Assistant usage
Your tier includes use of the assistant that renews by itself. There are no credits to buy or count, and no overage is ever charged: the price of your tier stays the same from one month to the next. The Hosting tier is the exception: its assistant is paid from a balance (see below).
On this page
The gauge
In an application's conversation, under the message box, “Assistant usage (team)” shows a percentage and a bar, then a sentence: “Comes back fully at 10:00 PM · Details”. The bar follows the fuller of two sliding periods, “over 5 h” and “over 7 days”, and the time says when it will be back to zero, after the day when it is not today: “Comes back fully on Monday at 12:00 PM”. Click “Details” to see each of the two periods, the number of changes published to the live app, and the link “See the usage per person”.
The usage is the whole organisation's: your colleagues and you draw on the same periods, whatever the application.
The “Assistant” tab of “AI usage”
In the menu, “AI usage” opens on its “Assistant” tab, which details the whole organisation's usage, without opening a conversation. Everything there is a share of your tier, never euros. Its administrators also have the “Applications” tab, your applications' AI, in euros: see AI in your applications.
- At the top, the two periods in large, “over 5 h” and “over 7 days”. Your own share shows in yellow on the organisation's bar (“Yours”). When the pace of the last few days would fill the week, a sentence says so: “At this pace, the 7-day period will be full: Saturday 11:00 AM.” The pace is the average of the last three days of activity.
- “When you get it back” shows how the 5-hour period empties if nobody asks anything more: each request leaves it five hours after it happened. The next three releases are listed, for example “04:20 PM: −11%”.
- “Day by day” shows, over 7 or 30 days, the share of the week used each day, split by application or by kind of work: scoping, plan, implementation, repair (what the assistant fixes after failed tests) and summary. The dotted line marks a seventh of the week: above it, the day took more than its share. “Where it went” shows the same split as bars.
- “What a change costs” places each change published in the last 30 days by the share of the week it took, plan and repairs included, and gives the median.
- “When the team works” shows the usage by day of the week and by hour.
- “Request log” gives one line per request the assistant handled: when, the application (and the test version), the kind of work, the duration, the share of the week and the outcome (published, done, answer, interrupted, error…). The messages themselves never appear there. Everyone reads their own requests there.
A request counts for the person who made it, even in a colleague's conversation. The shares are worked out on your current tier: if the tier changes, past percentages change with it.
Your organisation's administrator also reads:
- the split by person and the log of the whole organisation;
- “Per member”: each one's share over 5 h, over 7 days and over 30 days, and their published changes. Someone who has left the organisation keeps their row, marked “former member”, while their requests count;
- “To watch”, when the organisation's 7-day period goes over 80%, or when a member has taken more than a quarter of the week.
An organisation without a limit reads “No limit for your organisation”: the charts there count requests.
Kedalia's operator (the global administrator) also reads the cost in euros, the number of tokens and the AI model of each request, and picks the organisation with the “Organisation” filter, at the top of the page, or from an organisation's page.
The e-mails at 80% and 100%
When the organisation's 7-day period reaches 80%, its administrators receive an e-mail, “The assistant at 80% over 7 days”, which says how long until the usage is fully back. At 100%, a second e-mail, “The assistant is waiting: the 7-day period is full”, says when the assistant starts again. The person whose request is the latest of the week receives it too. Each e-mail is sent once, and again when the period has gone back under 60% and then crossed the threshold again.
When a period is full
At 100%, the assistant accepts no new request and answers: “Your tier has no assistant usage left for now. It comes back little by little: try again later.” The gauge then says “Reached” and the time the assistant is available again, for example “available again on Thursday at 02:30 PM”.
The period slides: what was used more than 5 hours, or more than 7 days, ago no longer counts. Usage therefore comes back little by little, without waiting for a fixed date.
What counts
Everything the assistant does for your organisation: its answers to your messages, writing the plan, the implementation and the fixes it makes so that the tests pass. An implementation weighs much more than a question.
A very long step may also stop by itself: “The assistant stopped partway.” Then send your message again, or click “Continue the plan” or “Continue the implementation”.
If Kedalia restarts while the assistant is working, for an update for instance, the work already done is kept: the conversation says so, and the assistant usually picks up where it left off by itself. If it does not, click “Continue the plan” or “Continue the implementation”.
When a step stops for another reason, the conversation says which: the artificial intelligence service did not answer, the assistant's work was cut off, or the reason is unknown. What the assistant has already changed is kept, not saved yet: one button, “Continue the implementation” by the message box, picks up from there and finishes the plan. Another request on the application first, a new plan for instance, erases it, and the work is done again. Continuing counts in the usage as a new step, less if you continue within five minutes. When the artificial intelligence service did not answer, the assistant tries again once by itself after a minute.
Making the usage last
- Describe your whole need before asking for the plan: a plan that is right the first time saves an extra implementation (see Asking well).
- For a small touch-up, “Adjust directly” costs less than a new plan.
- Start each new change with “New”, at the head of the conversation: a long conversation costs more with every message.
- A long conversation left a few minutes without a message summarises itself, while the assistant still remembers it while counting less towards your usage: the conversation shows “Conversation summarised to stay fast”. The summary keeps your decisions, your words, the plan and what was done; the assistant reads the files again when it needs to. Without it, the next message would re-read the whole conversation and cost much more. The summary takes a little of your usage over 5 h; while it runs, usually under a minute, the assistant is at work: send your message once it is done. If the summary could not be made, the “Long conversation” (or “Conversation on standby”) banner shows above the message field: it says what the next message costs, and its “New conversation” button starts afresh from the updated application.
What depends on your tier
The amount of usage included in each period depends on your tier, like the number of team members. If you regularly run short, an administrator moves to a higher tier from the “Subscription” page (see The subscription).
On the Discovery tier, the assistant uses a simpler model: it may make more mistakes on a complex application. A banner at the top of each conversation says so. A higher tier gives you the full assistant.
On the Hosting tier
The Hosting tier has no usage period: each request to the assistant is deducted from your organisation's balance, at its cost plus a margin, like your applications' AI: see AI in your applications. When the balance is used up, the assistant answers: “Your organisation’s balance is spent: an administrator tops it up from the Applications tab of AI usage to go on with the assistant.” A top-up, from the “Balance” card of the “Applications” tab, brings it back at once. The organisation's administrators get the balance e-mails, when it runs low and then when it is used up, even with no application's AI switched on. There is no gauge under the message box, and the “Assistant” tab shows no period: it says the assistant is paid from the balance and, for an administrator, leads to the “Applications” tab.