RealityRouterRealityRouter
HomeDocsGitHub

Get started

Quickstart

60-second install + your first routed request

Install RealityRouter, configure your providers, and route your first request — in about 60 seconds.

Requires Docker (recommended) or Python 3.10+. You will need an API key from at least one LLM provider (OpenAI, Anthropic, Gemini, Mistral, DeepSeek, or a local Ollama instance).

1. Install

The fastest way to get started is using the one-line installer.

Linux / macOS

curl -fsSL https://raw.githubusercontent.com/Lars-confi/RealityRouter/main/install.sh | bash

Windows (PowerShell)

Set-ExecutionPolicy Bypass -Scope Process -Force; iex ((New-Object System.Net.WebClient).DownloadString('https://raw.githubusercontent.com/Lars-confi/RealityRouter/main/install.ps1'))

The script will download the latest version, set up the environment, and launch the interactive setup wizard.

2. Run the wizard

The wizard walks you through the configuration of your decision engine:

Step 1 — Authentication

RealityRouter uses Reality Signal™ to estimate model success probabilities. You will be prompted to authenticate via GitHub, Google, or Microsoft SSO to receive your secure calibration token.

Step 2 — Routing Strategy

Pick your default behavior:

  • Expected Utility (Snap) — Single-shot routing to the best model. Lowest latency.
  • Tiered Assessment (Ladder) — Sequential escalation. Starts cheap, moves to smarter models only if validation fails.

Step 3 — Intelligence Coefficients

Fine-tune how the engine weights different factors:

  • α (Cost Sensitivity) — Preference for saving money.
  • β (Time Sensitivity) — Preference for speed/low latency.

Step 4 — Provider Credentials

Enter your API keys for providers like OpenAI, Anthropic, Mistral, and DeepSeek, or provide the URL for a local Ollama instance. The wizard live-validates your keys to ensure they are working before you finish.

Step 5 — Model Visibility

The router auto-discovers all available models based on your keys. Toggle models ON or OFF to define your routing pool. You will also select a cheap, fast model (like Gemini Flash) to act as your Sentiment Analyzer for the feedback loop.

3. Your first routed request

RealityRouter is 100% OpenAI API compatible. Just point your client to your local instance:

import openai

openai.api_base = "http://localhost:8000/v1"
openai.api_key  = "any"   # The router handles upstream auth

response = openai.ChatCompletion.create(
    model="auto",   # Let the router choose the best model
    messages=[
        {"role": "user", "content": "Write a high-performance Rust function to parse JSON."}
    ],
)

print(response.choices[0].message.content)

4. Control Center

Once running, visit the web dashboard to monitor your routing in real-time:

  • Dashboard: http://localhost:8000/metrics/dashboard

Here you can adjust your Cost vs. Speed preferences using live sliders and see exactly how much you are saving compared to using flagship models for every request.

Next