A multi-agent system for HR
that works like a teammate.

An AI agent is a system that understands the full context, reasons about the situation, plans toward a goal, acts autonomously within the limits people define, and learns from outcomes. In WorkVell, the “Vell” orchestrator understands your request and routes and combines work across specialized agents for attendance, payroll, expenses, recruiting, and more. It doesn't just answer questions — it moves the work forward.

Chatbots can't “think.” They respond to inputs using pre-programmed logic and can't learn from past experience or adapt to unfamiliar situations. Vell interprets the intent behind a request, breaks it into tasks, delegates them to specialized agents, and reasons over your company data and labor rules to act.

Yes. Beyond answering questions, Vell can draft attendance corrections, prepare leave and overtime requests, flag issues on expense claims, and run first-pass candidate screening — all within the permissions it's been granted. It takes on routine requests and streamlines complex processes, always under human approval and visibility.

It works in a loop: ① understand intent → ② route to the right agent(s) → ③ execute tools → ④ respond with sources. The Vell orchestrator reads the intent and conditions (period, department, etc.) and dispatches work to domain agents — attendance, payroll, expense OCR, knowledge, recruiting, analytics, drafting, and translation — then merges the results and answers with citations. Because each agent is specialized rather than one general-purpose model, accuracy in both answers and actions is higher.

Yes. It ingests work rules, in-app guides, announcements, training materials, and uploaded documents as a knowledge base, and answers from them with citations (RAG). For a question like “What's the parental-leave application process?”, it answers from your internal source of truth — not guesswork.

Yes. The expense OCR agent reads the photo, extracts amount, date, and merchant, auto-suggests a category, and drafts the claim (OCR is an add-on). Just say “File this receipt as an expense” and it gets started.

Yes. Remaining paid leave, today's/this month's attendance, overtime, your next shift, pending approvals, and open tasks are all available instantly in natural language. These run as direct queries against your data, so they're fast and incur no unnecessary AI cost.

Yes. The drafting agent helps write approval memos, company-wide announcements, and internal/external messages. The translation agent renders notices, policies, and pay slips into up to 26 languages while preserving domain terminology — so multinational teams share the same information in their own language.

Yes. Say “Schedule an interview next Wednesday at 3pm” and Vell interprets the intent, extracts the date/time and title, and creates the calendar event (JST-aware). It doesn't just answer — it carries out the action.

Every answer cites its sources — which records, policies, or requests it's based on. Vell operates only inside three guardrails: permissions, sources, and audit. It never touches data a user can't see, and every action is recorded in an audit log.

Role permissions are strictly enforced; users without access never receive that data. We do not use customer data across tenants to train models. Specific personal identifiers such as My Number are encrypted and audited.

We don't lock to a single model — Vell autonomously routes each task to the most suitable model (adaptive model selection), using different models for conversation, summarization, translation, and document reading to balance accuracy and cost. On top of that, navigation search, personal-data lookups, and a cache of common questions run locally without calling an LLM — so a model is used only when it's actually needed, keeping running costs low.

Yes. Vell reasons with an understanding of country-specific labor law — Japan's Article 36 limits and the karoshi line, Korea's 52-hour week, and more — so distributed organizations get decisions aligned to each site's rules.

It's ready to use out of the box, and can be customized in depth to match your workflows, permission design, and approval flows. It keeps improving with usage after rollout.

It augments rather than replaces. By taking on routine requests, Vell frees people to focus on higher-value judgment and human interaction. Decisions and final approvals always stay with people.

An analytics dashboard surfaces each agent's volume handled, time saved, and usage, so you can see which capabilities create the most value and act on it for investment and operational decisions.