Background work and Activity
What runs in the background, what runs on its own schedule, and what happens when a budget is reached.
Most of what CareerBot does takes minutes: reading a story, filling in a few hundred companies, ranking every open role, writing a tailored resume. It all runs in the background, so you can keep working or close the tab. What it produces appears as it's ready.
Activity
Activity sits at the bottom of the sidebar (in More on a phone). It shows how many things are running, with a thin progress bar, and opens into a list:
- Running work, with how far along it is where that's known ("24 of 38"). Cancel stops it: work not started yet is taken off before it runs; running work stops after the step it's on, which is still charged.
- What failed, with what failed ("Couldn’t read your Brightwater story", "Couldn’t check roles") and the reason under it, and what you cancelled, with how many steps it finished. Try again starts it again where that's possible, at about what the first try cost.
- What finished in the last day. Clear finished takes finished work off the list (with Undo); it's never deleted.
Under the list is this month's spending against your budgets. Waiting work shows why: "Waiting to start", "Paused until your AI budget allows" or "Paused until your Apollo budget allows".
An AI reply that comes back unreadable, or so long it's cut off, is asked for once more. That second call is paid like any other and counts in your spending under the same task. If it fails too, the work fails with "The model's reply wasn't readable. Try again." (or "…was too long and got cut off. Try again.").
Budgets pause work, they don't fail it
You set a monthly AI budget and the Apollo credits you set aside in Settings, under Budgets and spending. No AI work runs until an AI budget is set.
Before each paid call, CareerBot reserves what the call could cost. If that would go past what's left, the call doesn't run and the work pauses. Paused work isn't a failure: it picks up where it left off when you raise the budget, and on the 1st of each month (UTC), when budgets start again. Nothing that already finished is paid for twice.
Apollo has three modes, under Automated Apollo work:
| Mode | What it does |
|---|---|
| On | Finding companies, their details and job postings runs on its own, within the credits you set aside. |
| Only when I ask | Apollo is used only for what you start yourself. |
| Paused | No Apollo work runs. |
An Apollo call is never made past your Apollo account's balance, and nothing is spent when the balance can't be read.
What runs on a schedule
- Every morning, 9:00 UTC
- The job boards of every workspace with Targets or Maybe companies are read again for new and closed roles, and new roles are sorted and ranked. A company whose roles matching your titles opened or closed has its fit judged again, at about a cent for every 40 companies from your AI budget, unless Update fit as roles change is off.
- Every day
- With Google Drive connected, your CareerBot folder is compared with what it should hold, and synced only when something differs. Changes to documents, contact details, layout and sources also sync soon after they're made.
- Every five minutes
- Work cut off by an update or a crash is picked up again quietly, up to three tries. After that it fails with "This kept getting interrupted. Try again."
- Every hour
- Model prices are read again from OpenRouter, so what a call reserves follows the providers' prices.
- Every night
- Paid calls whose final cost wasn't known at the time are settled from OpenRouter's and Apollo's own figures, so your spending adds up.
- The 1st of each month
- Budgets start again, and work paused for a budget picks up again.
What starts on its own, and what waits for you
Some steps follow from what you do; the rest run only when you ask.
On its own:
- After a company search, new companies are screened and then filled in.
- When you approve a direction that has nothing filled in, its positioning, titles and criteria are filled in (Approve shows about what that costs).
- When a story or project has nothing left to review, your skills, tools and certifications are gathered and insights are looked for again. Approving one of the story's roles after that gathers again. The decision that starts it shows about what it costs. Changing a fact you already approved (an edit, a rewrite, rejecting it) starts neither.
- When you rate a company or approve, change or switch a limit, the list of roles is brought in step at once (no new AI calls).
- The daily roles pass, with updating company fit as roles change, and Google Drive syncing.
Only when you ask, each with its estimate shown first:
- Reading a story, reading it again, and reading a project.
- Read my goals, Fill in (again) on a direction, Fill in all, and Suggest more.
- Find companies, Check again on a company, and Check all again.
- Check roles again and Rank again.
- Writing resumes, letters, answers, messages and follow-ups; Check against facts and Update lines.
- Gather on Skills, finding duplicates and finding the same work.
Only one company search, one fill-in pass and one roles pass run at a time in a workspace. A company you ask to check again while a roles pass is running is read before that pass ends.
Edit this page on GitHub