What your team can ask the PeopleForce AI chatbot, and what’s coming next

It's 6:12 p.m. and the message reads: "Quick one: how many vacation days do I have left?" It isn't quick, it isn't one, and it isn't the first time today. The questions that fill an HR inbox are rarely hard; they're just constant, and every one of them interrupts something that actually needed you.

It’s in early access: the PeopleForce team switches it on for your account, then your admin decides which roles get it: HR only, managers too, or everyone. Routine questions can go to the chatbot first, before they reach you.

What the AI chatbot is, and where it lives

AI Companion is an AI chatbot that lives in a chat panel inside PeopleForce. A launcher with a sparkles icon sits beside the search box in the left sidebar on every page; clicking it opens the panel docked to the right, where it stays as you move around the product. It can be minimized to a small strip in the corner mid-conversation, keeps your 50 most recently updated chats (each titled automatically from your first message) and streams its answers live, showing what it's doing as it works: "Searching knowledge base", "Checking who's away", "Drawing chart".

The subtitle under the panel's name says what it is in three words: your HR assistant. It's an HR chatbot that answers from your company's own HR data.

The five questions that fill HR's inbox

Much of an HR inbox is the same few questions on repeat. Here are five the chatbot handles today, with prompts that work.

"What does our policy say?"

The chatbot searches your knowledge base semantically, so a paraphrased question still finds the right article: ask "can I take time off when my kid is sick?" and it lands on your parental and carer's leave policy even though the words don't match. It quotes or summarizes the actual policy wording and shows clickable Sources under the reply, so the reader can open the original document instead of trusting a paraphrase. Results are re-checked against the asker's permissions before the AI ever sees them; a restricted document can't leak into an answer.

The citations are also your error-catching mechanism. When an answer looks off, the source link is right under it, and the original document settles it in one click. And when the knowledge base simply doesn't contain the answer, a visible gap is a useful signal: it tells you which policy to write next. Answers are only as good as the knowledge base under them, so whoever owns your policies owns the chatbot's quality too.

The answer also knows who's asking. The chatbot sees the asker's position, department, location, manager, seniority, and employment type, and folds them into the search, so a notice-period question returns the variant that applies to that person's country and role rather than a generic average.

"Do we have someone who speaks German?"

People search in plain language: find a person by name, filter by department, location, position, hire date, or custom fields, get counts by team, or ask "who reports to me?" The question that used to mean exporting a spreadsheet and filtering it ("does anyone on the team speak German and sit in Warsaw?") becomes a sentence.

"Who's out this week?"

The Monday question. Instead of cross-referencing calendars, ask the chatbot: "who's away next week on my team?" The list comes back in seconds, scoped to what you're allowed to see.

"How many days do I have left?"

Leave balances, at the exact moment the question occurs, including 6:12 p.m. before booking flights. The employee asks the chatbot instead of messaging you; booking the actual leave still happens where it always did: the "+" (Quick add) button in the top bar → Request time-off, one click away. The chatbot can also tell you what happened to a request (pending, approved, rejected or withdrawn) or count the sick days you've taken this year.

"Can I get that as a chart?"

Ask for a breakdown and the chatbot draws it inline (bar, pie, or line) and formats multi-attribute answers as compact tables. "People per department as a bar chart" produces the chart in the conversation, no report builder involved. One cap worth knowing: a chart plots at most 50 data points, and past the cap it simply draws the first 50 and drops the rest. Weekly data for a full year is 52 points, so the tail gets clipped. Ask by month and the same year fits in 12.

Beyond these five, the chatbot reads three more things for the person asking: what's waiting in their own Inbox (approvals, tasks, surveys, reviews to write, documents to sign, courses), their 1:1s with the shared notes, talking points and action items, and the objectives they're allowed to see. Ask "what's waiting for my approval?" or "what did we agree in my last 1:1?" and the answer comes back in the same panel. Private 1:1 notes stay out of it.

Every answer is scoped to the asker's permissions

What the AI chatbot won't do yet, and what keeps your data safe

Here is where the line sits today.

The chatbot is in early access, and this first release is scoped to looking things up.

None of those abilities were built into it, so it won’t pretend it can. When a request needs an action, it says so plainly. If it knows where you’d do it yourself, such as booking time off, it tells you. If it doesn’t, it says so rather than guessing at a page name.

A few edges worth knowing before you roll it out

What keeps your data safe

Where it’s going

Acting on your data is the direction we're taking the chatbot, and we're actively working on it. What stays constant as it gets there is the security model underneath.

The assistant inherits your access model rather than getting its own.

Your processes and conversations

• Nothing changes behind your back: your approval chains and your records stay exactly as your admins configured them, because today the chatbot has no way to touch them.

• Your conversations are saved to your own account so you can pick them up later; the chatbot keeps your 50 most recently updated chats.

• A conversation can’t be deleted from the panel today.

• One reply makes at most 12 lookups, so split a very large question into smaller ones.

Already using Claude or ChatGPT? Connect it over MCP

The AI chatbot is for everyone inside PeopleForce. Some of your people live in an external AI assistant all day, and for them PeopleForce offers a second door: its own MCP server. MCP (Model Context Protocol) is the open standard that assistants such as Claude and ChatGPT use to reach business systems.

The point isn't to compete with the assistant your team already uses. Your HR data stays governed here and reaches their agent on your terms, instead of being pasted into a chat window as free text. What the assistant receives becomes part of the conversation there, under that assistant's own data settings.

The AI chatbot

Choose it if you want answers inside PeopleForce for everyone, with zero setup per user. It also reads your published knowledge base and draws charts in its replies.

The MCP integration

Add it if part of your team already works in an external AI assistant and should query people data from there. Over MCP, an assistant does not read the knowledge base or draw charts.

Running both is a valid answer; they serve different people in the same company.

Switching it on

1. Ask your Customer Success Manager (CSM) to enable AI Companion. It's an early-access feature, switched on centrally per account by the PeopleForce team; there's no self-serve toggle.

2. Grant the role permission. Settings → Roles → the "AI" section → "AI companion". Any role can hold it (HR only, HR plus managers, or everyone), except external recruiters, who are always excluded.

That's the whole rollout. People find the sparkles button beside search, and the panel greets them by name.

Availability and caveats

Built on the HR data you already keep

You don't train the chatbot or upload anything for it. It answers from what Core HR already holds: your org structure, who reports to whom, your published policies, and the permissions behind every role. When that data changes in PeopleForce, the answers change with it.

FAQ

Can the AI chatbot change or submit anything?

Not yet. It has no ability to send, submit, approve, create, schedule, or notify. None of those tools were built into this release, so it tells you plainly and points you to the screen where the action happens. We're actively working on letting it act on your behalf, and this article will be updated when that ships.

What happens when it doesn't know the answer?

Policy answers cite their sources, so a wrong or thin answer is visible and checkable in one click. If the knowledge base doesn't cover a question, that gap is your signal about which policy to document next.

Can it show someone a salary?

It checks the same compensation permission the web app checks (the one that decides whether someone can open an employee's Compensation tab), and it checks it person by person. An employee sees their own current base pay only if their role grants them compensation access to their own profile. The chatbot shows only the current base amount and currency, not the compensation history, additional pay, or bands the Compensation tab holds.

Where do the answers come from?

From your own PeopleForce data, nowhere else: the knowledge base, people and leave records, your Inbox, 1:1s and objectives. Policy answers quote the wording and cite the source article; results are permission-checked before the AI sees them.

Who in the company can use it?

Any role your admin grants the "AI companion" permission to, external recruiters excluded. You can start with HR, then widen to managers and everyone else.

Is the AI chatbot in the mobile app?

Yes. It's also in the PeopleForce mobile app, with the same permissions as on the web. The screens in this article show the web version.