# AI agents for HR and payroll | Peoplo.ai URL: https://peoplo.ai/in/en/ai-agents/ > Peoplo.ai agents check payroll, plan onboarding, prepare full and final settlements, read receipts and answer policy questions. You make the decisions. People, companies and figures in product examples on this page are sample data, not customers. AI agents # AI agents on your HR team. They prepare. You decide. They watch for events, read your records, do the checks and draft the work. Then they stop and ask before pay, tax, people records or money change. [See the agents in a demo](https://peoplo.ai/contact/) Meet the agents Your AI team, right now Illustrative status board Payroll anomaly check Watching Onboarding coordinator Working Exit coordinator Waiting for you Receipt reader Done Policy assistant Watching Weekly brief Working How an agent works ## Every agent follows the same five steps 01 ### Trigger Something happens: an offer is accepted, a claim arrives, month-end comes round. 02 ### Read The agent reads your records, policies and rules, with the same access as the role it works for. 03 ### Prepare It does the checks and the maths, and drafts the documents. 04 ### Ask you It shows you exactly what will change, and waits. 05 ### Record Your decision is applied. Important changes are written to an audit log: who did what, and when. Meet the agents ## One job each, and a clear limit Pick an agent to see what sets it off, what it does, what it hands you, and what it will never do. Trigger October payroll calculated for 48 employees What it does - 1 Read salaries, pay components and revisions - 2 Compared every payslip with September - 3 Checked PF, ESI, PT and TDS for each work state - 4 Found 1 person with no salary set - 5 Flagged an overtime amount twice the usual You approve Locking the run Never Lock or pay a payroll run. Payroll anomaly check Sample October payroll · 48 employees 3 changes since September PF went up for 12 people with the new ₹25,000 ceiling 1 flag: overtime for Rahul Verma is twice his usual Waiting for your approval Review and lock Trigger Arjun Mehta accepted the offer What it does - 1 Picked the Design onboarding template - 2 Set due dates from the joining date, Mon 2 Nov - 3 Assigned tasks to Arjun, his manager and HR - 4 Drafted the laptop and email request for IT - 5 Chose a buddy from the same team You approve The plan, before it goes out Never Send anything to a new joiner without approval. Onboarding coordinator Sample Arjun Mehta joins Mon 2 Nov Day-one plan with his manager and buddy 6 documents to collect, including PAN and UAN Laptop and email request for IT Waiting for your approval Approve and send Trigger Sara Khan’s last day set to Wed 25 Nov What it does - 1 Read monthly gross from payroll - 2 Counted 6 days of encashable leave - 3 Checked IT assets: laptop and ID card still assigned - 4 Worked out the full and final settlement - 5 Drafted the relieving and experience letters You approve The settlement Never Pay a settlement or revoke access on its own. Work out the same figures yourself: [Leave encashment calculator](https://peoplo.ai/in/en/tools/leave-encashment-calculator/) · [Gratuity calculator](https://peoplo.ai/in/en/tools/gratuity-calculator/) Exit coordinator Sample Sara Khan · last day Wed 25 Nov Full and final settlement worked out Laptop and ID card on the return list Relieving letter drafted Waiting for your approval Approve settlement Trigger Neha Iyer attached a receipt to a claim What it does - 1 Read the merchant, date and amount - 2 Matched the category to Travel - 3 Checked the amount against the travel policy - 4 Converted at October’s rate - 5 Filled in the claim line You approve The claim Never Approve or pay a claim. Receipt reader Sample New claim from Neha Iyer Cab fare · ₹640 · Travel Date matches her client visit Within the travel policy limit Waiting for your approval Approve claim Trigger An employee asked about carrying leave forward What it does - 1 Found the Leave policy 2026 - 2 Read the carry-over section - 3 Looked up the employee’s own balance - 4 Worked out what lapses on 31 Dec - 5 Answered with the source You approve Nothing. It answers, it never changes data Never Change any record. It only answers. Policy assistant Sample Can I carry unused leave into next year? Up to 10 days of paid leave carry over You have 12.5 left, so 2.5 lapse on 31 Dec Source: Leave policy 2026 · Carry-over No approval needed Answered Trigger Monday morning, week of 26 Oct What it does - 1 Counted leave, joiners and leavers - 2 Listed approvals waiting more than 3 days - 3 Checked statutory due dates - 4 Summarised the week for HR and managers - 5 Sent the brief You approve Nothing. It only reports Never Act on what it finds. It only reports. Weekly brief Sample Week of 26 Oct 4 people on leave, 2 joiners, 1 leaver 7 approvals waiting, 2 of them for more than 3 days TDS for October is due on 7 Nov No approval needed Read Names, dates and figures are illustrative. More agents ## More agents that prepare the work ### Payroll pre-run check Before a run exists, it looks at the month, reconciles attendance, leave and joiners, and explains pay changes. It never locks payroll. ### Attendance check It offers employees a filled-in correction for a missed punch, which they send themselves. ### Joiner and leaver handovers Each joiner or leaver opens one Help Desk case per team, with deadlines and one status view. For a leaver, it prepares final settlement drafts for HR to approve. ### Filing prep Peoplo.ai prepares the statutory files and reminds HR of due dates. A person downloads and submits them. ### Policy gaps Each week it groups Help Desk questions that had no answer and drafts the missing paragraph. HR edits and publishes it. ### Appraisal cycle runner It moves date-driven phases, chases late reviewers and flags calibration outliers. A person finishes calibration and closes the cycle. ### Monthly HR summary and drafts HR gets a monthly summary, and AI drafts employment letters and policies for HR to review. ### Custom agents HR describes a job in plain words, and Kili builds it from a fixed set of checks. It starts switched off and a person turns it on. Custom agents can check and remind, nothing more. Control and visibility ## See what agents do, and decide what stays with a person ### An AI Agents hub "Your team" shows what each agent is running, what it found and what is next. "Browse and create" finds an agent by HR area. HR can rename an agent and set its tone. ### Time saved and risks caught A Time saved card and a monthly report show the estimated hours saved and the risks caught before they reached the bank or ministry. ### Facts for approvers Approvers see a facts brief on vendor bills, exits, pay changes, attendance corrections and onboarding tasks. Each brief lists the checks and doesn’t recommend a decision. ### Leave auto rules HR writes rules, and matching leave requests are approved without a person. A manager or HR can reverse an approval within the time set in the rule. ### Ask an expert When an agent or Kili isn’t sure, HR can send the question to a person at Peoplo.ai. Only what HR writes is sent. ### Agent record for auditors Admins can export the agents’ work as a CSV. Important changes are also written to an audit log: who did what, and when. ### Approval for AI assistants AI assistants such as Claude or ChatGPT, connected over MCP, can propose changes, and the person approves them in their inbox. New API keys made in the app have this on by default. ### Kili knows Peoplo.ai too Kili answers questions about Peoplo.ai itself from the help articles, with a link. The help articles are in all 8 languages, in beta. Guardrails ## Powerful agents, with people in charge ### Approval before action Nothing that changes pay, a person’s record or money happens until a person confirms it. ### Your permissions, not more Agents and Kili work within the access of the role they act for. The password vault is never available to AI. ### Approvals on the record Important changes are written to an audit log: who did what, and when. ### An off switch for each agent Admins choose which agents run. Peoplo.ai works as a normal HRMS without them. How do other Indian HR products describe their AI? See the [best AI HRMS software in India](https://peoplo.ai/in/en/best-ai-hrms-software-india/) , side by side. Approval required for changes to pay, tax, people records and money. Payroll anomaly check Check every run before it can be locked Exit coordinator Prepare settlements when a last day is set Receipt reader Read receipts on new claims AI runs used this month 1,284 / 2,500 Trust ## You set the rules for the AI Decide which agents run and what they may prepare. Changes to pay, people records and money always wait for a person. Built for GDPR, SOC 2 and India’s DPDP Act. **Encrypted at rest** Personal data is encrypted in storage. **Two-step sign-in** Protect every account, and sign in with Google or Microsoft. **Audit log** Important changes are written to an audit log: who did what, and when. **DPDP and GDPR tools** Export or erase a person’s data when they ask. Ask Kili ## Answers from your own policies and records Employees, managers and HR ask in plain words. Kili answers from your company’s data and shows where every answer came from. Employee asks How many paid leaves can I carry into next year? Up to 10 days. You have 12.5 left, so 2.5 days lapse on 31 Dec unless you use them. Source: Leave policy 2026 · Carry-over Employee asks Can I claim my broadband bill? Yes, up to ₹1,500 a month under the remote work policy. Attach the bill to a claim and the receipt reader fills in the rest. Source: Expense policy · Remote work Employee asks Which tax regime am I on? The new regime, chosen in April. Declarations open again in January if you want to review it. Source: Your record · Tax declarations Manager asks Who on my team hasn’t done this week’s check-in? 2 of 7: Vikram Shetty and Omar Haddad. Want me to send them a reminder? Source: Check-ins · Week 44 HR asks What was our attrition last quarter? 3 people left out of an average of 48, which is 6.3% for July to September. Source: Reports · Joiners and leavers HR asks Who is due for a probation review this month? 2 people: Arjun Mehta on 12 Nov and Sara Khan on 25 Nov. Want me to remind their managers? Source: Employees · Probation Sample questions and answers. Bring your own AI ## Works with Claude and ChatGPT Each person can connect Claude Desktop, Claude Code or ChatGPT to Peoplo.ai over MCP with their own API key. The assistant sees only what their role allows, asks before making a change, and changes it makes are logged under that person’s name. Personal API keys you can revoke any time Same permissions as the app The password vault is never available to AI [How to connect ChatGPT or Claude](https://peoplo.ai/in/en/ai-agents/connect-chatgpt-claude/) AI assistant · connected to Peoplo.ai Sample Who on my team is on leave next week? ↳ Peoplo.ai · team leave calendar · read only Two people: Ananya Rao, Monday to Wednesday, and Vikram Shetty, Thursday afternoon. One request is still pending your approval. Approve Ananya’s request. ↳ Peoplo.ai · approve leave · change, asking first AI runs ## AI included in the plan An AI run is one AI-assisted action, such as a payroll check or reading a receipt. Fair use, shared across your company About 5 AI runs per employee a month. We warn you at 80%. At 100%, AI features fall back to working without AI until the next month. [See pricing](https://peoplo.ai/in/en/pricing/) Questions ## Frequently asked questions Straight answers about Peoplo.ai for Indian companies. ### Do Peoplo.ai agents change data on their own? Not pay, people records or money. Agents prepare drafts, run checks and flag issues, and anything that would change pay, a person’s record or money waits for a person to approve it. If an admin allows it, an agent can send reminders and add notes by itself. Agents that only answer or report, such as the policy assistant and the weekly brief, never change data. ### What happens if an agent gets something wrong? You see exactly what will change before it is applied, line by line. You can edit the draft, reject it, or approve it. Nothing is applied until you do, and your approval is written to the audit log. ### Can we turn agents off? Yes. Admins choose which agents run from AI controls, and can switch any agent off at any time. Peoplo.ai keeps working as a normal HRMS without them. ### Who can see what an agent did? Admins and HR. Important changes are written to an audit log: who did what, and when. ### What is an AI run? One AI-assisted action, such as a payroll check, an onboarding plan, reading a receipt or answering a question. AI is included with fair use shared across your company: about 5 AI runs per employee a month in India. ### What is the difference between agents and Kili? Agents work in the background on one job each, triggered by events such as an accepted offer or a new claim. Kili is the assistant you talk to: employees, managers and HR ask questions in plain words and get answers from company records, with the source shown. ### Can we use Claude or ChatGPT with Peoplo.ai? Yes. Each person creates their own API key and connects Claude Desktop, Claude Code or ChatGPT over MCP. The assistant sees only what that person’s role allows, changes are marked so it can ask first, and the password vault is never available to it. ## Put the agents to work on your data We’ll run them on your policies and payroll rules in the demo. [Request a demo](https://peoplo.ai/contact/) [Start free trial](https://app.peoplo.ai/signup)