FAQ
Answers to common questions about Trelium, including whether it replaces your software, whether it is a chatbot, who builds agents, where AI is used, what needs approval, how to pick a first job, what happens when a check fails, data and AI training, what an agent can access, single sign-on, and credits.
Each answer stands on its own. Links go to the page with the full explanation.
Does Trelium replace my current software?
No. Trelium does not replace your order system, accounting software, CRM, or any other tool, and there is no data to move. Your existing tools stay the official home for your data. A Trelium agent reads from them, does its work, and writes the result back into them. See What is Trelium?.
How does Trelium connect to my tools?
In one of two ways. If a tool offers a built-in connection for other software (called an API), the agent uses that. If it doesn't, the agent can work in the tool through a web browser, much like a person logging in: it signs in, opens pages, and fills in fields. That is how the agent works day to day, not a promise that every login challenge or interface change will succeed without help. See Integrations and Browser Agents.
Is a Trelium agent a chatbot?
No. You don't type questions to an agent and wait for a reply. An agent watches for work, such as a new customer email, does the job, and leaves the result where your team already looks, such as a draft order or a suggested reply. See Agents & Workflows.
Who builds the agents?
Both. Your team can describe a job in natural language and build the agent that way, and the Trelium team helps with setup. You do not have to build everything alone, and Trelium does not have to build every agent for you. See Your First Workflow.
What kind of work does Trelium handle?
Repetitive office work that follows the same pattern but arrives in different forms. Examples are entering purchase orders from emails, PDFs, and spreadsheets, answering order status and stock questions, matching supplier invoices to purchase orders, following up on proposals and overdue invoices, flagging supplier delays, and building reports. Each job is handled by its own agent. The full list is on the use cases page.
Where does AI fit in?
AI does the reading and interpreting. That includes pulling line items out of a PDF, working out which order an email is about, and noticing that a request is missing something. Fixed steps and your business rules do the rest. AI drafts a result, rules check it, and a person reviews anything unclear or risky. See How Trelium Works.
Can AI make mistakes?
Yes. Like a person, AI can occasionally misread something. That is why Trelium keeps AI to reading and interpreting, and lets fixed rules decide what goes into your systems. A draft that fails a check is flagged and sent to a person. See Validation & Exceptions.
What should need a person's approval?
Anything that involves money, promises to customers, inventory, or production. An agent can prepare a sales order and still wait for a person to release it. The workflow sets where approvals sit. See Human in the Loop.
Does a person have to review every run?
No. Routine cases that pass every check can go through without review. Your team defines which cases need a person when you build the workflow, such as unclear requests, failed checks, and unusual orders. The Trelium team helps you set that up. These rules are specific to your workflow, not one platform-wide review policy. A common approach is to start with approvals around most steps, then let more run on their own as the workflow handles your real exceptions well. See Human in the Loop.
How do I pick a first job for an agent?
Pick a job that happens often, has a clear start and a clear finish, uses information your team can already reach, and is easy to check. Start with one slice of the work, such as PO intake for one team. A job you can judge within a week or two tells you quickly whether the agent works. See Choosing a Workflow.
What happens when a check fails?
The agent doesn't quietly carry on. It flags the problem as an exception and sends it to a person, who sees what the agent proposed and why it was flagged. Sometimes the person approves the result. Sometimes they supply a missing detail and the run continues from there. See Validation & Exceptions.
Can I see what an agent did?
Yes. Every run is recorded in the agent's Activity log, including the steps it took, when each step ran, and how the run turned out. Your team can look back at any job and see what happened. See How Trelium Works.
Is my data used to train AI models?
No. Trelium's security page states that nothing processed through Trelium, including orders, customer records, and emails, is used to train or improve any AI model, by Trelium or anyone else. The security page is the source for this policy. As of October 2026, Trelium's SOC 2 Type II audit is in progress (not completed). The Trust Center shows the current status. See Security & Data.
What can a Trelium agent access in my systems?
Only the software you connect it to, and only as far as you allow. If the software has a built-in connection for other tools (an API), the agent can do only what that connection allows. If it logs in instead, it uses an account your team creates for it and can see and do only what that account can. Your records stay in your own systems. See Security & Data.
Does Trelium support single sign-on (SSO)?
Yes. Your team can sign in to Trelium through your company's identity provider using single sign-on, instead of a separate password. Ask the Trelium team to set it up for your organization. See Security & Data.
What are Trelium credits?
Credits are how Trelium measures an agent's work. They are used each time an agent does its job, not when it is built. A cancelled run can still show credits for the work it did before it stopped. The agent's Activity log shows how many credits each run used. An agent uses more credits the more often it works, the more there is to do each time, and the more tools and steps the job involves. Plans and prices are on the pricing page. See Credits & Usage.
How do I measure whether an agent is working?
Write down how long the job takes today before you change anything. Then measure results, not activity: time per job, hours given back to the team, backlog, how fast customers hear back, and how often a person has to fix the agent's output. See Choosing a Workflow.
Credits & Usage
What Trelium credits are, what uses them, where to see how many credits each run used, why some agents use more than others, and how to estimate usage before you build.
Glossary
Definitions of the terms used in the Trelium docs, including agent, workflow, Flow, credit, run, trigger, input, output, step, exception, human in the loop, deterministic step, validation, API, integration, and browser agent.