Tips for Token Saving

event

Remember this formula first: Token Cost = Input Size × Number of Calls.

What you spend each month comes down to two things： how much content you send in each time, and how many times you call the AI in total.

1. Control Input Size Per Call (Most Effective)

1. Only Show the AI What It Needs to See

Writing chapter 47? No need to load chapter 1. Polishing a dialogue? No need to read the whole chapter. How: Use @ to reference files precisely instead of opening everything by default. Select only the target paragraph when making edits. Close references you're done with. Estimated savings: 40–60%

2. Keep Your Rules Short

Rules are force-loaded every session and are a hidden major cost. Cut: Roleplay-style preambles (the AI already knows it's an AI). Trim: Convert paragraphs to bullet points — same information, half the tokens. Regularly clean out rules you no longer need. Estimated savings: 15–30%

3. Don't Add Background to Instructions

❌ Token-heavy: "You are a professional novel writing assistant. Please help me rewrite this dialogue to be more tense, so the reader can feel the conflict between the two characters…" ✅ Token-light: "Rewrite dialogue: increase conflict tension, maintain tone" Save frequently used instructions as Skills and call them with one click — zero description overhead. Estimated savings: 20–35%

2. Reduce Number of Calls

4. Align First, Then Generate

The most wasteful approach: ask the AI to write 2,000 words → not satisfied → redo → redo again. The right approach: Generate a chapter outline first (very few tokens) → confirm direction → expand into full prose → fine-tune tone at the end. Use Plan mode to discuss direction — a few rounds of discussion costs far fewer tokens than one full rewrite. Estimated savings: 30–50%

5. Ask the AI to Edit, Not Rewrite

Without clear instructions, the AI tends to regenerate the entire passage. ❌ "Help me improve this text" ✅ "Edit only the third paragraph — slow down the pacing. Output the revised paragraph only, no explanation needed." Adding "no summary needed" saves even more. Estimated savings: 20–40%

6. One Window, One Task

The longer a conversation, the longer the history — and the larger the input for every new message. Finish a chapter, open a new window for the next. Edit a paragraph, open a new window for the next. One task per window. Estimated savings: 10–30%

In a Nutshell

Control context scope + reduce redundant generation. Get these two things right and your monthly bill drops by at least half.

Advanced Tips

  • Turn high-frequency workflows into a Workflow template — just fill in the chapter number each time and let it auto-assemble.
  • With 16GB+ RAM, install a local model (LM Studio) for drafting, and only send polished work to the cloud — reduces cloud token usage by 60%+.
Tips for Token Saving - SoloEnt