AI behind your tools,hosted in Canada
An advisor opens an AI tool to summarize a client’s file before a meeting; the file contains a statement, an account number and an address. The tool sends that text to a language model, and that model is what Agent Libre serves, from servers in Canada. The gateway masks personal information before sending and records every call in a log compliance can read.
Talk about your financial services projectThree requirements before any AI project
Before approving an AI tool, compliance, internal audit and finance each put one question to the model provider.
Knowing where a loan application goes
A mortgage application brings together a pay stub, a notice of assessment, account statements and a piece of ID. When your provider’s tool sends those pages to a model, compliance’s vendor questionnaire asks for the country of the servers, who operates them and who can read the data in transit. A provider whose servers are in the United States sends the project back into review for months.
Answering the audit without rebuilding
At year end, internal audit asks for the list of tools that sent client data to a model over the past twelve months, with which fields and under whose approval. If each team took its own key from a provider, the answer is rebuilt one email at a time.
Staying within the budget approved in the fall
The AI budget is approved once a year. Then come RRSP season, tax slip production and policy renewals: the tools work far more for two months. A usage-based bill climbs in those months and gets justified in committee after the fact.
The tool does the work, Agent Libre serves the model
Your service platform, your document processing tool or your internal assistant keeps doing its work. The model that answers it comes from Agent Libre, for a fixed amount per month.
One tool at a time, without redoing the tool
Your provider or your team points the tool at the model served by Agent Libre instead of Anthropic’s or OpenAI’s: the address, the key and the model name. The tool’s screens, scripts and rules do not move. You can start with a single tool, the internal assistant for instance, and connect the others one by one.
Masking rules per tool
Agent Libre’s gateway spots account, card and social insurance numbers, names and addresses in transit and replaces them before the text reaches the model. The rules are set tool by tool: an assistant on credit policies does not need the same rules as an agent answering clients.
A log and a quota per team
Every call is recorded with the tool, the team, the model, the time, the volume and the rules applied. A quota per team keeps one tool from taking up all the speed reserved for the others. Your teams pull the reports themselves, with no request to the provider.
Data in Canada
The model runs on servers hosted and operated in Canada. Requests do not leave the country, are never used to train a model and are not kept after processing.
Three common uses
Document processing
Your provider’s tool reads a loan application, pulls out the income, the employer and the account balances, and fills in the credit file. When an insurance claim arrives, it classifies the file and extracts the date, the amount and the type of loss. See Document processing.
Assistant on internal documents
An advisor asks whether a self-employed client without two years of declared income qualifies for a product. The assistant set up by your team answers from the credit policy in force and cites the page. See Internal document assistant.
Customer service
Your provider’s agent answers evening and weekend questions about a pending payment, a deposit that has not shown up or a policy renewal. Complaints and disputes go to a person. See Customer service.
What each team decides before connecting the tool
Four decisions are made before changing the model behind a tool. None of them falls to Agent Libre.
Choose the tool that goes through the gateway first, the one that touches the fewest clients live, and write the rule that every new AI tool goes through it too. Without that rule, you will have two inventories a year from now: the gateway’s, and the keys taken directly from providers.
List, tool by tool, the fields the model does not need to read to do the work: account number, social insurance number, date of birth, address. The gateway masks them before sending, and that list goes into the privacy impact assessment Law 25 requires for the project.
Set now what the log must contain for your next cycle: the tool, the team, the model, the time, the volume and the rules applied. Before going to production, have someone who did not take part in the project pull a report; if they manage it alone, the auditor will too.
Ask for the free 24-hour to one-week audit before setting the plan: it measures what your tools actually consume at the current provider, without changing anything in your tools. Then set the amount for the full fiscal year, knowing it is revised at renewal from measured consumption, up or down.
What we get asked on this topic
Do our clients’ files pass through an American provider?
No. The model runs on servers in Canada, operated by Agent Libre. No request is sent to a foreign provider, and nothing is kept after processing.
Does the model see account numbers and personal information?
Only what you let through. The gateway replaces the fields you have listed before the text reaches the model. To summarize a file, classify a claim or answer a question about a product, the model does not need those values and the answer stays the same.
How do we know the model is good enough for our statements, policies and loan applications?
By comparing it on your own documents. During a period when both models answer side by side, your team submits a batch of already-processed files to both and compares the answers on its own criteria, starting with the error rate on extracted fields. No public evaluation replaces that batch.
What happens in RRSP or tax slip season, when our tools work far more?
The month’s amount does not change. If your tools exceed the reserved speed, requests wait their turn, with no cut-off and no fees. For an agent answering clients live, that wait matters: reserve the speed for those months when setting the plan.
Can our banking or insurance platform use Agent Libre’s model?
Yes, if your provider lets you choose the model in its tool’s settings, which most tools built to call a language model do. Agent Libre provides neither the platform nor the agent, only the model that answers behind it. Ask your provider before signing; without that setting, the tool stays with its model provider.
See also
Governance and compliance
The monthly report, the inventory of AI tools and the rule review your compliance team produces from the gateway, with no request to IT.
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.
What “hosted in Canada” actually means
Servers, operator, applicable laws: what to check behind the claim.
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.