Skip to content
Back to overview
|

We Set Up Weldall CLI on Your Servers for Free

A central access layer between your employees' AI agents and your company APIs, and why we install it for you at no cost
Martin Seibert
Co-Founder & CEO, Seibert Group
Weldall CLI connects AI agents to company APIs

Weldall is a central access layer between your employees’ AI agents and your company systems. In one place you decide which agent may reach which API. The code is on GitHub, and you can install and configure it yourself. Or we do it for you: we set up Weldall on your servers, configure it to your needs and charge you nothing for it.

This article is also available in German: Wir richten Weldall CLI kostenlos auf deinen Servern ein.

Why a Software Company Gives Away the Setup

That sounds like a strange offer, so here is the reason. We want to learn how companies actually run Weldall: which instances they operate, which identity provider they use, which APIs, which agents. We learn the most when we install it ourselves. In return for the free setup, we see how you run it.

The setup costs you nothing. How long we offer it, we decide after the first setups.

What Weldall Does

An AI agent is a chat assistant that can also act. It runs programs on your computer, opens tickets, reads the wiki or calls an API. More and more employees use one. To reach company systems, the agent needs access. The usual answer is an access key per person and per tool, pasted into some configuration file. Afterwards nobody knows who can reach what.

Weldall sits between the agents on one side and your company systems on the other. It has two parts. The first runs on a server you host: the authorization server with the admin interface, where you write the rules. The second is a small command-line program, the CLI, on each employee’s computer. It signs the person in and sends the requests.

In the admin interface you define in one place:

  1. What agents may do. Each capability is a written instruction for the agent, a skill. The skills are published in one central list, the catalog.
  2. Who gets it. Per person, per group or per machine.
  3. Where it applies. Each connected system, a resource, defines its own permissions. One permission that a request asks for is a scope, for example “read expenses” but not “change expenses”.

The agent asks the CLI what it may do and gets the skills meant for this person. You change the catalog once, and every agent in the company sees the change.

The Agent Never Holds a Token

This is the part I care about most. A token is the digital key an app uses to call an API. With Weldall the agent never sees one. The CLI keeps the credentials in the operating system’s secure store and sends the requests itself. The tokens it uses are short-lived and bound to the device: a token copied from one machine does not work on another.

Every granted and every denied request lands in the audit log. If something goes wrong, you can see which agent tried what, and when.

People sign in with the single sign-on they already use, your identity provider, the system that holds the accounts. Machines get their own registered identities. Groups can come from LDAP or Active Directory.

What You Can Connect

  • Built-in connections for Google Workspace, Jira and Confluence. The tokens for these services stay encrypted on the Weldall server.
  • Your own internal apps. There are code libraries for TypeScript and Python, so your developers don’t build their own sign-in check for every internal app.
  • The agents your people already use. Weldall works with any agent that can run a command-line program. The README recommends open-source agents such as OpenCode, Codex CLI, Gemini CLI and Goose; for Goose there is a ready-made extension.

What It Looks Like Day to Day

At Seibert, Weldall CLI is installed on every Mac. Each employee runs weldall login once. After that the agent runs the commands, not the person. They look like this:

Terminal window
weldall status # who am I, what can I do?
weldall skills find expense
weldall request --scope expenses:read https://expenses.example.com/api/expenses

The last line is the important one. The agent asks for exactly one permission, here “read expenses”, and the CLI checks the target before it sends anything.

If you want to see what that looks like in a chat, I showed the Weldall Client with contracts, sales pipeline and ticket risks in my article about MCP Apps.

Loading content...

The License

The code is public on GitHub under FSL-1.1-ALv2, the Functional Source License. You may use, modify and distribute it for any purpose except offering it as a competing commercial service. Each released version turns into Apache-2.0 on its second anniversary.

How to Get the Free Setup

Write to agentsinteams@seibert.group and tell me what your setup looks like: your identity provider, the systems your agents should reach, the agents your people use. Unusual setups are welcome; those are the ones we learn the most from.

If you want to do it yourself, start here:

If you want to try an AI chat assistant on EU servers: SeibertGPT is a chat assistant that runs DeepSeek on our own server in Helsinki. It is not the backend of the setup I describe here.

Atlassian news, monthly

Our newsletter brings you the month's Atlassian updates. Unsubscribe anytime.

Submitting sends you a confirmation email. Only the link in it activates the subscription, and you can end it again at any time.