Self-Hosted AI Chat for Teams: Per-Group Control Without the SaaS Tax

If you've ever tried to roll out an AI chat tool across a company, you know the pain. One team wants a 200k context model, another just needs a fast thing for ticket summaries, and finance is asking why you're burning money on a per-seat license for an API you could call yourself. Most self-hosted solutions give you a chat window and nothing else. This project actually thinks about permissions and budgets.

HiveChat solves that specific headache. It's a self-hosted AI chat interface designed for teams, but the key differentiator is granular control. You can set which models each group can access and, more importantly, cap their token usage. That's the kind of feature that lets you run this thing in production without a panic attack on your monthly bill.

What It Does

At its core, HiveChat is a web-based chat application. You connect it to your AI backend (think OpenAI-compatible APIs, local models, whatever you have), and it gives your team a clean, familiar chat interface.

The real functionality is in the group management. You create groups, assign users, and then define policies per group. For a specific group, you can say "these users can only use the gpt-4o-mini model" or "this group gets access to your internal fine-tuned model, but they have a hard cap of 500k tokens per day." It's about being able to say "yes" to AI adoption without letting costs spiral out of control.

It handles the boring but essential stuff too: user authentication, conversation history, and admin panels. It's built to be a drop-in internal tool, not a toy.

Why It's Cool

The token limiting per group is the standout feature. Most self-hosted chat apps treat token limits like an afterthought. HiveChat makes it a core permission. This is genuinely useful.

Here are a few things that stood out to me looking through the repo:

Model routing by policy. You don't have to give everyone access to your expensive reasoning models. You can lock the fancy stuff to a handful of senior engineers and give the customer support team a cheap, fast model. That's smart architecture.

Cost control is enforced, not suggested. The token caps aren't a soft suggestion in the UI. They're enforced at the API layer. If a group hits their limit, the API stops answering until the window resets. That's the difference between a dashboard and a system that works.

Built for a real organization. It has user roles, group invites, and an admin interface. It feels like it was built by people who actually had to manage an AI budget, not just someone experimenting with a chatbot.

Self-hosted means your data stays put. If you have strict compliance requirements, this is huge. You control the model backend and the data flow.

How to Try It

Getting it running is straightforward, assuming you have Docker and an AI API key handy. Check the repo's README for the latest specifics, but the general flow is:

1. Clone the repository.
2. Copy the .env.example to .env and add your AI provider keys (OpenAI, Anthropic, local endpoints, etc.).
3. Run docker-compose up -d (or use the provided setup script).
4. Access the admin panel, create groups, and assign model permissions.

There isn't a hosted demo that I found, which makes sense for this kind of tool. It's meant to live on your infrastructure. The setup is well-documented, so you should be up and running in under ten minutes if you're comfortable with basic Docker.

Final Thoughts

This is a tool for developers and IT leads who are tired of battling shadow IT or worrying about the bill at the end of the month. It's not flashy, but it's practical. The per-group limiting is the killer feature that most self-hosted options just don't have.

If you're already running a homelab or a small VPS and you want to standardize AI chat for a small company, this is worth a weekend spin. You get the control of self-hosting with the guardrails you actually need to keep the project alive. For us, that hits the sweet spot between "cool experiment" and "production-ready internal tool."

Found this interesting? Follow @githubprojects for more useful repos and tools.