Bring your own AI key means your project tool uses an AI provider account that you choose and pay for, instead of a model bundled into a subscription. In AlianHub, AI runs on your own LLM API key, a person's own AI app, or a model you host yourself. You pick the provider, you see the bill, and you decide where your data goes.
Key takeaways
- Your key, your model, your bill. No AI markup hidden in a seat price.
- AlianHub offers three ways to run AI: a workspace key, a personal AI account, or a self-hosted model.
- A self-hosted model means nothing leaves your network.
- Spend caps and daily budgets stop a busy agent from surprising you.
- No key? Nothing breaks. AI is optional in AlianHub.
What "bring your own AI key" means
Many tools now ship with AI built in. That sounds convenient until you ask three questions: which model is it, where does my data go, and what does it really cost?
With a bring-your-own model, the answers are yours. The tool talks to the provider using credentials you control. You can change provider later without changing tools.
Bundled AI is a decision someone else made for you. A key you bring is a decision you can change.
The three ways to run AI in AlianHub
The AI controls page describes three options, chosen by owners and admins. They are not exclusive. Many teams will use more than one.
| Option | Who pays | Where it runs | Best for |
|---|---|---|---|
| Workspace key | The company, capped per agent | Server-side, from AlianHub | Scheduled agents, such as a daily digest |
| Personal AI account | The person's own subscription, $0 workspace spend | On their machine | Hands-on work, on their own terms |
| Self-hosted model | Nobody per run; power and one GPU | Inside your network | Regulated teams, air-gapped installs |
Workspace key
This is one API key set in settings. AlianHub's admin guide lists the supported providers as OpenAI, Anthropic and DeepSeek, set with LLM_PROVIDER plus the matching key and model.
A workspace key powers server-side agents, including ones that run on a schedule. Everything they do stays on the allow-list, and spend is capped per agent.
Personal AI account
Here a person connects the AI they already use, such as Claude Code, to AlianHub. No key is held by the workspace. The AI works only while they are at the keyboard and can do only what they can do.
Offboarding is clean: revoke the token and the link dies. For the setup steps, see how to connect Claude and ChatGPT.
Self-hosted model
If you run Ollama or vLLM on your own GPU, AlianHub can use it. Nothing leaves your network. The trade-off, stated plainly on the controls page, is that it can be slower on hard reasoning.
For teams with strict data rules, that trade-off is often worth it.
Privacy: who sees your project data
Project data is sensitive. Task titles name clients. Comments mention prices. Docs hold plans that are not public yet.
When you bring your own key, the path is short and visible: your server, your chosen provider, and back. With a self-hosted model the path ends inside your own network.
A few other privacy details are worth knowing:
- AI memory is private. Each person can tell AlianHub how they like to work. Only they can see it, admins included.
- Memory imports are cleaned. You can bring memories from ChatGPT, Gemini or Claude, and passwords, keys and card numbers are removed first.
- Ask never widens access. It only reads projects you can already open.
Tip: decide where data may go before you choose a provider. If the answer is "nowhere outside our building", start with a self-hosted model and add a cloud key later for the jobs that need it.
Cost: seeing the real number
Bundled AI hides the cost inside a price per seat. A key you bring shows it to you directly, from your provider, at their rates.
AlianHub adds controls on top so cost stays predictable:
- A monthly spend cap per agent.
- A daily action limit per agent.
- A daily AI budget a workspace can set.
- Rate limits per agent and a one-click pause for all agents.
- An AI health and spend view with cost per agent, per model and per feature.
The cheapest AI is the one you can see. When the bill comes from your own provider account, there is nothing to reverse-engineer.
Pick the model per job
You can pin a model per agent or follow the workspace routing policy. In Ask, you can choose the model per question.
That means a heavy model for a hard planning question and a light, cheaper one for routine digests. You are not stuck paying top-tier prices for a two-line summary.
Freedom to switch
Providers change prices, release new models and retire old ones. If your AI is bundled, you wait for your vendor to react.
With your own key, switching is a settings change. The skills, the allowed actions and the limits live in AlianHub. Only the model is external, so swapping it does not touch how your agents are governed.
What happens without a key
Nothing breaks. The admin guide is clear: nothing in the product requires AI, and the AI pages simply stay empty without a key.
Projects, tasks, views, time tracking, chat and reports all work as normal. You can add AI when you are ready, and only for the teams that want it. Browse the rest of the product on the features page.
How to do this in AlianHub
- As the owner, open Settings › Instance › Settings and go to the AI group. Choose the provider (OpenAI, Anthropic or DeepSeek), then add the key and model. Secrets are stored encrypted.
- Open AI in the left menu. Ask, the agents and the AI project generator now have a model to use.
- For each agent, pin a model or leave it on the workspace routing policy, and set its monthly spend cap and daily action limit.
- For people who prefer their own AI, send them to AI, then Connect your AI. Claude Code users choose Create a token, pick the projects and permissions, and copy the token, which is shown once.
- Watch Usage & spend on the AI controls screens. One click pauses all agents if spend looks wrong.
A business scenario
A small consultancy wants AI-written weekly client updates but has strict rules on client data. It sets a workspace key for a scheduled Reporter agent with a modest monthly cap, and its two developers connect their own Claude Code accounts at $0 workspace spend.
The result: client reporting goes out on time every week, the AI line on the budget is a number the owner chose, and moving to a self-hosted model later is a settings change, not a migration.
FAQ
Which AI providers can I use with my own key?
The admin guide lists OpenAI, Anthropic and DeepSeek for the workspace key. You can also run a self-hosted model with Ollama or vLLM, or let people connect their own Claude, ChatGPT or Claude Code.
Does bringing my own key cost extra in AlianHub?
AlianHub is free and open source under AGPL-3.0, with no per-seat pricing. You pay your AI provider directly, at their rates, or nothing per run with a self-hosted model.
Can I limit how much an agent spends?
Yes. Each agent has a monthly spend cap and a daily action limit, and a workspace can set a daily AI budget. You can pause all agents in one click.
Do I need AI to use AlianHub?
No. Everything else works without a key, and the AI pages stay empty until you add one.
Try it on the live demo, or self-host AlianHub free and plug in the key you already have. If your team is debating bundled AI, send them this post.