Governance
AI & Models
Add AI provider keys, choose which models the workspace may use, and set the image and embedding models.
Everything Next does, from drafting a PRD to answering a question about a knowledge base, runs on an AI model from an outside provider. AI & Models is where your organisation connects those providers with its own API keys and decides which of their models are allowed.
Open it from Governance → AI & Models.
Before you start
- AI & Models has its own permission, separate from the rest of Governance. If you cannot see this page, your organisation may not use it, or your role may not include it. See Understanding roles.
- Adding, replacing and removing keys, and choosing the embedding model, need the AI: Manage permission.
- Turning models on or off and choosing the image model need permission to change AI settings. Without it, those controls are switched off and show You don’t have permission to do this when you hover over them.
- Have your provider’s API key ready. You can create one in the provider’s own console.
Understanding the tabs
The page has one tab for each kind of AI work. Each tab has its own provider keys, so a key added on one tab is used only for that kind of work.
| Tab | What it covers |
|---|---|
| Text Inference | The models that write, summarise and reason for this workspace. |
| Image Inference | The models that generate and read images. |
| Embedding Inference | The model that turns documents into searchable embeddings, which is how knowledge bases are searched. |
Providers on each tab
| Provider | Text | Image | Embedding |
|---|---|---|---|
| OpenAI | Yes | Yes | Yes |
| Anthropic | Yes | No | No |
| Scaleway | Yes | Yes | Yes |
| Yes | Yes | Yes | |
| Mistral | Yes | Yes | Yes |
Azure, AWS, OpenRouter and Moonshot AI are listed at the bottom of each tab as Coming soon. They cannot be set up yet.
Reading a provider row
| What you see | What it means |
|---|---|
| Key set · updated time | A key is saved for this provider on this tab, and when it was last changed. |
| Not set | No key has been added on this tab. |
| Using the Text inference provider key until an embedding key is added | Embedding tab only. This provider has no embedding key of its own, so it embeds using its Text Inference key for now. |
| Managed by Paradym (or another note) | The key is managed for you and cannot be changed here. |
| Model catalogue · n of m enabled | The models the key unlocks, how many are switched on, and when the list was last fetched. |
Keys are write-only. Once a key is saved, its value is never shown again — only whether one is set.
Add a provider key
- Open the tab for the kind of work the key is for, such as Text Inference.
- Find the provider and select Add key.
- Paste the key into the field.
- Select Save. The button reads Checking key… while the key is checked with the provider. A key the provider refuses is never stored, and the reason is shown under the row.
Once saved, the row shows Key set and a Model catalogue with every model the key gives access to.
To cancel without saving, select Cancel or press Esc.
Replace or remove a key
- Replace a key by selecting Replace, pasting the new key and selecting Save. The new key is checked in the same way.
- Remove a key by selecting Remove, then Remove again to confirm.
Choose which models the workspace can use
Every model a key unlocks is switched on by default. Only enabled models are offered wherever someone chooses a model elsewhere in Paradym Next.
- Select the Model catalogue line under a provider to expand it.
- Switch off any model the workspace should not use. The change saves straight away.
If a provider row says Re-enter this key to load the models it gives access to, select Replace and enter the key again to fetch its model list.
Choose the image model
On Image Inference, the Image model setting at the top decides which model is used for images.
- Make sure at least one provider has a key and some image models switched on. If the list is empty, it reads Add a provider key below to choose.
- Open the Image model list and pick a model. The choice saves straight away.
Choose the embedding model
On Embedding Inference, the Embedding model setting decides how every knowledge base in the workspace is embedded. It is chosen once and then locked.
- Add a key for the provider whose embedding model you want, if you have not already.
- Pick the model in the Embedding model list.
- Select Use this model.
- Read the confirmation and select Lock in.
Once locked, the setting shows the model name and Locked.
If someone without the AI: Manage permission opens this tab before a model is chosen, they see Not chosen yet. Someone with the AI: Manage permission chooses it.
What happens next
- Models you enable here appear wherever a model is chosen in Paradym Next, such as on agent packs.
- New knowledge bases are embedded with the locked embedding model. Each knowledge base shows its embedding model in Governance → Knowledge Bases → Catalogue.
Good to know
- Keys on different tabs are separate. Adding an OpenAI key on Text Inference does not enable OpenAI image models; add a key on Image Inference as well.
- Switching tabs clears anything you were in the middle of, such as a half-typed key.
- Anthropic does not appear on Image Inference or Embedding Inference because it does not offer those kinds of model.