Activity
Activity shows what your agents did over a period of time and what it was
charged: how many tasks they handled, how many met their goal, which agents
and AI models did the work, and the AI usage behind it. It is a page of the
admin area at aptiveai.io/admin/activity, the
Activity row in the admin menu.
| Role | Sees Activity |
|---|---|
| Owner | ✓ — your own enterprise |
| Admin, Member | — Activity is visible to your enterprise’s owner. An admin can open Productivity instead |
The page has five tabs: Overview, Models, Trends, Explore and Tasks. Refresh sits at the end of the tab row. The time range and the tab are kept in the page’s address, so a link you share opens on the same view.
The figures
Section titled “The figures”The same word means the same thing on every tab, and on Productivity.
| Figure | What it means |
|---|---|
| AI usage | What your enterprise was charged for AI over the period: every step your agents took, in US dollars. AptiveAI’s own background checks are never in it |
| Tasks | Tasks started in the period. A task belongs to the day it was started, even if it finished later. Deleted tasks are not counted |
| Goals met | Tasks that met their goal — see What a goal met is |
| Not graded | Tasks that have no verdict yet, or will never have one. They are never counted as met |
| AI usage per goal met | The AI usage of all the period’s tasks, divided by the number of goals met. Tasks that did not meet their goal are included, so it answers “what did each success cost, failures included”. Building agents and Try it are never in it |
| AI steps | Each call an agent makes to its AI model. A task that reads many pages, or thinks through a long problem, takes more steps |
| Computer time | How long your agents’ computers were on, including the few minutes one is kept warm after a task so the next starts faster |
What a goal met is
Section titled “What a goal met is”Each night, AptiveAI reads the tasks that have finished — what each was asked to do, the agent’s final summary, and what it actually did — and decides whether it met its goal, partly met it, was blocked, or did not meet it. Only a task with that verdict can count as a goal met: a task that completed but has not been judged is not graded, not met.
A task is not graded when:
- it is still in progress, or was stopped or skipped — its outcome is already known;
- it finished today — the verdict arrives the next night;
- grading is off for your enterprise, or the task’s workspace is marked as regulated, so its tasks are never read.
The verdict is a judgement, not a fact. The counts beside it — records read, connector actions, approvals, pages visited — are counted, not judged.
The time range
Section titled “The time range”The time range in the page header picks the period every tab shows:
Last hour, Last 3h, Last 6h, Last 12h, Last 24h, Last week, Last 30 days, Last 90 days, This month, Last month, This year, All time, or a Custom range with From and To.
- ‹ and › step to the previous or next period of the same length.
- All time starts at your enterprise’s first AI usage.
- Rollup — Auto, Hour, Day, Week or Month — sets how the charts group time. Auto picks hours for up to two days, days up to about two months, weeks up to about six months, and months beyond. A choice that would draw too many points for the period is turned off.
- Times are in your browser’s time zone, so a day starts at your midnight.
- vs prior period compares a figure with the period of the same length just before.
The same time range is on Productivity.
Overview
Section titled “Overview”Five tiles:
| Tile | What it shows |
|---|---|
| AI usage — Included in your seats | What was charged for AI in the period, with the change vs the prior period |
| Tasks — N goals met | Tasks started in the period, and how many met their goal |
| AI usage per goal met | AI usage of all the period’s tasks ÷ goals met |
| AI steps — Each call an agent makes to its model | With the change vs the prior period |
| Wallet charges — AI beyond what your seats include | What was actually taken from your wallet for AI in the period |
Then four cards:
| Card | What it shows |
|---|---|
| Team capacity | Everyone in your enterprise, with the Tasks they started and their AI steps. There is no dollar figure per person, anywhere on this page. Until seat tiers are enabled the card says Capacity appears here once seat tiers are enabled. |
| Wallet | Its Balance now, and what was Charged this period. Pays for AI beyond what your seats include. Everyone in your enterprise draws from it. |
| Agents | Each agent’s Tasks, Goals met, AI usage and AI usage Per task |
| Models | Each AI model’s AI steps and AI usage. Premium models use your team’s capacity faster. |
An agent that has since been deleted reads A deleted agent, and someone who has left your enterprise reads A former member. A model without a display name reads A model.
Models
Section titled “Models”Impact of model by agent puts each of your agents next to each AI model it used in the period, so a change of model can be judged on real results.
- A task counts once, under the model it used the most.
- When an agent used more than one model, its rows sit together, so the difference is side by side.
| Column | What it shows |
|---|---|
| Agent, Model | By name |
| Tasks | Tasks of that agent on that model |
| Goal met | The share of graded tasks that met their goal, with of N graded · M not graded beside it |
| AI usage per task | The AI usage of those tasks ÷ tasks |
| AI usage per goal met | The AI usage of those tasks ÷ goals met |
| AI steps | Steps those tasks took |
What the same work would cost on another model
Section titled “What the same work would cost on another model”Compare with picks any model an agent can use. The table then gains a column, Same tokens on ‹model› (estimate): what the same work each row did would have been charged on that model, per task.
Beside each estimate, own model at list price gives the same work on the row’s own model at its standard price. That is the figure the estimate is measured against, so the change shown is the difference in price alone, and comparing a model with itself shows no change. When a model has no standard price, the row is measured against what it was actually charged instead and says so: its own model has no list price.
Trends
Section titled “Trends”Three sections — Models, People and Agents — each a chart over the period beside a Trending list: the last 7 complete days against the 7 before, with the change (or New when there was nothing in the earlier week) and a 14-day line. Today is in neither week, so a day still in progress never reads as a drop.
Models and Agents can be measured in AI usage or AI steps. People are measured in AI steps only.
Explore
Section titled “Explore”Pick a Metric (AI usage or AI steps), Group by — Agent, Person, Model or Activity — an optional Then by, and Show top 5, 10 or 25. It starts grouped by Agent. The chart draws the top five and Other; the Breakdown table under it gives each row’s Sum, its smallest, largest and average per period, the latest period and its % of total. Export CSV downloads exactly the table.
Activity sorts AI usage by what it was for:
| Activity | What it is |
|---|---|
| Tasks | Tasks a person started |
| Automations | Tasks an automation started |
| Agent building | Building and editing agents: the builder’s chat, Try it, and generating an agent’s icon |
| Overhead | Small jobs around a task, such as writing its short title |
Click a row or a bar to drill in: it becomes a filter, and the grouping moves one level down. Remove filters one at a time, or Clear all. Once a person is in the grouping or the filters, the metric is AI steps only.
| Tile | What it shows |
|---|---|
| Tasks | Tasks started in the period: how many were automated and how many started by hand |
| Success | One rate over every finished task: graded tasks that met their goal, plus ungraded tasks that completed. The hint gives each half |
| Median active time | How long an agent actually worked on a typical task, with the mean beside it. Time spent waiting for a person, a question or an approval is not counted |
| AI usage per task | And AI usage per goal met |
| Computer time | Time on tasks, and time kept warm for the next task |
Then:
- Tasks by outcome — each period’s tasks by their verdict, or by how they ended when they were not graded.
- Why tasks didn’t complete — how tasks ended (stopped by a person, an approval declined or timed out, AI usage ran out, no progress) and what stood in the way of the ones that did not meet their goal (missing information, a missing file, a sign-in, the live view, not allowed, a system error, out of scope).
- By agent — tasks, goal met, average active time, turns, tools and approvals per task, computer time, AI usage and AI usage per task.
- Top tools — the ten tools your agents called most.
- People — the tasks each person started. A count, never money.
Activity and what was taken from your credit
Section titled “Activity and what was taken from your credit”AI usage on Activity and Productivity is worked out at today’s prices for every date in the period. What was actually taken from your credit, step by step, is on Transactions, and the two can differ for older dates. The Wallet charges tile and the Wallet card read Transactions.