The assistant's credential
The assistant runs on an Anthropic API key shared by the whole organization, or on a cloud provider. It is set up in Configuration → Assistant, Credential tab.
Set it up
- Create an API key at console.anthropic.com.
- In the Credential tab, paste it into Anthropic API key.
- Click Test and save. NappAI checks the key before saving it (“Testing…”).
Once saved, it is only shown masked (sk-ant-•••••••• and a fingerprint to recognize it). To replace it, click Change. The key is stored encrypted.
If the check fails, the key isn’t saved and you’ll see the reason:
| Reason | What to do |
|---|---|
| The API key isn’t valid or has been revoked in Anthropic. | Check that you copied all of it, or create another one. |
| The Anthropic account for this key has no credit. | Top up your credit in Anthropic and try again. |
| The key couldn’t be checked right now. | It’s a temporary problem: try again in a few minutes. |
The model provider
If your installation allows it, under Model provider you can choose where the assistant’s models come from instead of an Anthropic API key: for example, to use your company’s cloud account.
| Provider | What it asks for |
|---|---|
| Anthropic (API key) | The API key, as above. |
| Amazon Bedrock | The AWS Region with Bedrock (e.g. eu-west-1) and the Authentication: a Bedrock API key or IAM keys (Access key ID and Secret access key). Optionally, the Endpoint URL. |
| Google Vertex AI | The Google Cloud Region (e.g. europe-west1, us-east5 or global), the Project ID (empty = the service account’s project) and the Service account (JSON): paste the whole .json file. It needs permission to invoke Vertex AI models. |
| Microsoft Foundry | The Foundry resource (what goes before .services.ai.azure.com) or its URL, and the Resource API key. |
| Compatible gateway (experimental) | An Anthropic-compatible gateway (DeepSeek, LiteLLM…): a Tested gateway or Other, with its Base URL, the Token and the Token header. It only appears if your organization allows it in Policies. |
With a cloud provider you also fill in Models: each model’s id in your provider (in Foundry, the name you gave each deployment; in the gateway, the model that plays each one). Whatever you leave empty isn’t offered in the model selector.
Each provider warns about what changes with it, and the panel’s selector shows those same warnings to each person:
- On Bedrock the assistant has no web search, and prompt caching may be missing in some AWS regions: long turns may cost more.
- On Vertex AI, each region has its own models: check that yours has the ones you set.
- The gateway is experimental and not supported by Anthropic: tools, reasoning, caching and compaction may behave differently or fail. Non-Claude models are not validated for building flows: use them for testing, not in production.
Saved keys and tokens are not shown again: leave them empty to keep the current ones. Click Test and save, as with the API key.
The status
Below you’ll see the credential’s Status, the date of the last update and, if there is one, the last error:
| Status | What it means | How to fix it |
|---|---|---|
| Valid | The assistant works. | — |
| Invalid | Anthropic rejects the key. | Create a new key at console.anthropic.com and paste it here. |
| No credit | The Anthropic account has run out of credit. | Top up at console.anthropic.com → Billing; the assistant works again as soon as there is credit. |
| Not configured | There is no key. | To start using the assistant, add an Anthropic API key. |
While the credential isn’t working, the assistant panel explains it to each person before they type: that the assistant isn’t enabled in their organization yet and they should ask their administrator to set it up in Configuration → Assistant, or that the key is no longer valid or has run out of credit.
Delete it
Delete credential deletes it, after you confirm. The assistant stops being available for the whole organization until another one is set up.