In launch month, the agent connected to your CRM qualifies three times as many requests, marketing turns out variants by the hundreds and the model provider’s bill follows. That tool, put in place by your provider or your team, calls a language model. With Agent Libre behind it, that model is served in Canada for a fixed amount per month, and prospect records do not leave the country.
Talk about your salesSales management, marketing and the privacy officer each set a condition before putting a model behind their tools.
The campaign plan is set in January, with a budget per month. In September, the launch multiplies inbound requests, follow-ups and ad variants for three weeks. Billed on usage, that month arrives with a bill nobody wrote into the plan, and the marketing director is the one who explains it.
A CRM record holds the name, email, phone number and call notes of a person who is not yet a customer. That person agreed to receive a newsletter; that consent does not cover sending their call notes to a foreign provider. The privacy officer asks where the model processes those records before approving the tool.
A campaign in Quebec goes out in French first, then in English, with the same price, the same offer and the same tone in both versions. Marketing requires a model that writes everyday Quebec French, without calques from English, and that follows the company’s templates rather than inventing a slogan.
Your CRM, your email platform or your writing tool calls a language model. That model can come from Anthropic, from OpenAI or from Agent Libre. The tool stays the same; only the model’s address changes.
You reserve a processing speed for your tools. When the campaign exceeds that speed, requests line up; nothing is cut off and nothing is billed on top. A follow-up email that goes out thirty seconds later changes nothing. An inbound request that waits an hour does.
Before proposing an email, the agent re-reads the record, the call notes and past exchanges with that prospect. That re-reading is part of the monthly amount, with no volume limit under fair use. The rep can therefore let the tool re-read an account’s whole history rather than a summary.
Agent Libre’s gateway masks in transit the name, email and phone number contained in the request before it reaches the model. Masking rules are set per tool. The log shows which tool called the model, when and with what volume, and marketing can read it without going through IT.
The model is hosted and operated in Canada, on Agent Libre’s servers. No request is sent to the model’s publisher or to a foreign provider, and nothing is kept after processing. Your CRM stays where you installed it.
A website form, a demo request, an email asking for a price: your provider’s agent reads the request, finds the record in the CRM, proposes a score and a first email that the rep corrects or sends before noon. See Autonomous AI agents.
An eighty-page tender arrives on a Thursday, with the response due in ten days. The tool spots each requirement, matches it to your past answers and prepares a draft per section; the team fills in the prices and signs. See Document processing.
Thirty product sheets to update before the fall launch, ten ad variants to test, one email per segment. Your team’s assistant writes from the current catalogue and templates, in both languages, and marketing reviews before publishing. See Internal document assistant.
Four decisions to make before changing the model behind your CRM and content tools. None of them falls to Agent Libre.
Decide which inbound requests the agent may handle on its own, such as a price request on a catalogue product or a meeting booking, and which go straight to a rep, such as an existing account or a custom quote. Write those rules into your tool’s scenario before changing the model behind it; the model does not set them for you.
Schedule the free 24-hour to one-week audit on a campaign day: the peak is what sets the speed to reserve. Then keep twenty product sheets and twenty emails produced with the current model, have the same ones produced with Agent Libre’s model, and give both sets to the same person for review.
Check that your CRM, your email platform or your provider’s tool lets you change the address and key of the model it calls; without that setting, no other model can sit behind it. That is three lines of configuration per tool. Make the change on a morning outside any campaign, one tool at a time, keeping the old address at hand.
List the record fields that are not needed for writing, such as the name, email, phone number and account amounts, and ask that the gateway mask them before they are sent to the model. Add this processing to the privacy impact assessment required by Law 25 and to the privacy policy your prospects read.
Your provider or your team, in HubSpot, Salesforce, Pipedrive or another CRM, with an off-the-shelf tool, n8n or a home-built script. Agent Libre serves the language model that agent calls; the agent itself stays yours. Ask your provider whether the model’s address can be changed in its tool; that is the only condition.
Yes, in French and in English. The model served, GLM 5.2, takes the tone of your templates and examples; without examples, it writes neutral French. Compare on your own product sheets and emails before switching the reps over, with both models answering side by side for a week.
If the reserved speed is exceeded, yes: they wait their turn, with no cut-off and no extra charge. The campaign day measured during the audit is there precisely to reserve a speed that holds on that day. If peaks repeat, the speed is revised at renewal.
What the tool sends it, minus what the gateway masks. If the tool sends the full record, the gateway removes the fields you listed before the request reaches the model. The model keeps nothing from one request to the next; every email starts again from what the tool gives it.
The current amount does not move before renewal. At renewal, the reserved speed is revised from the consumption measured over the year, up or down, and the new amount is written in the contract. A team that doubles its reps reserves a higher speed; it does not pay for each text.
What an agent that re-reads its history on every turn costs, and what changes when the model behind it is served in Canada for an amount known in advance.
What a model served in Canada changes behind your support agent, and what your team decides before connecting it.
Why a volume that spikes costs more billed on usage than on an amount written in the contract.
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.
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.