Analytics

Know what your agents cost, and what they deliver.

Every run in ctxflows is recorded with the item it worked, the workflow it followed and what each agent session spent. So you see what a piece of work cost, whether it passed the first time and where it waited.

Cost per item, milestone and workflow

Analytics adds up what every agent session spent, so you see the estimated cost per task: per item, per milestone and per workflow.

Split it by model, with cached tokens kept apart, and see the costliest sessions and what work that never landed spent. It's the vendor's published rate applied to the tokens used, there to compare work: you pay your AI vendor on your own plan.

Burn-up per milestone

A burn-up draws a milestone's done work rising toward its scope, so work finished and work added show on one chart.

Throughput and cycle time stand beside it, per board or milestone. The lines show a trend, not a deadline: the roadmap holds no dates, so nothing is marked late.

First-pass rate and rework

First-pass rate is the share of runs that end without a redo or a send-back; rework is what the others needed on top.

See both per workflow, with the node that needs the most attempts and what the redos cost.

Which workflows work

For each workflow you see how its runs end, where they fault and how many rounds its loops take to pass.

Compare how often it ships with how often it sends work back, and spot loops that keep running to their limit.

Time waiting on you

Time waiting is how long runs stand still until a person answers, by what they wait for: an approval, a question or a fault to look at.

See the typical wait, the slowest and the oldest still open. A waiting run spends no tokens, so this is time your work loses, not money.

Context per task

Context per task shows how much each task reads before it starts, so you see where your rules and docs weigh most.

Trim what a task doesn't need, and the next run starts lighter and costs less.

Improve from your own runs

The numbers feed your own workflows: a scheduled workflow reads past runs and their cost, and proposes changes you accept or dismiss.

It notes what slowed or stopped each run in a dataset, so the next review reads the history. How to build one is on Workflows.

One milestone's burn-up, item by item

You open the "Checkout works" milestone to see why it's taking longer than the last one.

  1. Read the burn-up

    Done work is rising, but the scope line stepped up where gaps became items.

  2. Sort its items by cost

    One item cost far more than the rest, and its runs show redo after redo at review.

  3. Open that item's runs

    Each redo came back from the same failing check, and most of the item's spend went to runs that never landed.

  4. Check the waits

    Much of its cycle time went to waiting overnight for a review.

  5. Change the workflow

    You add a loop that repeats the build until that check passes, so review only sees passing work.

On the next milestone, the same workflow passes review the first time more often, and the cost per item shows it.

Questions about analytics

Is the cost per task exact?

No, it's an estimate: the vendor's published rate applied to the tokens each session used, by model. You pay your AI vendor on your own plan, so the estimate is for comparing work, not a bill.

Can I compare models?

Yes. Cost, tokens and outcomes are kept per model, so you can run the same kind of task with two agents or models and see which one delivers for less.

Where do my runs wait?

Time waiting shows it by kind, such as approvals, questions or faults, with the oldest waits still open, so you know which answers hold work back.

Can I export the numbers?

Yes. The runs, items and decisions behind every number export as plain files at any time, and a workflow can write the figures you track into a dataset you export as CSV.

See what every piece of work costs

Cost, burn-up and rework, for every item your agents work on.

Beta accessSee pricing