud APIs or license tokens.

It’s fast, flexible, and easy to switch models or versions depending on your hardware or needs.

Note

We are always looking to improve Ranger's AI capabilities. If you have a preferred LLM backend or model that you think would enhance Ranger, please open an issue.

How the AI Provider Works

When a user sends a message — whether via chat UI, n8n, or CLI — the flow looks like this:

  1. Prompt comes in from Open-WebUI (or another frontend).
  2. Ranger intercepts it via its API, instead of letting the frontend talk to Ollama directly.
  3. Ranger evaluates the prompt:
  4. If it’s a language-only query (e.g. “Can you generate a python 101 module for data science?”), it forwards it to Ollama and returns the answer.
  5. If it’s an actionable command (e.g. “Start 3 NPC agents”), Ranger interprets and executes it using the appropriate GHOSTS workflow.
  6. Ranger returns the final result to the user.

This proxy architecture ensures all requests pass through Ranger, allowing it to mediate, enrich, or act on them intelligently.

Swapping Out the Brain

Ranger isn’t married to Ollama. It can talk to:

  • Any OpenAI-compatible API (e.g. OpenAI, Azure, Groq)
  • Claude (via a proxy or plugin)
  • Local APIs you host yourself
  • Ollama, with any model you install

To switch models in Ollama:

ollama list
ollama pull mistral:latest
ollama run mistral

Or use Open-WebUI’s model selector if running interactively.

You can even point Ranger at different model endpoints using environment variables:

LLM_API_BASE=http://ollama:11434
LLM_MODEL=llama2

Isolation by Design

A key design decision: LLM execution happens outside Ranger.

Ranger never loads models into its own process. Instead, it treats the LLM like an external compute resource — asking questions, getting completions, and deciding what to do next.

This gives you:

  • Fault isolation (LLM crashes don’t affect Ranger logic)
  • Better performance (no model bloat inside the API)
  • Deployment flexibility (you can swap the model backend at any time)

TL;DR

Ranger’s AI Provider is the pluggable AI layer that handles language understanding. By default it uses Ollama, but it’s backend-agnostic. This keeps Ranger flexible, lightweight, and adaptable to your environment — whether you’re using open models, fine-tunes, or proprietary APIs.

Want to plug in your own brain? Just point Ranger at it.