How the job finder works
A run is a fixed sequence of named steps, not one big question to a model. Three of the four steps cost nothing at all - only the scoring is expensive, and that is exactly where the caps bite.
Searching through the same path as the job scanner
The agent has no second search path. It calls the same scan you can start by hand: watched career pages and your saved searches in the German Federal Employment Agency job board. Whatever is filtered out there never arrives here.
Pre-sorting by rules, not by feel
From the findings the agent takes the newest ones, drops whatever is on your blocklist and cuts the list down to your daily cap. This step is pure arithmetic: no model, no cost, and you can follow it.
Scoring against your master profile
Every remaining posting goes through the same evaluation as a job check you start yourself - percentage, criteria and the questions that may come up in an interview. After each single score the run saves its state, so an abort does not cost the whole batch.
A posting that cannot be read is counted and skipped. The run does not stop over it.
Reporting what clears your threshold
At the end there is a summary in your agents area, and an email or a push message if you want one. Reported is what clears your threshold - everything that made it through the pre-sorting was scored.
Example view
- Searching through the same path as the job scanner
- Pre-sorting by rules, not by feel
- Scoring against your master profile
- Reporting what clears your threshold
From the findings the agent takes the newest ones, drops whatever is on your blocklist and cuts the list down to your daily cap. This step is pure arithmetic: no model, no cost, and you can follow it.
Documents, if and when you want them
The second agent does not work on a hunch. You hand it a list of finished job checks and it prepares exactly those - twenty at most in one go.
You pick the applications
What documents get built for is your call. The agent does not decide it. Every application you hand over is checked first: is it yours, and is its evaluation finished? Otherwise nothing runs.
Tailoring the resume, writing the cover letter
The resume is tailored to the posting and the cover letter written to match - through the same generation path the 1-click application uses. The document language follows a fixed rule, not the model.
A fact check that flags instead of rewriting
A mechanical check then compares cover letter and resume against your master profile and marks whatever is not in there. The findings sit with the application. Nothing is edited in - "not evidenced" is not "disproven", and that distinction belongs to you, not to a model.
Attachments from your own selection
Which references travel along is in your settings and never comes out of a model. If you have never picked, all references are the default; if you deliberately picked none, none go.
- You pick the applicationsDocuments
- Tailoring the resume, writing the cover letterDocuments
- A fact check that flags instead of rewritingDocuments
A fact check that flags instead of rewriting
What keeps you from being surprised
An agent that runs without you needs hard limits - otherwise it is a risk, not a tool. There are four, and they hold independently of each other.
Your explicit consent first
Turning it on the first time asks for its own consent, and the server refuses to start without it - not just the button. No agent runs before that, at night or otherwise.
A daily cap you set yourself
You decide how many scorings a day may cost; your plan caps that value on top (currently 20). Two runs of the same kind are at least 6 hours apart.
A quota that cannot be overdrawn
Autopilot brings 150 job checks and 150 applications per billing period. Pro gets 3 complete runs to try things out - once, for the lifetime of the account, not monthly. When a pot is empty the run stops instead of quietly billing on.
A kill switch that closes when in doubt
Above all of it sit two cost ceilings and a switch that halts every run at once. If its state cannot be read it counts as pulled - in doubt nothing runs, rather than in doubt everything.
- A quota that cannot be overdrawnControl150
- A daily cap you set yourselfControl20
- Your explicit consent firstControl
A kill switch that closes when in doubt
What the agents do not promise
Automatic here means the groundwork happens without you. It does not mean anyone applies. Four limits that belong on this page rather than in a footnote.
You do the sending
Jobvin never sends an application to an employer. That is a decision, not a missing feature: an application arriving from an address that is not yours reads as unprofessional. You get the finished files and the text blocks and send from your own mailbox.
No promised cadence
What is reliable is the run you start yourself. This page does not promise you a fixed rhythm in the background - that would hang on a schedule outside the product, and a promise someone else would have to keep is no promise.
The fact check corrects nothing
It marks what is not in your profile and tells you where to look. It deletes no sentence and rewrites none. Whatever ends up in your application, you have read and released.
Agents are part of Pro
Turning them on, starting a run and seeing the results are tied to Pro, and that is checked on the server. The full quota comes with Autopilot; Pro has the trial run. Without either you still have the job scanner and the job check by hand.
- You do the sending
- No promised cadence
- The fact check corrects nothing
- Agents are part of Pro
Automatic here means the groundwork happens without you. It does not mean anyone applies. Four limits that belong on this page rather than in a footnote.