Getting Started with Orion

Orion is a desktop app for testing and iterating on LLM prompts, built by Rebellion Software. Think of it as Postman for language models — a place to organize your prompts, run them against any provider, and track what works. This guide covers the three things you'll do first: create a collection, write and run a prompt, and export your work.

Orion is a desktop app for testing and iterating on LLM prompts, built by Rebellion Software. Think of it as Postman for language models — a place to organize your prompts, run them against any provider, and track what works. This guide covers the three things you'll do first: create a collection, write and run a prompt, and export your work.

Before you begin: add an API key

Orion needs at least one provider API key to run prompts.

  1. Open Settings using the gear icon at the bottom of the sidebar.
  2. Go to the Providers tab.
  3. Click Edit next to the provider you want to use (OpenAI and Anthropic are built in).
  4. Click Set API Key inside the edit form and enter your key.

Your key is stored in your operating system's secure keychain — never in a file on disk.

Once a provider is configured, you're ready.

Part 1: Create your first collection

A collection is a folder that organizes related prompts. Collections map directly to folders on your disk — by default inside ~/orion/. You might use one collection per project, per use case, or per model you're evaluating.

Create a collection

  1. In the sidebar, find the "Collections" heading at the top.
  2. Click the folder-plus icon in the top-right corner of the sidebar.
  3. A name input appears at the top of the sidebar. Type a name — for example, support-drafts or GPT-4 experiments — and press Enter.

The collection appears in the sidebar immediately. Click it to expand it (it'll be empty at first).

Tip: You can rename a collection at any time by hovering over it, clicking the ··· menu, and selecting Rename — or just double-click the collection name directly.

Link an existing folder instead

If you already have a folder of prompts outside ~/orion/, click the link icon (a folder with an arrow) in the sidebar header to link it as an external collection. Orion reads from and saves to that folder without moving anything. External collections appear with a distinct icon in the sidebar.

Part 2: Write and run your first prompt

Open a new prompt

  1. Hover over your collection in the sidebar. A small + button appears on the right side of the row.
  2. Click it. A new tab opens in the editor with a blank prompt ready to fill in.

Fill in the prompt

Every prompt has two text areas:

  • System prompt (optional) — sets the model's behavior, role, or constraints. Example: You are a concise technical writer. Respond in bullet points.
  • User prompt — the actual message you're sending to the model. This is what you're testing.

The system prompt sits above the user prompt. Fill in one or both.

Choose a provider and model

At the top of the editor, there are two dropdowns side by side:

  1. The first selects the provider — for example, OpenAI or Anthropic.
  2. Once you choose a provider, the second shows its available models. Pick the one you want.

If a provider isn't appearing, it may not have an API key saved yet. Go to Settings > Providers to add one.

Run the prompt

Click the Run button (triangle icon) in the editor toolbar.

The response streams into the panel below in real time. When it finishes, you'll see the full response text along with latency and token usage.

Save your prompt

Press Cmd+S (Mac) or Ctrl+S (Windows/Linux), or click the Save button in the toolbar. The prompt is saved as a .json file inside your collection's folder. You can rename it by double-clicking its name in the sidebar.

Part 3: Export a collection

Exporting bundles an entire collection — its metadata and all prompts — into a single portable .orion.json file. Use this to share prompts with a teammate, back up your work, or move a collection to another machine.

Export

  1. Hover over the collection in the sidebar and click the ··· menu.
  2. Select Export collection…
  3. A save dialog opens. Choose a location. The default filename is derived from your collection name — for example, support-drafts.orion.json.

The file contains your collection's name and description, plus every prompt with its full configuration: system prompt, user prompt, model settings, and parameters. API keys are never included in the export.

The file is plain JSON, so it's easy to inspect, version-control, or diff. Its top-level structure looks like this:

{
  "orion": "1.0",
  "collection": { "name": "...", "description": "...", ... },
  "prompts": [ ... ]
}

Import a collection

To load a .orion.json file — one you exported or one a teammate shared:

  1. Scroll to the bottom of the collection list in the sidebar.
  2. Click Import collection…
  3. Select the .orion.json file in the file picker.

Orion creates a new folder in ~/orion/ for the imported collection. If a folder with the same name already exists, it adds a number suffix to avoid overwriting anything.

Note: You can also import a single prompt into an existing collection. Hover over the collection, open the ··· menu, and choose Import prompt… to load an individual prompt .json file.

What's next

With a collection and a working prompt in hand, a few features worth exploring next:

  • Variables — use {{variable_name}} anywhere in your prompts to create reusable templates. Orion fills in the values before each run.
  • Batch runs — attach a CSV file to run a prompt against many inputs at once and see results in a table.
  • Run history — every run is saved automatically. Click the history icon in the response panel to browse past responses for any prompt.
  • Model parameters — adjust temperature, max tokens, top-p, and more from the parameters panel alongside the editor.
  • Custom providers — add any OpenAI-compatible API endpoint in Settings > Providers to test against local models or third-party services.
Back to Getting Started