What is a compute unit
Who this is for: admins who want to understand their usage figures or estimate what a workflow will use.
Before you start: nothing.
A compute unit (CU) measures how much work a run used. It combines three things: how long the run took, how much CPU its scripts used, and how much memory they needed. Reading files and data in built-in steps counts too (see below). One number makes runs of different kinds easy to compare and to add up per month.
The formula
Section titled “The formula”Each attempt of a run gets:
- 0.1 CU per second the attempt took,
- plus 1 CU per billion units of script CPU (the sandbox counts CPU in “fuel”; about one second of busy script work is 10 billion units, which is also the default limit per step),
- plus 0.05 CU per second for each 64 MB of peak script memory.
Written as a formula (version 1):
CU = 0.1 × seconds + fuel / 1 000 000 000 + 0.05 × seconds × (peak memory in MB / 64)Most workflows spend their time waiting for other systems, not computing, so the first part is usually the largest.
Examples
Section titled “Examples”| Run | Seconds | Fuel | Peak memory | Compute units |
|---|---|---|---|---|
| Fetch a list, build a CSV, send an email | 2 | 0.2 billion | 32 MB | 0.2 + 0.2 + 0.05 = 0.45 |
| Wait 10 seconds for a slow API, little script work | 10 | 0.05 billion | 16 MB | 1.0 + 0.05 + 0.125 = 1.175 |
| Heavy script that transforms many rows | 5 | 8 billion | 128 MB | 0.5 + 8 + 0.5 = 9.0 |
A workflow that runs the first example every weekday uses about 0.45 × 22 = 10 CU a month.
Good to know
Section titled “Good to know”- Usage is counted per attempt. If Liitora tries a run again because of a temporary problem on our side, each attempt counts.
- Reading data counts like script work. Built-in steps read XML, CSV, Excel, fixed-width and zip files, JSON answers of
http.requestand AI models, error replies of Microsoft 365 email senders, and webhook bodies in the same safe sandbox as scripts. The CPU and memory that reading uses are added to the run’s fuel and peak memory, so a run without scripts can still use some CPU. A step’s own fuel figure includes its reading too, and each unit is counted once. Reading a small file costs very little; a large file costs more, in the same way as a script that reads it. - Usage is stored exactly, in millionths of a CU (micro-CU), so monthly totals always add up.
- The formula has a version number. If it ever changes, each day’s usage says which version was used, and we tell you before.
- Test runs are recorded separately (as
test_run) and don’t count toward your plan’s monthly compute units (cu_per_month). Only real runs (run) do. - The approval page shows the Expected cost of a workflow per month. It estimates from recent runs or, for a new workflow, from its test durations at 0.1 CU per second.
When your company reaches its plan’s cu_per_month limit, new production runs end at once with quota_exceeded until the next month (UTC), and admins get an alert. See Limits per plan.