22 Aug AI Agent in MintHCM – an intelligent assistant for the HR world
The rise of artificial intelligence tools is no longer just a technological curiosity – it’s becoming a real support system in everyday work. Within MintHCM, a new project has been launched – the AI Agent the first free and open source AI chatbot for HCM systems, created by eVolpe.ai. Its mission is to simplify and accelerate work in the system by ending direct, natural interaction.
Key Takeaways
- The MintHCM AI Agent is the first free, open source AI chatbot for HCM systems – built on LangGraph, with a human-in-the-loop mechanism that requires user confirmation before any significant action.
- The agent is not tied to a single AI provider – it works with OpenAI, Anthropic Claude, Google Gemini and other LLMs, giving each organization the freedom to choose the model that fits their needs.
- The current version operates as an external application – it requires a separate backend and is not yet embedded directly in MintHCM. Additional setup is needed before it can be used in production.
- The roadmap moves toward agents defined as records inside MintHCM – triggered by record saves or automation buttons, with awareness of the user’s current location in the system.
- The project is open source – available on GitHub, open to community contribution and extension.
What is an AI Agent in MintHCM?
The MintHCM AI Agent is an intelligent assistant integrated with MintHCM, available as a chat interface embedded within the system. Users communicate with it in natural language and issue commands that are translated into specific actions in the system – making HR tasks significantly more intuitive. Instead of navigating forms to schedule a meeting or search for a candidate record, a user can type the request directly and the agent handles the execution, asking for confirmation before anything is saved.
How does the MintHCM AI Agent work?
The agent functions as an external server, communicating with MintHCM through the API and direct database queries. Access is provided via a chat widget embedded in the system’s interface.
Behind the scenes, it uses an architecture built with LangGraph, featuring dedicated components:
- gear_manager-node – initializes a new conversation and sets the prompt
- history_manager-node – manages conversation memory, summarizes and removes old messages
- llm_node – communicates with the chosen language model (e.g. OpenAI or Anthropic)
- tool_controller-node – verifies whether a selected tool is safe to use
- tool_node – executes the actions chosen by the LLM
The human-in-the-loop mechanism ensures that before performing any significant action, the agent asks for the user’s approval. This keeps the right balance between automation and control.
What can the AI Agent do today?
The agent supports users across a range of daily tasks:
- Checking schedules – retrieving information about upcoming meetings or tasks
- Employee and candidate lookup – quick access to system records by name or criteria
- Creating new entries – adding candidates, scheduling meetings, creating records
- Updating records – editing existing entries or adding participants to meetings
Does the AI Agent work with any language model?
Yes. One of the key design decisions in the MintHCM AI Agent is that it is not tied to a single language model provider. Its architecture allows integration with:
- OpenAI models (GPT series)
- Anthropic’s Claude
- Google Gemini
- Other providers supporting modern AI interfaces
Each organization can choose the model that best fits their requirements – whether based on cost, data privacy standards, or performance. This is consistent with the broader open source philosophy behind MintHCM – no vendor lock-in at the infrastructure level, and no lock-in at the AI model level either.
What are the current limitations of the AI Agent?
The honest answer is that the AI Agent is not yet production-ready, and we want to be clear about that.
The agent was built as a proof of concept – a working demonstration of what an open source AI chatbot for HCM systems can look like. The code is available on GitHub and can be installed, but it requires a separate backend, technical setup, and is not embedded in MintHCM as a native feature. Most organizations will find it is not ready to use in day-to-day HR work without significant additional development.
Known technical limitations include:
- Non-deterministic responses – similar prompts may trigger different tools, leading to inconsistent behavior
- Conversation history – managed through token-based summaries, which the project itself describes as a rough approximation rather than a reliable solution
- No page context – the agent does not know where in the system the user is located
- Data fabrication risk – when the agent lacks information needed to complete a request, it may fill in missing details rather than asking the user
We are sharing this because we think honesty about where a project stands is more useful than overstating its readiness. The architecture is sound, the direction is clear, and the roadmap below describes where we are taking this.
Where is the AI Agent heading? The roadmap for agents in MintHCM
The current AI Agent is the first step in a longer journey. The direction we are working toward looks significantly different from what exists today.
Agents defined as records in the system
The next iteration moves away from a single external chatbot toward a model where agents are defined directly inside MintHCM. An agent becomes a record: you write a prompt, select which tools it has access to, and define how it is triggered. No separate deployment, no external backend – the agent lives in the system alongside your HR data.
To illustrate what this looks like in practice, here are the kinds of interactions the system is being built to handle:
| Task | Example prompt | What the agent does |
|---|---|---|
| Schedule a meeting | “Schedule a meeting with the candidate tomorrow at 10 AM” | Creates a calendar entry, asks for confirmation |
| Find a candidate | “Show me active candidates for the PHP Developer position” | Searches the Candidatures module, returns a list |
| Update a record | “Change Anna Nowak’s position to Senior Specialist” | Updates the employee record after user approval |
| Check calendar | “What do I have today?” | Returns the day’s events from the calendar |
Trigger-based automation
Agents will be callable in two ways: attached to a record save event, or launched manually via an automation button in the interface. An agent triggered when a meeting record is saved could automatically update the candidate’s status, add a note, and prepare a summary – without the recruiter touching anything. One save, one outcome, no extra steps.
Page context awareness
One of the most meaningful improvements in development is giving the agent awareness of where the user is in the system. If you are viewing a candidate’s profile and type “schedule a follow-up meeting”, the agent will know which candidate you mean – without you needing to specify. The system context becomes part of the conversation.
A network of specialized agents
The longer-term vision is a system where agents can call other agents as tools. A recruitment coordinator agent could invoke a calendar agent, which in turn calls a notification agent – each specialized, each operating within its own permission boundary, composing into a workflow that handles the whole sequence autonomously.
We are building this step by step, in open source, with no fixed delivery dates we are not ready to commit to.
Why does it matter?
The MintHCM AI Agent demonstrates that AI in HR systems does not need to be limited to large, closed platforms. Thanks to its open source nature and flexible architecture, each organization can build a version tailored to their workflows – and contribute improvements back to the community.
The combination of open source, model flexibility, and human-in-the-loop control makes this a foundation worth building on – for organizations that want AI assistance in their HR processes without giving up control over how it works or which provider powers it.
If you want to follow the progress, contribute, or try the current version, the repository is available at github.com/eVolpe-AI/AI-HR-Agent.
Frequently asked questions
Is the MintHCM AI Agent free to use? Yes. The AI Agent is an open source project, available on GitHub at no cost. Organizations are responsible for hosting the backend and for any costs associated with the language model provider they choose to connect.
Which language models does the MintHCM AI Agent support? The agent supports OpenAI (GPT series), Anthropic’s Claude, Google Gemini, and other providers that support modern AI interfaces. The model is configurable per organization.
Is the AI Agent available in the standard MintHCM download? Not yet. The current version operates as a separate external application that connects to MintHCM via API. Embedding the agent directly into MintHCM as a native feature is part of the development roadmap.
What does human-in-the-loop mean in practice? Before the agent executes any action that modifies data – creating a record, updating a status, scheduling a meeting – it presents the proposed action to the user and waits for explicit confirmation. No changes are made without approval.
Can the agent access all data in MintHCM? The agent operates within the permissions of the authenticated user. It cannot access modules or records that the user themselves does not have access to in the system.
How is conversation history handled? The agent uses LangGraph’s history management, which summarizes older messages to keep sessions manageable. This is a functional but imperfect approach – very long sessions may lose some earlier context.
Can we add custom tools to the AI Agent? Yes. The modular architecture is designed to be extended. Developers can add custom tools that connect to additional MintHCM modules or external systems. Documentation and examples are available in the GitHub repository.
Sources
- Building the First AI-Enabled Open Source HCM. MintHCM, 2025.
- LangGraph documentation. LangChain, 2025.