A chat inside PDVS: how to submit a request, what awaits approval, where to find a menu item – answers from the company knowledge base with live settings. When asked, the assistant tells the user their own leave balance, schedule and request statuses, drafts a training course, and gives no data about other employees. The provider is Azure OpenAI, OpenAI or a local company model.
Key functions and business benefits
Artificial intelligence with rules – boundaries, limits and a log.
- 01A chat with the knowledge baseThirty Lithuanian topics – sign-in, requests, substitutions, approvals, SLA deadlines, training, business trips – with live company values (permitted negative balance, approval deadlines) and menu paths based on the user's permissions; what it does not know, it does not guess.
- 02The user's own dataLeave balance as on the home page, the work schedule by its latest version, submitted requests and what awaits my action – only when asked; the assistant gives no data about colleagues or subordinates even to a manager, and mentions sick leave neutrally, without a code.
- 03Work on documentsCandidate CV reading into record fields, trip receipt reading into expense fields, an RFQ e-mail draft for purchasing, position-hierarchy and role suggestions – a person checks every suggestion before saving.
- 04A training course draftOn request in the chat, a course is generated by a strict schema – title, code, sections, a test with questions; the course is created only once the user confirms the card, and the right to create content is checked both when listing tools and when executing.
- 05A provider of your choiceAzure OpenAI in an EU region, OpenAI or a model running on the company's server via vLLM or Ollama with an OpenAI-compatible interface – requests never leave the company network; a connection and tool-call test from administration.
- 06Limits, budget, logA daily message limit per user, a monthly token and cost budget, a conversation retention period, a privacy notice, a token report with prices; personal-tool results are not stored in the database.
The flow
Question → tool → answer – every step with a permission check.
- 01QuestionA conversation list, suggestion chips · a daily limit · a privacy bar
- 02ToolsA knowledge-base topic · live values · personal data only when asked
- 03AnswerMenu paths by permission · “I don't know” instead of guessing · referral to HR
- 04AccountingTokens and cost · budget brakes · a retention period · audit
AI assistant (Azure OpenAI / local model)
Integration active · last successful connection 2026-10-24 08:12| Week | Conversations | Input tokens | Output tokens | Cost |
|---|---|---|---|---|
| 40 | 71 | 248 300 | 31 900 | 0,00 € |
| 41 | 84 | 296 100 | 38 400 | 0,00 € |
| 42 | 79 | 281 700 | 35 200 | 0,00 € |
| 43 | 78 | 262 900 | 33 800 | 0,00 € |
Personal-tool results are not stored in the database; conversations are deleted when the retention period ends. A local model costs the client 0 €, tokens are still counted.
The point
Artificial intelligence in an HR system is useful only when it answers from your rules rather than from the internet, and does not carry out data it should not. The PDVS assistant gets one user, their permissions and the company knowledge base; the provider – in the cloud or on your own server – is chosen by the administrator.
Related solutions
Recruitment and hiring
Candidate pipeline, AI CV parsing, Teams interviews, job offers, conversion to employee.
Read more → M-12Training management
Courses and tests in the system, mandatory training with deadlines, position training plans, PDF certificates, seven reports.
Read more → M-10Business trips
Request to travel, per diems by country, an automatic order, advances for accounting, expenses with receipts and the expense report.
Read more →