How Devias is built for the agentic era

In an earlier post we argued that once AI agents join the team, a stack of separate tools is not enough, and a business needs an operating system: the whole business in one place, your own words, limits you control, data you own, and the freedom to add what is missing. Devias is built as that operating system, around those five. Here is how, in plain terms.

The whole business in one place

Devias is a core with modules on top, the way an operating system has programs on it. Customers, Quotes, Products, Invoicing, Payments, Inventory, Expenses and Email are modules. Workflows, tasks and documents are part of the core. They keep their records in the same place, so a customer, their quote, the invoice and the payment are linked records, not four tools. The agent works across all of it, with the same view you have.

Shaped to your business

The modules are starting points, not a boundary. You can use some of them, change them, or begin with none and model your own: the things you keep track of, the fields they have, and the views you want.

You can also tell the agent what you run, and it proposes the model for you: the objects, the fields, the views. You approve each change before it is made. What you add sits beside the modules and uses the same records, so nothing is a second-class citizen.

An agent that asks first

This is the part we spent the most care on, because an agent that can change your data needs limits.

• It acts as the person who asked. It uses the same actions as the app, with that person’s permissions, so it can never do more than they could.
• It asks before it changes anything. Reading and searching happen right away. Anything that changes your data waits for a yes: a card shows what the agent wants to do, and you approve or reject it. You can choose, tool by tool, what runs on its own and what asks first.
• Every change is on the record. The history shows what was changed, which agent changed it and for whom.
• When it is unsure, it asks. It puts a question to you instead of guessing.

Work that goes on without you

Routine work does not need you at all. A workflow starts when something happens, such as a record changing, a button, a schedule or an event from a connected service, and runs its steps. Every run is kept, step by step, so you can see what happened. You can also hand a task to an agent: it works on it and gives it back to you for review.

Connections to outside services build on this, and each of them uses your own account. We have written about three:

• Get paid online with Stripe: customers pay an invoice by card, and the payment records itself on the invoice.
• Send email from Devias with Resend: emails go out from your automations, and you hear back what happened to each one.
• Fill in your contacts automatically with Context.dev: people in your CRM are looked up and completed, without overwriting what you wrote.

Yours to run and to change

• It runs on your own server, in a database you control. Nothing leaves unless you turn on something that needs to, such as sending email, taking card payments or the AI provider.
• You choose the AI provider for the agents, Anthropic, Google or OpenAI, with one setting.
• It is open source. The whole platform is on GitHub under the AGPL-3.0, and the toolkit for building on it from outside is MIT. The software is free: you pay for your server and for what you connect, such as email or the AI provider.
• If something is missing, you can add it. A developer can write it as a module, install it like any other, and it sits beside the rest.

Where we are

Devias is young and still changing quickly. Today you work with the agent inside Devias. Letting other agents, such as Claude or ChatGPT, use Devias is planned, through MCP, along with channels and a hosted cloud version, which has no price yet.

Try it

The code is on GitHub and runs on your own server. If you would like a guided start, get a demo and we will help you set it up.