Public chatbots are not right for sensitive work
Customer data, offers, and internal information need better control than copy-paste into an external tool.
Self-hosted Hermes AI agents
AI agents for real company workflows
OEJ builds AI agents that fit into your company’s real workflows. We start with one clear place, such as Gmail, Calendar, or Drive work through Telegram, and expand the agent when the value is visible.
- Start with familiar tools
- Custom agent, not a generic chatbot
- Important actions need human approval
Custom workflowThe solution is built around your company’s repeated work, not a ready-made SaaS template.
Limited accessThe agent receives access only to the channels and data needed for the workflow.
Local development partnerOEJ OÜ is a small company based in Tallinn. You speak directly with the developer.
Problem
Small businesses do not need AI hype. They need controlled help with real work.
A public chatbot can help write text, but it does not safely solve customer requests, document summaries, or internal follow-ups on its own. For that, AI needs to fit the workflow and access must be explicit.
Automation must reach the workflow
An agent should help with actual tasks: find information, prepare drafts, summarize material, and queue next actions.
Trust comes from architecture
Before building, we define where data moves, which permissions the agent has, and what it is not allowed to do.
Data sovereignty
OEJ’s core promise: the company stays in control.
The point is to avoid company knowledge and customer data moving uncontrolled through third-party services.
EU-hosted
Processing and storage are designed to stay inside the European Union where the chosen integrations allow it.
Minimal access
The agent receives only the permissions needed for the selected workflow.
Clear audit trail
You should be able to see what the agent did, when, and based on which input.
Example workflow
What does a practical agent look like?
Example customer-request agent: AI prepares the work, but the important decision stays with a person.
A request arrives
A customer writes through email, a form, or a website channel. The agent receives only the context needed for that workflow.
The agent prepares the work
The agent summarizes the request, finds relevant context, and prepares a reply or next action.
A person approves
A company owner or employee reviews the result, edits if needed, and sends or starts the action themselves.
Solution
What does OEJ build?
OEJ builds AI agents that connect to a company’s tools and agreed workflows. An agent can help sort emails, summarize documents, prepare reply drafts, queue follow-ups, or find internal information.
- Work in existing channels — the first step often fits Gmail, Calendar, Drive, and Telegram.
- Custom workflows — customer requests, quotes, internal summaries, follow-ups, or other repeated office work.
- Integrations — website forms, email, Telegram, Google Workspace, or the client’s own internal tools.
- Clear boundaries — access rights, data sources, and approval points are agreed before development.
- Natural growth — when the first workflow saves time, the same agent can expand to more work and integrations.
Pilot
Start with one useful workflow.
The first step does not need to be a big project. We choose one repeated job where an AI agent can show value quickly and where the boundaries are easy to agree.
Choose the first workflow
Often the easiest first step is a Google Workspace and Telegram workflow, because the company does not need to adopt a complicated new tool.
Agree permissions and approvals
We define what the agent may read, where it may create drafts, and which actions require human approval.
See whether it is worth expanding
If the first workflow genuinely saves time, the same solution can be expanded calmly from there. If not, that is also a useful answer.
Founder note
Built for local business owners.
OEJ OÜ is a one-person company based in Tallinn. The goal is to make self-hosted AI agents understandable and useful for local businesses.
If you have one repetitive process where assistance would save time or reduce data risk, it is a good pilot candidate.
Contact
Send workflow descriptionSend one repeated workflow.
A good start is a short description of one task that takes too much time or too often gets stuck. I’ll reply whether it could become a sensible first AI-agent workflow.