A personal AI agent for your working day
I set up an agent on your own server or computer that works in familiar messaging apps, remembers agreed details and helps with recurring tasks. You choose the model, accounts and permissions.
- Your own server
- Remembers your work
- Asks permission for risky actions
Personal assistant in your channels
A personal AI agent is an assistant you can talk to in a messaging app you already use. I connect agreed channels such as Telegram, WhatsApp, Slack or email. Across different conversations, the agent can use persistent memory about your projects and preferences.
The agent needs a language model. You can choose a model provider or use a locally running model. Messages sent to a cloud model are processed by that model provider; with a local model, data processing depends on the chosen setup.
What the agent does for you
I configure recurring routines around your work. For example, the agent can prepare a morning overview, sort inbox messages into drafts, remind you of agreed items or search for information when you ask. Drafts remain for you to review.
The agent can use persistent memory and reusable skills. Some agent software lets the agent create skills and reuse them later. The exact tools and scheduled tasks available depend on the software and agreed permissions.
Setting up the agent
I choose open-source agent software suitable for the use, install it on your server, VPS or computer and connect the language model. One example is Hermes Agent, an open-source agent from Nous Research under the MIT licence. It is one example, not the only option, and I am not its author.
Then I connect only the messaging channels, accounts and tools you choose. I show you where to change instructions, routines and access. I build AI systems; I select agent software based on the task and your preferences.
Permissions and approvals
I limit the agent's access to accounts and tools you have approved. Sending a message, making a payment and deleting something require your approval first. I do not give the agent access to bank accounts or passwords.
Some agent software also supports approving commands and running them in an isolated container. I use these features when the chosen software and environment support them; they reduce risks but do not make the agent infallible.
Price and schedule
- First conversation · freeWe discuss how you will use it and suitable channels.
- Fixed priceI provide the price in writing before work begins.
- First version · 1–2 weeksI set up the agent, connections, permissions and agreed routines.
- Maintenance · if requestedMonthly maintenance scope depends on the work and is agreed after the first version.
Limits
The agent can give a wrong answer or make a mistake. Important answers and decisions must be checked. If you use a cloud model, its provider processes the messages sent to it; when using a local model, data flows depend on the chosen setup.
This is an assistant for one person or a small team, not a replacement for employees. I do not give the agent access to bank accounts or passwords. Approvals and limited permissions help keep actions under control but do not guarantee error-free operation.
Frequently asked questions
Where does a personal AI agent run?
I set up the agent on your own server, VPS or computer. I choose a suitable location with you, considering its use and the options available.
Which language model does the agent use?
You can choose a model provider or use a locally running model if it suits your needs and device. If you use a cloud model, you pay the model provider directly.
Which messaging apps can the agent connect to?
I connect the agreed channels, such as Telegram, WhatsApp, Slack or email. The choice depends on the agent software's capabilities and your accounts.
Can the agent send messages or delete files on its own?
I configure risky actions to require approval. Sending a message, making a payment or deleting something requires your permission first, and I grant access only to agreed accounts and tools.
How much does personal AI agent setup cost and how long does it take?
The first conversation is free. I provide the fixed price in writing before work begins. The first working version is ready in 1–2 weeks. If you wish, we can agree on monthly maintenance after the first version; its scope depends on the work.
Does the agent always give the right answer?
No. The agent can make mistakes, so its answers and important decisions should be checked. I do not give it access to bank accounts or passwords, and it is an assistant for one person or a small team, not a replacement for employees.
See also: AI development, AI agent for business, AI automation and AI consulting. Read about me.
Start planning your AI agent.
Tell me which routines you want to make easier and which accounts the agent may use.