Every Vendor Contract Is a Different File Type. Your AI Agent Doesn't Care Anymore.

Your last three vendor contracts arrived as a PDF, a Word file, and a photo of a signed page. Nobody in your business has actually read all three.

Marcus runs a six-person specialty foods importer in Manchester. Twelve suppliers across four countries. Every one of them sends paperwork differently.

The Italian olive oil producer sends a PDF with a stamp on it. The Polish bakery sends a Word document with the payment terms buried in paragraph four. The Spanish cheese co-op sends an Excel price list that changes every quarter. One supplier still faxes a scanned page, then emails a photo of it.

Marcus signs all of it. Nobody re-reads it once it is signed.

You Are Not Behind on Contracts. You Never Had a System.

This is not a Marcus problem. This is every small business with more than three suppliers, three landlords, or three long-term clients.

Here is what actually happens to a vendor contract after it is signed:

• It gets saved into a folder named after the supplier, or a folder named "contracts," or an email thread nobody will find again.
• The payment terms, the renewal date, and the price break clause live inside the document, not in anyone's calendar.
• Three months later, the supplier quietly renews at a worse rate, because the renewal notice window closed while the PDF sat unread.
• The only person who remembers the actual terms is the person who signed it, and they are not always available when the question comes up.

You do not have a filing problem. You have a reading problem. Every contract is a small, one-time data-entry job that never gets done, because nobody has an hour to retype a PDF into a spreadsheet. So it does not get done, and the business runs on memory instead of terms.

What Changed This Week

On 6 August, Teknium, the lead engineer behind Hermes Agent at Nous Research, posted that Hermes can now "read anything you throw at it: PDF, Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, all auto-converted to clean Markdown the moment the agent reads the file, all locally." It runs on a new open source engine called anydoc, built by the web-scraping company Firecrawl, written in Rust, with zero setup.

The tweet is not a small update buried in a changelog. Within a day it had pulled over 274,000 views, nearly 3,000 likes, and more than 1,600 bookmarks, and Hermes-focused accounts were still reacting to it the next morning, calling the feature "built different."

Here is why that number matters more than a typical feature launch. Document reading used to mean picking a lane: OCR software for scanned pages, a different converter for Word, a different one again for Excel, and usually a cloud upload step that made anyone nervous about a signed contract with a client's bank details on it. Anydoc collapses all of that into one step, and it runs on the machine in front of you, not on someone else's server.

How Do I Get an AI Agent to Read My Vendor Contracts Automatically?

You point it at the file and ask a specific question. That is the whole trick, and it works whether you are using Hermes Agent, Claude with a document-reading skill enabled, or ChatGPT with file upload turned on.

The difference between a useless answer and a useful one is the prompt. Compare these two:

• Chatbot prompt: "Summarize this contract." You get three vague paragraphs and nothing you can act on.
• Agent prompt: "Read this file. Extract the supplier name, the payment terms in days, the renewal date, the notice period required to cancel, and any price break clauses. Output it as a table row." You get a row you can paste straight into a tracker.

The first is a chatbot answering a question. The second is an agent doing a job. Hermes Agent Tips, an account that closely follows Nous Research's releases, put it plainly a few hours after the anydoc launch: the agent "processing" a file is worthless on its own. The value is in the specific, repeatable extraction you ask it to do every single time a new document lands.

Isn't This Just OCR Software?

No, and the difference is the whole point. OCR reads text off a scanned image. Dedicated bookkeeping tools like Dext, Docyt, and Booke AI do a version of this already, but only for receipts and invoices, and only inside their own subscription.

A general-purpose agent with anydoc-style reading is not limited to one document type or one workflow. The same skill that reads a supplier contract also reads the lease renewal, the insurance certificate, the freight forwarder's terms, and the client NDA, because it is not a specialized tool, it is a general reading capability you can point at anything.

And because it can run locally, on your own machine or a VPS you control, the contract never has to leave your hands to get read. For a business handling signed agreements with payment details and client information, that is not a small detail.

What you're comparing	Dedicated bookkeeping SaaS (Dext, Docyt, Booke AI)	General AI agent with document-reading skills
File types covered	Receipts and invoices only	PDF, Word, Excel, PowerPoint, OpenDocument, RTF, EPUB, scanned pages
Monthly cost	Per-seat subscription, usually $20 to $50+/month	Whatever you already pay for the agent, often the same subscription you use for everything else
Where the document goes	Uploaded to their cloud	Can run entirely on your own machine
Flexible to new document types	No, built for one workflow	Yes, same skill reads contracts, leases, certificates, reports

Build It: A Contract-Reading Assistant in Five Steps

1. Pick one folder. Every signed vendor contract, lease, and supplier agreement, whatever format it arrived in, goes into one place. Stop sorting by supplier first. Sort after the agent has read it.
2. Write the extraction prompt once. "For every file in this folder, extract: supplier name, contract start date, renewal date, notice period to cancel, payment terms, and any price or volume clause. Output one row per file in a table." Save this prompt. You will reuse it dozens of times.
3. Run it against your backlog first. Point the agent at every contract you already have, not just new ones. This is the part that actually surfaces the renewal you were about to miss.
4. Turn the output into a tracker. A simple spreadsheet or table with a column for renewal date is enough. The goal is not a fancy dashboard, it is one place where every deadline lives outside of someone's memory.
5. Add a standing reminder. Ask the agent to flag anything renewing in the next 30 days, on a weekly schedule if your setup supports cron jobs, or as a manual Monday-morning check if it does not. Either way, the notice window stops depending on who happens to remember.

None of this requires a developer. It requires one prompt, one folder, and the willingness to run your existing paperwork through it once.

What You Have After This

A week in, you have read every contract sitting in that folder for the first time since it was signed, and you already know which renewal is closest.

A month in, every new supplier agreement gets the same five-minute read the day it arrives, instead of the day someone finally has time, which in practice was never.

Six months in, you have never missed a renewal notice window, you have caught at least one supplier quietly raising a price past what you agreed to, and the person who signed the contract is no longer the only person who knows what it says.

Same contracts. Same suppliers. A business that finally reads its own paperwork.