Your AI Is Reading 1,000 Pages to Answer One Question
You have a stack of PDFs. A new SSM guideline. A supplier contract. A tax circular from LHDN. You open Claude Desktop, drag the file in, and ask a single question: “What are the filing requirements for Form 13?”
The answer comes back. Great. But here is the hidden inefficiency most business owners miss. The moment you ask a follow-up question—or start a new chat on the same document—your AI reads the entire document again from scratch. Every page. Every paragraph. Every single token.
For a Malaysian SME owner, this “token tax” quietly drains your AI resources. You waste time waiting. You risk your sensitive business data floating around a foreign server. And worst of all, you never quite trust the answer because the AI might have hallucinated a clause from a different section.
TL;DR — The One-Minute Fix
A new open-source tool called Token Saver acts as a local brain for Claude Desktop. It keeps your PDFs on your machine, searches them intelligently using a mix of keyword matching and semantic understanding, and feeds the AI only the specific paragraphs relevant to your question. The result? You save up to 99% of the token waste, your data never leaves your computer, and your answers come with exact page citations you can verify.
What “Token Saver” Actually Means for Your Workflow
Right now, when you talk to an LLM about a large PDF, the entire text (often 1,500 to 3,000 tokens per page) is re-queued on every single interaction. Token Saver completely changes this model. It installs as a light background program on your machine (using the Model Context Protocol, or MCP).
You ask your question. The local server searches your PDF using Local Hybrid RAG. This combines:
- Keyword Matching (BM25): To find exact terms like “SSM Section 14” or “Pindaan 2024”.
- Semantic Search (Cosine Similarity): To understand what you mean, even if you don’t use the exact wording from the document.
The server then hands Claude only the most relevant snippets (capped at a tiny payload). Claude synthesises the answer. The original file is never uploaded. This is what a massive reduction in computational waste looks like in practice:
| Document Type | Pages | Whole Doc Tokens | Returned Tokens | Resources Saved |
|---|---|---|---|---|
| FDA Drug Label | 33 | 23,959 | 1,021 | 95.7% |
| GDPR Regulation (EU 2016/679) | 88 | 70,260 | 996 | 98.6% |
| SFFA v. Harvard (Legal Ruling) | 233 | 133,349 | 740 | 99.4% |
The savings grow exponentially with the size of your documents. The bigger the file, the bigger the waste it prevents.
How This Applies to Your Malaysian SME
1. The Owner vs. The Filing Cabinet
You have dozens of supplier contracts, tenancy agreements, and NDAs. A partner asks, “What is our notice period for the Shah Alam warehouse?” Instead of scanning 200 pages manually or uploading an entire folder to an AI chat, you just ask the question. Token Saver searches your local folder, finds the exact clause, and gives Claude the precise text to summarise. It cites the page number so you can open the PDF and verify instantly. No data leaves your machine. No time wasted.
2. The Compliance Officer’s Shortcut
LHDN guidelines, Customs duty classifications, and HRDF regulations are notoriously dense. A single misinterpretation is costly. Token Saver retrieves the exact text from the original PDF. Instead of the AI guessing whether the penalty applies, it pulls the official wording and summarises it. You get accuracy and the ability to check the source immediately. It turns a 200-page tax guide into a personal assistant that answers only the relevant part.
3. The Operations Manual Maze
Manufacturing, F&B, or logistics SMEs live by SOPs and equipment manuals. “The machine is showing error code E-403. What do I do?” Instead of flipping through a 300-page manual or uploading the whole thing again, you ask your local system. It pinpoints the exact troubleshooting subsection. It is faster than Googling, more accurate than guessing, and completely private.
4. A Setup That Doesn’t Require a Tech Degree
Most RAG tools are painful to set up. They require Python environments, terminal commands, and JSON configurations. Token Saver is packaged as a single .mcpb bundle. You download it, open Claude Desktop Settings, click Install Extension, and point it at your document folder. If you can install an app on your phone, you can set this up. No coding required.
“The real friction in using AI for your business documents isn’t the AI itself. It’s the inefficiency of feeding it entire libraries when you only need a single sentence. Token Saver eliminates that friction entirely. Your data stays local. Your AI stays accurate. Your time stays yours.”
Your Practical Action Plan
- Stop uploading whole documents to cloud AIs for every query. You are wasting resources and exposing your data unnecessarily.
- Designate a single “Reference” folder on your computer. Put your active contracts, manuals, and guides there.
- Download the Token Saver
.mcpbfile from the project’s GitHub Releases page. - Install it into Claude Desktop via Settings -> Extensions -> Install Extension.
- Start asking questions differently. Instead of “Read this PDF for me”, ask “What is the penalty for late filing of Form E?” and let the system find the exact paragraph.
The Bigger Picture for Malaysian Business Owners
We are moving away from the era of the “generic” AI assistant. The next phase is personalised, private, local AI agents that truly understand your business documents.
For Malaysian SMEs, this is a massive equaliser. You do not need a huge IT team or an expensive cloud computing budget. An open-source tool on your laptop can now give you the same document-retrieval power as a large enterprise. It uses a Local Hybrid RAG system that blends keyword matching and semantic search to deliver the perfect answer every time.
The concept is simple: the AI reads your library so you do not have to, without ever exposing your most valuable data. This is the standard that smart SMEs will adopt to run faster, with more privacy, and far less wasted effort.
Ready to Streamline Your Operations?
Your business should run itself. AutoRunBiz deploys AI agents to automate your daily operations — WhatsApp orders, invoicing, customer follow-ups, and accounting. Book a free 15-min ops audit to see where automation fits your business →