Open source AI tools for business give you something closed tools cannot: you can read how the AI is controlled, run it on your own servers and choose which model it uses. AlianHub is a working example. It is a free, AGPL-3.0, self-hosted project management and ERP workspace where AI agents do routine work, and the rules that limit them sit in code you can inspect.
Key takeaways
- With open source, "trust us" becomes "read it yourself".
- Self-hosting decides where your project data lives. A self-hosted model decides where your prompts go.
- AlianHub's agent limits come from an action registry, not a policy document.
- No per-seat pricing means you can give AI to the whole team, not a licensed few.
- Nothing in AlianHub requires AI. You add it when it earns a place.
Why open source matters more for AI than for most software
When a normal app has a bug, a button breaks. When an AI tool has a gap in its controls, it can change work nobody asked it to change.
So the questions you ask an AI tool are sharper. What exactly can it do? What can it never do? Where does my data go? With a closed tool you get answers in a sales deck. With an open source tool you can check them.
For AI, the source code is the safety policy.
Three kinds of control open source gives you
| Control | Closed AI tool | Open source, self-hosted |
|---|---|---|
| Code | Hidden | Readable, under AGPL-3.0 |
| Data location | The vendor's cloud | Your own server |
| Model choice | Whatever is bundled | Your key, or your own model |
| Pricing | Often per seat | AlianHub has no per-seat pricing |
| Exit | Export, if offered | You already hold the data |
Code you can read
AlianHub's agents can only take actions that exist in its action registry. Some actions are missing entirely: deleting a project or task, billing, deploying to production, merging code, removing a member, changing permissions and marking a task Done. Because the project is open source, that claim is something your team can verify, not just believe.
Data that stays with you
AlianHub runs on your own servers, using Docker Compose or Node.js and MongoDB. Your tasks, docs, comments and timesheets live in your database.
A model you choose
The workspace key can use OpenAI, Anthropic or DeepSeek. For stricter needs, AlianHub can use Ollama or vLLM on your own GPU, so nothing leaves your network. People can also connect their own Claude, ChatGPT or Claude Code over MCP. More on this in bring your own AI key.
The AlianHub AI features you get
Open source is the foundation. Here is what sits on it, all described on the AI page:
- Ask: plain-word questions across projects you can open, with Research, Plan changes and Skills.
- AI agents: members with an owner, a run budget, a skill, allowed actions and one autonomy level, L0 to L3.
- AI Inbox: where wider changes wait for a person to approve or decline.
- Audit log and Undo: every action, proposal and refusal is logged; changes made on their own can be undone.
- Skill library: versioned playbooks such as
digest.ceo,brief.parse,prd.draftandwiki.upkeep. - Team packs: role agents for 15 teams, starting paused.
- AI memory: private notes on how you like to work, up to 50 things.
- Budgets: a monthly spend cap and daily action limit per agent, plus a workspace daily AI budget.
Tip: some features are switched on by whoever runs the server, such as team packs, connecting your AI by address and skills that read outside your workspace. If the demo shows one as off, that is the reason.
How to do this in AlianHub
- Install AlianHub with the Docker Compose file and a
JWT_SECRETin.env, then rundocker compose up -d. The app opens atlocalhost:4000. Full steps are in self-host AlianHub with Docker. - Finish the setup wizard. On the last step, "One last step", you can connect your AI now or choose Skip for now.
- As the owner, open Settings › Instance › Settings, then the AI group, and set your provider, key and model. Or point AlianHub at your own model.
- Open AI in the left menu. Try Ask on your sample project.
- Switch on a team pack, review each agent in the Agents tab, and unpause one at L0 or L1.
- Check AI Inbox and Audit log daily for the first week.
A business scenario
A manufacturing business wants AI help with purchasing and production tracking, but its supplier prices and drawings cannot go to a third-party cloud. It self-hosts AlianHub, runs a model inside its own network, and switches on the relevant team packs for purchasing and production.
The result: agents draft routine updates and flag stuck orders, staff approve the wider changes, and the IT lead can show auditors both the code and the audit log. The business gets AI help across every team without paying per seat for it.
What open source does not solve for you
Honesty matters here. Open source does not make AI automatically safe. You still need to:
- Choose sensible autonomy levels for each agent.
- Set spend caps that match your budget.
- Keep your server updated and behind HTTPS.
- Decide who on the team reviews the AI Inbox.
Open source gives you the controls. Using them is still your job.
The AGPL-3.0 license also has a rule worth knowing: if you run a modified version as a public network service, you must publish your changes. Internal use and self-hosting are free.
How to evaluate any open source AI tool
Use these questions on AlianHub or anything else:
- Is the license clear, and does it allow self-hosting?
- Can you run it with your own model, or a model inside your network?
- Are agent limits enforced in code, and can you find where?
- Is there an audit log that includes refusals?
- Can you pause all AI in one click?
- Is there an active community for questions and bug reports?
For the last one, AlianHub has GitHub Discussions, docs and a contributing guide on the community page.
FAQ
What are the best open source AI tools for business?
Look for tools with a clear license, self-hosting, your choice of model and enforced limits on what AI can do. AlianHub covers project management and ERP work with AI agents under AGPL-3.0.
Is AlianHub really free?
Yes. It is free and open source under AGPL-3.0, with no per-seat pricing. Paid help with hosting, customization and support is available from Aliansoftware if you want it.
Can AlianHub's AI run without the internet?
With a self-hosted model such as Ollama or vLLM on your GPU, AI runs inside your network and nothing leaves. Connecting outside apps like Claude needs an address reachable from the internet.
Can I see the code that limits agents?
Yes. AlianHub is open source, so the action registry and the rest of the code are on GitHub for anyone to read.
Open the live demo, or self-host AlianHub free and read the code before you trust it. If your team is weighing open source AI, share this post in this week's AI conversations.