Setting up AI
Ithura's assistant runs on your own model key. You bring a key from a provider you already have an account with, and prompts go to that account.
That is a deliberate choice rather than a limitation. Your prompts go to a provider you chose, under terms you agreed, and you can see and cap the spend on their side as well as ours. Nothing is routed through a shared pool.
A workspace admin sets the key once, in Settings, then AI.
This key is for the in-app assistant. Agents that run at your provider use a separate Agent provider key, set on the same page, just below it. The two cards look alike and are not interchangeable: setting one does not set the other, and an agent will not start on the assistant's key. See Hosted agent sessions.
What you need
- A key from one of: Anthropic, OpenAI, xAI, Groq, or any OpenAI-compatible endpoint you host or buy.
- Workspace admin access, to set it.
- The Pro plan, which is what the assistant itself needs.
Setting the key
- Open Settings, then AI.
- Pick the provider.
- Optionally name a model. Each provider has a sensible default, so you can leave this empty unless you want a particular one. A custom endpoint is the exception: there is no default to fall back on, so name the model yourself.
- Paste the key and save.
The key is tested before it is stored. Ithura makes one small call to the provider, and if that call fails the key is not saved and you see what the provider said, so a typo or a key without credit is caught at the moment you paste it rather than the first time somebody asks a question.
After it is set
- The key is stored encrypted, and only its last four characters are ever shown again. Nobody, including a workspace admin, can read it back out of Ithura.
- Admins can test it again, replace it, or remove it.
- Members see whether AI is on, not the key or its hint.
- A custom endpoint must be https, and it cannot point back at Ithura itself.
What it costs
Prompts go to your provider account, so your provider bills you for them. There is no AI usage charge from Ithura on top.
AI spend reports what that work cost, broken down by task, agent, person and model. On this model it is a record of what you spent at your provider, not an invoice from us. It is worth watching for the same reason: it is the number that tells you whether an agent is being expensive.
Troubleshooting
| What you see | What it means |
|---|---|
| "The assistant is part of Pro." | The workspace is on Free or Starter. The plan is checked before the key, so this is about the plan, whatever the key is. |
| "AI is not set up for this workspace yet." | The plan is fine and no key is set. A workspace admin can set one in Settings, then AI. |
| The key is refused when you save it | The provider rejected the test call. The message is the provider's own, so a bad key, no credit, or a model you cannot use all read differently. |
| A custom endpoint is refused | It has to be https, and it cannot be an Ithura address. |
| It worked and then stopped | Test the key from Settings. A key revoked or out of credit at the provider fails there, not here. |
Related
- AI spend: what the work cost, and who spent it.
- Plans and limits: what the assistant needs.