NetBird Assistant early demo: looking for feedback

[TechHutTV]

Hey everyone,

Most of the team is focused on bugs and stability for 1.0, but we've also been building a new feature alongside that work: the NetBird Assistant. It's an AI assistant in the dashboard that reads your NetBird setup, helps you configure and troubleshoot it, and answers questions about how things work.

Demo video: https://youtu.be/JFGNbasscVw

In the video, the assistant:

• works out why one peer can't reach another by reading your peers, groups, and policies
• drafts policy, group, and DNS changes in the Control Center
• sets up Agent Network access to the Anthropic provider
• answers setup questions using the docs and Knowledge Hub

Also depending on the page picks up context from the page you're on. If you have a peer open, it knows which peer you're asking about.

The assistant is read-only. Any change it suggests shows up as a draft in the Control Center, and nothing is applied until you review and deploy it yourself. IP addresses, emails, and other sensitive data are redacted before anything goes to the model, and it can only see what the person using it has permission to see. For self-hosted, the plan is to ship it as a separate, optional container, so if you don't want it, you don't run it.

This is an early build and a lot of it can still change, which is why we're posting it now. Below are items we are still discussing.

• deployment model for self-hosted
• licensing for the assistant
• AI provider in cloud

Some things we'd like to know:

1. Would you use this, and for what?
2. What would you want it to handle that isn't in the video? A few we've talked about are auditing who changed a policy, checking who can reach a resource, and bulk changes across groups and policies.
3. Would you rather use NetBird from your own AI agents and tools, for example through an MCP server, instead of a built-in assistant? Or both?
4. What would make you comfortable, or uncomfortable, letting an AI work with your network config?

Comments are open below. "I wouldn't use this" is a useful answer too if you tell us why.

[barbobandsuch]

Personally this isn't something I want or would use. I'm tired of software getting bloated with assistants that add uneeded complexity. I switched to NetBird because I felt like it gave me more direct control over my system. I'd rather see more work put into other aspects of NetBird, especially the Android client which seems to be a battery hog. I can see where some people might find this useful, but to me, it seems like energy could be spent making improvements elsewhere.

I really appreciate that you've asked for input.

[shaban00]

Personally I wouldn't use this. This is one of the NetBird features that has gotten a lot of push back especially on Reddit.

I don’t think an AI assistant is necessary in NetBird. Most users already have access to tools like ChatGPT, Claude, or local LLMs. NetBird would benefit more from excellent documentation, a strong CLI/API, and an official MCP server that lets users connect their preferred AI tool.

[antybubbs]

Yeah, not sure I am with you guys on this one. I personally would not use it and I certainly dont think its a good idea to have some form of LLM within Netbird. I get that its read only.

My LLM knows netbird inside out (I fed it your docs) and so I just ask it for help if need be.

Normal users could just use CHatGPT etc if they are not confident/sure on what they want to do.

So thats my 2 cents, for what its worth. If it does go ahead PLEASE have a setting in the backend where we can switch it off.

okloveyoubye

[horzadome]

I would use it, but not if it's just a generic agent with rudimentary docs MCP. IMHO this would be awesome if:

• generic model (no need for frontier, any can do it) + Netbird's up to date elaborate MCP
• or a customized Netbird's model

I don't see much value in interpreting Netbird's server side view.
What would be awesome for me:

• agent's ability to decide if it needs to trigger debug bundle collection from client and automatically parse results (but only if needed)
• MCP directly into Netbird's KB, issue tracker, etc. Not just docs + tools for working with Netbird API. Also MCP into Netbird's support portal so that it notices if there's currently a well known issue (naturally, this would need to be either curated or passed throug anonimization models).

In any case - 👎 for a generic agent (even with rudimentary MCP), but massive 👍 if it's super-focused on Netbird and networking.

[NetHorror]

It would great if you implement in client too " the assistant works out why one peer can't reach another by reading your peers, groups, and policies"

[ressys1978]

Can it be disabled?

[keyboardhair]

Agents love CLI.

And (almost) everybody love Ansible.
https://github.com/netbirdio/ansible-netbird

The remaining people need NetBird Assistant. It can probably make onboarding easier. At one point they are onboarded, and their needs and requirements might change.

You should give all 3 groups what they need, and make sure all approaches are up to date on every release. Your agents will make this happen, including keeping docs up to date.

[bomtobe-hgkim]

I work on Orke at BOMTOBE; I haven’t tried this feature. Could policy drafts show evidence, the diff, and affected peers before deployment? That would make reachability changes easier to review.