Agents / HR Assistant

HR Assistant: answers your employees' questions on policies, leave and benefits — 24/7

MANAGED SELF-HOSTED 97.5% closed without a human

What it does

  • Answers questions on vacation days, sick leave, bonuses, raises and benefits — grounded in your actual policies.
  • Knows each employee's own context: "will my training be reimbursed?" gets a personal answer, not a policy quote.
  • Works where your people already are: web chat, a Telegram bot open only to your staff, or embedded in your intranet through an API.
  • Escalates to a human when a question is out of scope — and logs every conversation for HR review.
  • Speaks your employees' languages — EN, UA, PL out of the box — and copes with the way people actually write: slang, typos, half-sentences.
  • Shows which policy each answer came from, so an employee can check it and HR can trust it.
  • Takes the recruiting side too: reads incoming CVs in any format — PDF, Word, text, photos of paper — and writes clean, normalised fields into your ATS or CRM. No more re-typing.

What we'll need from you

Honest list — preparing this together is most of the project and it's why the agent answers correctly.

  • a. Your HR policies and procedures — in whatever state they're in today; structuring them is part of the work.
  • b. An export of employee records (roles, dates, benefit entitlements) — or read access to your HR system.
  • c. One person on your side who can answer "what's the actual rule here?" during data preparation.
  • d. A pilot group of 10–20 employees willing to ask it real questions for two weeks.
  • e. For CV parsing: 30–50 real CVs from your pipeline (anonymised is fine) and the fields your ATS expects.
The economics

A person's capacity ends at about 25 dialogs a day. The agent's ends at your budget.

HR generalist
≈€2.50per dialog
Capacity~25 dialogs a day
Grows with volume byhiring people
Virtual HR
≈€0.10per dialog
Capacityno ceiling, 24/7
Grows with volume bycents per dialog
Implementation is paid once — data preparation, integrations, pilot. After that the cost grows only with usage, never with headcount.
Dialog = up to 6 questions and 7 answers. Agent: Claude Sonnet 5, ~7.5k input and ~400 output tokens per answer (policy excerpts, employee profile, history) — ≈€0.08 with prompt caching, ≈€0.13 without. Human: ~10 minutes per dialog.
Managed

Runs on my EU infrastructure; answers are generated through the Claude API under a DPA. Fastest to launch — you pay per dialog.

Self-hosted

Deployed inside your perimeter on a local model — employee data never leaves the company. Hardware: Mac mini M5 Pro with 64 GB, ≈€3,100 one-time.

Live demo
Ask this agent your own questions

The demo runs on a fictional company — its HR policy and the card of one employee. Access is per person: write me and I will send you a login, usually the same day.

UNDER THE HOOD retrieval over your structured policies · local or cloud language models, benchmarked on your cases · regression-tested on 30 fixed questions across 13 categories before every change · session history · prompt-injection protection · full conversation logs · CV parsing into a fixed schema, export to ATS/CRM
See it answer real questions
A live demo on sample company data — 30 minutes.
Request a demo