TokenTrim

TokenTrim compresses your prompts before they reach ChatGPT, Claude or Gemini. You get the same answer while sending fewer tokens. Most prompts carry words the AI does not need: "Could you please…", "I would like you to…", "make sure that you…", repeated instructions, greetings and thank-yous. TokenTrim removes them, turns requests into direct instructions and restructures long asks into short lists, while keeping every requirement, name, number and example intact. Before: "Could you please analyze this information and provide me with a detailed summary of the most important points, but please make sure that you focus specifically on the things that could potentially impact our marketing strategy going forward?" After: "Summarize the key points, focusing on what could impact our future marketing strategy." HOW IT WORKS A small "✂ ~47 tokens · Optimize" button appears above the prompt box. Click it (or press Alt+O) to see the original and optimized versions side by side, with a word-level view of what changed. Choose Replace to edit further, or Replace & Send to submit right away. Turn on automatic mode and TokenTrim optimizes every prompt as you send it. THREE MODES Light – removes redundancy and filler, keeps your writing style. Balanced – compresses wording and restructures instructions. Max – aggressive savings: numbered lists, no pleasantries, no context the AI infers anyway. TOKEN COUNTER A live estimate of the tokens in your prompt while you type, plus before/after counts and the percentage saved for every optimization. IT LEARNS THE WORDS YOU WANT TO KEEP (new in 0.2) Like to say "please"? Open View changes, click any struck-through word, and TokenTrim never removes it again. Your list stays in your browser and can be edited in Settings. ONE CLICK TO A FASTER MODEL WHEN THE TASK IS SIMPLE (new in 0.2) Rewrites, summaries and translations rarely need a reasoning model, and those models can spend many times the tokens. When TokenTrim sees a simple prompt and a thinking or premium model selected, it offers to switch. On Claude and Gemini one click selects the fast model, with Undo; on ChatGPT it opens the model picker for you. It only acts when you click, never changes your prompt, and can be switched off. SAVINGS TRACKER See how many prompts you optimized and how many tokens you saved today, this month and overall. WHAT IT NEVER TOUCHES Code blocks, inline code, URLs, e-mail addresses and quoted text are always left exactly as written. If a rewrite would not be shorter, your original prompt is kept. PRIVACY Everything runs locally in your browser. Prompt text is never sent to any server. Prompts are not stored. Only aggregate numbers (prompt count, token totals) are saved for the tracker, and you can reset them at any time. No account, no tracking, no ads. SUPPORTED SITES chatgpt.com, claude.ai and gemini.google.com. More sites are on the way. Token counts are estimates: every AI provider uses its own tokenizer, so exact numbers vary slightly. TokenTrim is an independent project and is not affiliated with OpenAI, Anthropic or Google.