Ships empty
On day one the platform has people, groups and an audit trail, and no agents. Nothing can act until an administrator installs something.
Runs on your own computer or in your own environment. Every agent is reviewed before it is installed, checked on every call, and recorded. It ships with no agents; you add the ones you need.
Available on request: write to ahmed@decentai.io. Made by DECENT INTELLIGENCE ARTIFICIAL SERVICES, Abu Dhabi.
The model proposes; the platform decides.
On day one the platform has people, groups and an audit trail, and no agents. Nothing can act until an administrator installs something.
Before it is installed, an agent declares each thing it can do, how far each one reaches, the records it keeps, the secrets it needs and the hosts it connects to.
An administrator reads the declaration and installs one exact version, checked byte for byte. Nothing updates itself: a new version is a new review.
Every call is checked against the agent's declaration, your organization's rules and the chat's trust. Above that trust, a person approves first.
Each agent runs in a sandbox of its own, with its own files and its own network, and reaches only the hosts it declared.
An agent is a program and a manifest, written to one published standard. Your ERP or internal portal can have its own agent, under the same controls.
Run the platform once; people use it from a browser, and soon from their desktop and phone.
The server side of DecentAI: people, agents, records and chats, the sandbox agents run in, and the audit trail. Runs with Docker, on one computer or on your servers.
The app people use in a browser: chats, approvals, agents, records and the organization's settings. Connects to any DecentAI server.
Thirty-eight agents for mail, calendars, files, documents and records you keep, and a real web browser. Open source, in two public catalogs.
For Windows and macOS: connect to a DecentAI server, or run DecentAI on the computer itself.
For iPhone and Android: chats, approvals and notifications from your DecentAI server, wherever you are.
Nothing about your people, your conversations or your data passes through us.
On your own computer, or on your own servers or cloud account. There is no hosted version: the platform runs where you run it. One installation can serve several organizations, each kept apart.
Choose from over two hundred providers, or a model you host yourself. You hold the keys. The model sees the conversation; it never sees a credential.
Conversations, documents, records, credentials and the audit trail stay in your deployment. Secrets are encrypted, and reach only an agent you listed, during a call.
DecentAI is proprietary, and we grant access case by case. You run it; we never see it. Updates are installed when you choose, and agents' updates through your own review.
What leaves your environment is listed and yours to limit: the conversation to the model you chose, an agent's calls to the hosts it declared, and notifications where you send them.
| Aspect | A general AI tool | DecentAI |
|---|---|---|
| Where it runs | The vendor's cloud | Your environment |
| What it can do on day one | Whatever it was connected to | Nothing, until you install an agent |
| How capability is added | A vendor connector, for everyone | An agent you review, install, allow and can uninstall |
| Your credentials | Held by the tool | Encrypted, released only to an agent you listed, during a call |
| Actions that change things | When the model decides | Checked against your rules; above the chat's trust, a person approves |
| What you can see afterwards | A chat history | An audit trail of every call, approval and refusal |
| Your own systems | A vendor roadmap item | An agent your developers write, in any language, to one standard |
A general tool asks you to trust the model. DecentAI asks you to approve a declaration and read a trail.
Write to ahmed@decentai.io with your organization and what you'd like to do. We answer within a few working days, with access or a demo.
On your own computer or your servers. One command sets DecentAI up with Docker and generates every password and key; then name your organization: you are its administrator.
Connect the AI model you choose, add an agent catalog, install each agent after reading what it declares, and ask for the work in plain words.
An agent is a program and a manifest, in any language. The standard is one public document, and the project's own agents, open source, show every part of it.