An assistant that readsyour documents in Canada
An employee asks how many vacation days they have left or what notice a supplier contract requires, and the answer sits in a document nobody re-reads. The assistant your provider or your team puts in place reads those documents and answers by citing the passage. With Agent Libre behind it, the model that writes the answer is served in Canada for a fixed amount, whether employees ask ten questions or a thousand.
Talk about your assistantThree problems specific to a document assistant
Teams that put an assistant in front of their policies and contracts run into the same three obstacles.
Two versions of the same policy
The assistant answers from what it finds. When the 2023 telework policy still sits on the shared drive next to the 2025 one, it cites either with the same confidence, and the employee has no way to know which one applies.
The bill follows the reading
To answer a two-line question about group insurance, the assistant sends the model the twenty pages of the plan. In November, during the benefits enrolment period, questions multiply and the metered bill climbs with them.
Clauses that leave the organization
A supplier contract holds negotiated prices, penalties and names. Every question sends the relevant clause to the model. When that model is served abroad, the legal department asks where that text goes, who can read it and how long it stays.
You, your tool, and the model that answers it
Your tool stays the same. What changes is the model that answers it.
Your organization
Your teams, your data and your rules. You decide what the tool does, and with which data.
Your internal document assistant
The assistant your employees ask about vacation days, group insurance or a supplier contract. Your provider or your team connects it to your files and decides which documents it may read.
The model that answers the tool
Every time the tool needs an answer, it calls a language model. That model can come from different providers. This is where Agent Libre comes in.
What Agent Libre changes behind your assistant
The assistant stays the one from your provider or your team. Four things change when the model answering behind it comes from Agent Libre.
The same assistant, another model behind it
The assistant calls a language model to write each answer. That model can come from Anthropic, OpenAI or Agent Libre. Switching it changes neither how the assistant finds documents, nor the access rights, nor the screen employees see.
The whole document, on every question
The text sent with each question is part of the plan. A sixty-page collective agreement can go out whole, every time, without moving the amount. Your team chooses between the full document and the useful passages based on answer quality, no longer on spending.
A reserved speed
You reserve a processing speed. At peak hours, if questions arrive faster than that speed, answers wait a few more seconds; none is refused and none is billed extra.
Data in Canada
Your policies and contracts stay in your systems. Only the text attached to each question goes through Agent Libre’s gateway, on servers hosted and operated in Canada. Nothing is kept beyond processing and nothing is used to train a model.
Three rules for a document assistant
One reference per answer
The assistant sends each passage with its reference (document, section, page) and asks the model to return it with the answer. An answer without a reference is fixed in the assistant’s instructions.
Access rights first
The assistant sends the model only the documents the person asking is allowed to read. The model does not know your access rights; that rule applies before the call.
Masked in transit
Agent Libre’s gateway masks personal information in the text sent to the model and logs every call. You see who queries what, with which model.
What each team decides before connecting the assistant
Four decisions to make before changing the model behind the assistant. They weigh more on answer quality than the choice of model itself.
Designate the official version of each policy and remove the old ones from the folder the assistant reads. It answers from what it finds: an outdated policy left on the shared drive gives an outdated answer, reference included. Name the person who updates that folder at each revision.
Decide whether the assistant may summarize a contract clause or must quote it word for word and refer to the person in charge of the file. Write that rule into the assistant’s instructions before changing the model behind it. The model follows the instruction; it does not set the level of caution for you.
Check that your platform lets you change the model’s address, the key and the model name: those are the three configuration lines to edit. Gather thirty real questions employees asked last month, with the right answer, and run them through both models side by side for two weeks before switching everyone.
List what contracts and files contain that is not needed to answer: employee names, social insurance numbers, home addresses. Have the gateway mask those fields before sending to the model and ask for the call log. The assessment required by Quebec’s Law 25 relies on those two elements.
What we get asked on this topic
Does Agent Libre provide the assistant?
No. The assistant comes from your provider or your team: it finds the documents, applies access rights and displays the answer. Agent Libre serves the language model it calls to write that answer.
Does the model answer as well as Anthropic or OpenAI on our policies?
On reading a document and answering with a reference, the gap between the large models is thin. Quality depends first on the passages the assistant sends and the instructions it gives. The only measure that counts is made on your documents, with your questions, both models side by side.
What does the model see of our documents?
Only the text the assistant attaches to each question: a passage, a chapter or the whole document, depending on your team’s setting. It does not browse your shared drive and keeps nothing from one question to the next.
What happens if usage doubles, for example when the assistant opens to the whole organization?
The monthly amount does not move. If the new pace lastingly exceeds the reserved speed, answers slow down at peak hours, and the plan is revised at renewal from measured usage, up or down.
How do we know what our assistant consumes today?
Through the free 24-hour to one-week audit, with no commitment. Agent Libre’s gateway sits in front of your current provider and counts, question by question, the text sent and the answers returned. You leave with the measurement, whether you sign or not.
See also
Document processing
What a model served in Canada changes behind a tool that reads documents by the thousand, and what each team decides before connecting it.
Public sector and municipalities
What a ministry, a city, a hospital or a college decides before putting a model served in Canada behind its tools, and what the auditor finds there.
What “hosted in Canada” actually means
Where the servers are, who operates them and what remains of a question’s text once the answer is returned.
Your company has two supply chains
The first one you already know. Materials come in, they are transported, stored, transformed, delivered. Every step has a supplier, a lead time, a unit cost. When the margin tightens, you walk down that chain step by step, and you generally know where to look.
The second one moves information instead of materials. A price request comes in and becomes a quote. An order comes in and becomes a production order, an invoice, a confirmed delivery. A site report becomes a change order, or it does not. This chain runs through every one of your departments and it employs a large share of your people.
The 24-hour to one-week audit
We measure your actual consumption at your current provider: who calls what, what it costs, and what is leaving your walls. Free, no commitment, and the report is yours even if you go no further.