Autonomous AI agents,without usage billing
Your agent reads an incoming email, opens the attachment, checks the customer record and drafts a reply: one task, some thirty calls to the model, and on every call it re-reads everything it has already done. With Agent Libre behind the agent from your provider or your team, that re-reading costs nothing more. The model is served in Canada for a fixed amount, written in the contract.
Talk about your AI agentsWhat breaks when an agent is billed by usage
Three effects teams notice once the agent is in production.
Every turn re-reads the whole task
An agent that settles a claim in fifteen steps re-reads the opening email, the customer record and its own notes fifteen times. With a usage-billed provider, each re-read is billed in full. The final answer is a small share of the total.
Retries do not show in the result
When a tool does not respond or a page has changed, the agent retries, with its whole history, until the task goes through. Those attempts do not appear in the delivered result, but they are all on the bill. An agent left alone overnight racks up dozens.
The team shortens the agent’s memory
To contain spending, your team cuts the history after a few turns, limits the agent to office hours or refuses to run several in parallel. The agent forgets what it read at the start of the task and quality drops.
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 autonomous agent
The agent that works through the steps of a task on its own: reading, searching, calling your tools, writing. Your provider or your team builds it and writes its instructions.
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 tool
The agent stays the one from your provider or your team. It calls a language model; that model, Agent Libre serves. Here is what changes once the address is swapped.
The orchestrator stays the same
n8n, LangGraph, LangChain or your home-built agent keep calling a language model. Your team replaces the model address and the key in the orchestrator. The scenarios, the connected tools and the agent’s instructions do not move.
One speed shared by all your agents
You reserve a processing speed for all your agents together. Ten slow agents or two fast ones use the same reservation, and adding an agent does not change the bill.
A 24-hour to one-week audit before you sign
Agent Libre places a gateway in front of your current provider for 24 hours to one week, with no commitment, and measures what your agents really consume: the re-read share, the peak hours, the number of agents active at the same time. The proposed plan starts from that measurement.
Data in Canada
The model is hosted and operated in Canada. Everything the agent sends goes through Agent Libre’s gateway, which logs the call and masks personal information before it reaches the model. Nothing is kept after processing.
What holds for every agent served by Agent Libre
Fixed amount
The amount is written in the contract for twelve months. It does not move during the year, whatever the number of tasks processed.
Re-reading included
The history re-read on every turn is part of the plan, with no volume limit, under fair use.
No training
Your agents’ requests are never used to train a model, neither Agent Libre’s nor anyone else’s.
What each team decides before connecting the agent
Four decisions to make on your side before changing the model behind the agent. Agent Libre provides the model and the gateway; the rest belongs to your provider or your team.
List the tools the agent is allowed to call (email, database, calendar, payment) and what it may write there without human approval. That list lives in the orchestrator: changing the model behind it does not make a write more careful, and does not authorize it either.
Replay a week of already-handled tasks with the old model and Agent Libre’s side by side, and compare the results task by task before switching. Also check that the agent accepts an answer that arrives a few seconds late without re-sending the request; otherwise it doubles the work at peak hours.
On your last three invoices, note the share for input, what the agent re-reads, and the share for answers. The first is part of the plan; it is what decides whether the proposed amount is lower than what you pay now. Then put the plan in the year’s budget, like a licence.
Take stock of what the agent sends to the model: email contents, customer records, documents opened during a task. Tell Agent Libre’s gateway which fields to mask before sending (names, card numbers, addresses). The setting is made once and applies to all your agents; the controls are described on the Governance page.
What we get asked on this topic
Does Agent Libre provide the agent?
No. The agent is put in place by your provider or by your team. It calls a language model, and that model can come from Anthropic, OpenAI or Agent Libre. Agent Libre serves the model that answers behind your agent.
What does my orchestrator have to allow?
One thing: choosing the model address and the key. n8n, LangGraph, LangChain and most home-built agents allow it. If your provider imposes its model without that setting, Agent Libre cannot sit behind it.
Does the model follow the agent’s instructions as well as the one I use now?
The model served, GLM 5.2, is built for this work: following an instruction, calling tools, returning an answer in the requested format. The only measure that counts is your own tasks; compare both models on the same cases before switching, no public ranking replaces that test.
How many agents can run at the same time?
As many as you want. The reserved speed is shared among all active agents; the more of them at the same moment, the longer each waits its turn. The 24-hour to one-week audit shows at what time your agents crowd together, which helps choose the speed to reserve.
What happens if my agents’ usage doubles during the year?
The amount written in the contract does not change. If your agents exceed the reserved speed, requests wait their turn, with no cut-off and no charge; your agents run slower at peak hours. At renewal, the plan is revised from measured consumption, up or down.
See also
Coding agents
What a model served in Canada changes behind Claude Code, Codex or Cursor, and what the team decides before switching.
Customer service
What a model served in Canada changes behind your support agent, and what your team decides before connecting it.
Re-reading: the line item nobody looks at
Why the history re-read on every turn weighs more than the answer in an agent’s bill.
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.