ai
Cheap models, visible suggestions, human decisions
HireFlow uses AI where it saves you time, labels it wherever it appears, and never lets it act alone.
The rules AI follows here
Suggestions only
No automatic rejection, no AI text sent without a person, no final grade without someone confirming it.
Redacted first
Email, phone, address, date of birth and salary are removed before any model sees a resume or an answer.
Always labelled
Anything AI produced carries an “AI suggested” tag, in the colour HireFlow reserves for machine output.
Low-confidence goes to you
Below 60% confidence a suggestion is shown as “grade this yourself”, not pre-filled.
Low-cost routing
Writing tasks run on Claude Haiku with DeepSeek as the fallback; you can flip the order per task. Small typed judgements (a rubric score, a yes/no) run on the same models — or, where your company opts in, on TypeSafe Jev.
| Task | Default | Typical cost |
|---|---|---|
| Generate a 20-question assessment | Claude Haiku → DeepSeek | about 2.5¢ (Haiku) · about 0.6¢ (DeepSeek) |
| Suggest a grade for a written answer | LLM decision (or Jev) | a fraction of a cent |
| Fill gaps in a parsed resume | LLM decision | a fraction of a cent |
| Summarise a debrief | Claude Haiku | under a cent |
Budgets and your own key
Set a monthly AI budget. At 80% owners are told; at 100% everything pauses except grading suggestions. On the free plan, AI runs on your own Anthropic, DeepSeek or OpenAI-compatible key, stored encrypted and never shown again. Without a key HireFlow still works — assessments use built-in templates.
Start with the free plan
One job, three teammates, 200 applicants a month. No card, no time limit.