Make every token go further: a practical guide for Orion™ and Sirius™

Most token savings don't come from technical expertise. They come from a handful of habits in how you write, attach files, and carry on a conversation with an AI assistant. This guide gathers the ones that make the biggest difference day to day, and it applies equally to Orion™ and Sirius™. None of them requires giving up quality. If anything, the opposite is true, because a clear, focused message usually gets you a better answer.

What a token is and what uses them

A token is a fragment of text: a short word, a syllable, or a punctuation mark. As a rough guide, a word typically works out to between one and two tokens, so a 100-word paragraph runs about 150. AI assistants measure their work in that unit, and every Orion™ and Sirius™ plan includes a monthly allowance.

What uses tokens isn't just the question you type. Four things count toward the total:

1. Your message, including anything you paste into it.
2. Any files and images you attach.
3. The conversation history, which the assistant rereads on every turn to keep track of the thread.
4. The response you get back.

That's why the biggest savings rarely come from cutting polite phrases. They come from being deliberate about what information goes along with each turn, and that's what the rest of this guide is about.

Prompts: clarity over length

A good message isn't short or long; it's precise. The difference shows up mostly in the number of back-and-forth exchanges. A vague instruction forces you to clarify, correct, and ask again, and every round adds tokens.

Four details are enough for the assistant to get it right the first time: what you need, who it's for, what format you want, and how long it should be. Compare these two messages:

• Before: I need help with an email to a customer who complained about a late delivery. Can you help me write something?

• After: Write an email to a customer whose order arrived five days late. Keep the tone friendly and professional, 120 words max, with a brief apology and a 10% discount on their next purchase.

The second message is longer, but it will almost always get the job done in one try. The first usually takes two or three rounds of adjustments.

A few other habits help:

1. Group related questions. Every additional message makes the assistant process the history again, so three questions on the same topic in a single message save you two of those rereads.
2. Fix only what needs fixing. If one paragraph isn't working, ask for “rewrite only the second paragraph with a more direct tone” instead of requesting the whole text again.
3. Include an example when format matters. A two-line sample explains a style better than a paragraph of description.

Conversations: history counts too

Think of a conversation as a desk: the more papers pile up, the longer it takes to go through them all. In most assistants, each new message travels along with everything that came before it, so a long chat costs more on every turn than a fresh one.

Three simple habits keep that weight in check:

1. Start a new chat when you change topics. Orion™ and Sirius™ keep your conversation history, so nothing is lost. You can go back to any of your chats whenever you need to.
2. Summarize before you pick up a long project again. If a project stretches over several days, ask for a summary and use it to open the next chat.
3. Give context just once. At the start, say who you are, what the project is about, and the tone you want. The assistant will keep it in mind without you repeating it.

For the summary, an instruction like this is all you need:

• Summarize in six lines the decisions made and the open items from this conversation so I can continue it in a new chat.

Each assistant handles memory in its own way. Orion™ is built for quick, agile conversations and manages context efficiently; Sirius™ works with a longer context. That reach is an advantage when the work calls for it, such as analyzing a long document. When it doesn't, it just adds weight to every turn.

Files and documents: send only what's needed

Attaching an entire document is convenient, but it's rarely necessary. Most questions can be answered with just part of the material, and the assistant does its best work when it receives exactly that part.

1. Trim before you send. If you're asking about one clause in a contract, share that clause and the definitions that affect it, not all forty pages.
2. Clean up the text. Headers, footers, repeated legal notices, and signatures take up space without adding anything. If you're copying from a multi-column PDF, check that the text doesn't come through scrambled. Clean text is easier to interpret, too.
3. Work in stages with long documents. Ask for an outline or a summary of the structure first, then dig into only the sections that matter.
4. Be selective with spreadsheet columns and rows. If you're after a formula or a structure, the headers and about ten sample rows are enough.
5. Narrow down code and error logs. Share the relevant function and the lines around the first error, not the whole file or the entire log. Sirius™ is built to generate and debug code and to interpret error logs, and it performs even better when it gets precise input.

Here's a simple test: imagine handing the material to a colleague. If they wouldn't need to read all of it to answer you, there's no need to send all of it.

Images and screenshots

An image can explain in seconds what would take a paragraph, which is why it's worth deciding when to send one. In general, the larger and more detailed an image is, the more tokens it uses.

1. If the image is just text, copy the text. Pasting it is usually cheaper and more accurate than asking the assistant to read it from a screenshot.
2. Crop to the relevant area. An error window says more than a whole desktop, and it's lighter.
3. Don't go for maximum resolution. It's enough for the content to be clearly legible.
4. Pair the image with a specific question. “What does the red line on the chart indicate?” works better than “analyze this image.”
5. Send only images that add something different. Three nearly identical screenshots cost three times as much and add no new information.

Orion™ offers fast image analysis, ideal for one-off questions. Sirius™ adds more advanced vision capabilities, which come in handy for screenshots or charts with a lot of detail.

Responses: ask for what you'll actually use

The response uses tokens, too, and it takes time to read. A response that runs longer than necessary doesn't add value, just more text to review.

1. Specify the format and length. “In five bullet points,” “in one paragraph,” or “in a three-column table” head off explanations you didn't ask for.
2. Ask it to lead with the conclusion. With an instruction like “give me the recommendation first, and the details only if I ask,” you get the essentials right away and decide whether you want more.
3. Break long texts into two steps. Ask for an outline first, adjust it, and then ask for the full draft. That way you avoid throwing away a long text that started from a misunderstood idea.

That last practice is probably the one that saves the most: reviewing a ten-line outline costs far less than correcting a three-page document.

Which assistant fits which task

Choosing the right assistant is also a way to save. A simple task doesn't need more horsepower than it requires, and a complex task gets solved in fewer attempts with the assistant that's built for it. As a guide:

As for the monthly allowance, Orion™ plans range from 100,000 to 600,000 tokens depending on the tier, and Sirius™ plans from 500,000 to 3 million. With the habits in this guide, that allowance goes further.

Five questions to ask before you hit send

With practice, these checks become second nature. In the meantime, a ten-second review before you send is enough:

1. Does my message say what I need, in what format, and how long it should be?
2. Is the conversation still about the same topic?
3. Am I sending only the part of the file that's needed?
4. Can I crop the image or replace it with text?
5. Is this the right assistant for the task?

Saving tokens, at its core, comes down to giving the assistant just what it needs to help you well. If you'd like to put these ideas to the test, you can check out Orion™ and Sirius™ in our store, and if you have questions about which one fits your work best, our team is happy to help.