Logs & Audit Trails
How the Activity log records what an agent did, how to use it to review a piece of work, and what it does not cover.
A finished result only tells part of the story. When a reply looks wrong or a record needs checking, you also want to know what happened along the way. Trelium keeps that record in each agent's Activity log, so your team can follow the work step by step instead of judging it only by the end result.
What the Activity log shows
Open the agent's workspace and select Activity log, next to Overview and Canvas. Each row is one run, which is one pass through the agent's workflow. A run can cover one piece of work or several. For each run you can see:
- Whether it finished, is still going, or ran into a problem
- Which apps it used
- How many credits it used
- When it started and how long it took
Open a run to see its steps. Each step shows when it happened and how long it took. See How Trelium Works for how those steps fit together.
One run opened up in an approval agent's Activity log. Each line is a step, with the app it used, when it ran, and how long it took.
Reviewing a piece of work
Start with the job you want to check, such as an inquiry, an order, or a reply. Note which agent handled it and roughly when. Then find that run in the agent's Activity log and open it.
The log tells you what the agent did. To judge whether it did the right thing, set it beside two other things: the rules the workflow is meant to follow, and the original request and records. A run with every step completed is not proof that the agent used the right facts.
| Question | Where to look |
|---|---|
| What did the agent do? | The steps in the run, and the result in the tool it updated. |
| What information did it use? | The original request and the records it was working from. |
| Why did it go that way? | The workflow's rules and checks, and any problem it flagged. |
If the log does not show the detail you need, check the original source. Do not fill the gap with a guess.
Example: a draft that does not look right
Say an agent writes replies to email inquiries using a reference document, and your team finds a draft with the wrong answer.
Find the run
Identify the agent and when it handled the inquiry, then open that run in its Activity log.
Follow the steps
Check that it did what you would expect: read the inquiry, checked the reference document, and prepared the draft.
Compare with the sources
Put the original inquiry, the reference document, and the draft side by side. This is where you see what went wrong.
Fix the cause
If a rule was missing or the wrong information was used, fix that first, then try the work again. See Runs, Retries & Failures.
What the log does not cover
The Activity log is a record of the agent's steps and results. It can show that a review email was sent or that a Record Decision step ran. It is not, on its own, a complete native audit of who approved, edited, or rejected something inside Trelium.
Some workflows write a separate review log into a customer system, such as a Google Sheet. If you need a durable record of human decisions and actors, keep it in the tools and process you already use for approvals, or ask the Trelium team what your workflow records.
Using the log for an audit
The Activity log shows what the agent did. If an audit also asks who approved a change, or what a record looked like at the time, you will need other evidence alongside it. Check what an audit needs before relying on the log alone. For what Trelium records and protects beyond the Activity log, see Security & Data.
Where to go next
Validation & Exceptions
How checks flag work that needs attention, and what is recorded.
Runs, Retries & Failures
What to check when work cannot finish or needs another attempt.
Human in the Loop
Where people review work and make decisions.
Browser Agents
How an agent works in a website when an API is not available.
Security & Data
What an agent can reach, where your data stays, and where to find Trelium's security policies.
Triggers & Inputs
How a Trelium agent knows there is work to do and what it receives when a job starts. Covers event and schedule triggers, the kinds of input an agent can read, and what happens when an input is unclear or incomplete.
Testing a Workflow
Use past work to check a workflow's results, test how it handles problem cases, and decide what is safe to put into use.