Skip to content

Choose your AI

Scry's assistant is powered by an AI model. You bring your own — Scry doesn't resell AI or route your conversations through a Scry server. There are two ways to provide one, both set up under Settings → AI providers:

Provider Where it runs Cost Choose it when…
OpenRouter (recommended) Cloud Free open-weight models, or pay-per-use for premium ones You want one key that unlocks hundreds of models and the best quality.
Ollama On your own hardware Free You want to run fully offline, on an air-gapped robot, or keep everything on your network.

OpenRouter is the default and the simplest place to start: a single key gives you access to 300+ models — including Claude, GPT, Gemini, Llama, and Qwen — and there's a free tier so you can begin without spending anything.

One API key, 300+ models, a free tier to start. You pick the specific model from the model chip at the top of the chat screen, and you can switch models at any time.

Scry works by calling tools on your robot, so the model you pick must support tool calling. If you attach photos or ask Scry to read a camera frame, it also needs vision. Three good starting points:

For… Model Notes
Best free, general use openai/gpt-oss-120b:free The default. Free, reliable, no credit card. Text only.
Best free, with images google/gemma-4-26b-a4b-it:free Free and multimodal — handles text and images and supports tool calling. Use it when you attach photos or inspect camera frames.
Best paid, deepest reasoning anthropic/claude-haiku-4-5 Fast, inexpensive, excellent at multi-step diagnosis.

Picking your own model

Browse the full catalogue at openrouter.ai/models. Filter to Free, and check that a model lists tools support (and image input, if you need it). Any model that does will work in Scry.

Set it up

  1. In the app, open Settings → AI providers → OpenRouter.
  2. Tap Sign up to create a free OpenRouter account at openrouter.ai. No credit card is needed for the free tier.
  3. Tap Get API key, create a key at openrouter.ai/keys, and copy it.
  4. Paste the key into OpenRouter API key and save.
  5. Open the model chip at the top of the chat screen and pick a model — start with the free default above.

Option B — Ollama (local, offline)

Ollama runs AI models on your own machine — the robot itself or any computer on the same network. Nothing leaves your network, there's no API key, and it's free. The trade-off is that local models generally reason less sharply than the cloud options and may occasionally need a retry, but they're a great fit for offline or air-gapped robots.

Set it up

  1. Install Ollama on the robot or any host on your network — see ollama.com/download.

  2. Download a model that supports tool calling (add a vision model if you want image support):

    ollama pull qwen2.5:7b      # good general model with tool support
ollama pull qwen2.5-vl      # optional: adds image/vision support

  3. Make Ollama reachable on your network, not just locally:

    OLLAMA_HOST=0.0.0.0:11434 ollama serve

  4. In the app, open Settings → AI providers → Ollama. Leave Base URL blank to let Scry find Ollama automatically on your network (it checks the robot's host on port 11434 first), or enter the address yourself, e.g. http://192.168.1.50:11434.

  5. Pick the model you downloaded from the chat model chip.

Switching models

You can change provider or model at any time from the model chip at the top of the chat screen — even mid-conversation. The choice is remembered on your device. Your API keys are stored encrypted on the phone and are only ever sent to the provider you chose.