AI

What it is

Honeybase is bring-your-own-key: it does not resell model access, so an AI connection holds your provider credential and calls go out on your account and your bill.

An AI connection powers two things — AI nodes inside workflows, and Barry, the in-product assistant. An organization needs at least one before Barry is available at all.

Before you start

Have an account with the provider and a payment method on it. Most providers reject API calls from an account with no billing configured, which surfaces as an authentication-looking failure that no Honeybase setting will fix.

Setup fields

These are the fields the guided setup collects — both the Add integration form and the setup card Barry raises in chat. Additional settings live on the connection's settings page once it exists.

Connection settings

Field	Type	Required	Secret	Notes
name	Text	Yes	No	Display name for this connection. This is what you pick from in a workflow node.
provider	Choice	Yes	No	Which model provider to call — OpenAI, Anthropic, Google, Amazon Bedrock, Vertex AI and others. Your choice decides which credentials apply.
defaultModel	Choice	Yes	No	The model to use when a node does not name one. Nodes can override it.
apiBase	Text	No	No	Override the provider's API base URL. Set this for a self-hosted, proxied or Azure-style endpoint; leave it empty otherwise.

Credentials

This integration has 3 credential options. You supply the fields for the one you pick.

API key

For OpenAI, Anthropic, Google AI Studio, Groq, Mistral and the other providers that authenticate with a single key.

Field	Type	Required	Secret	Notes
apiKey	Password	Yes	Yes	The provider's API key. See the per-provider steps below.

AWS access key

For Amazon Bedrock, which authenticates with AWS credentials rather than a provider key.

Field	Type	Required	Secret	Notes
accessKeyId	Text	Yes	No	The AWS access key ID for the IAM user or role that may call Bedrock. An identifier, not a secret — it is stored as an ordinary setting.
secretAccessKey	Password	Yes	Yes	The AWS secret access key that pairs with the access key ID.
sessionToken	Password	No	Yes	The AWS session token, required only when the credentials are temporary (STS). Leave empty for a long-lived key pair.
region	Text	Yes	No	The AWS region to call Bedrock in. Model availability differs by region.

Google Vertex AI service account

For Google Vertex AI, which authenticates with a Google Cloud service account.

Field	Type	Required	Secret	Notes
serviceAccountJson	JSON	Yes	Yes	A Google Cloud service account key, pasted whole. Google shows this JSON once, at key creation.
project	Text	No	No	The Google Cloud project ID that Vertex AI runs in.
location	Text	No	No	The Vertex AI region, for example us-central1.

Getting a key

OpenAI — platform.openai.com/api-keys → Create new secret key. Shown once.

Anthropic — console.anthropic.com → API Keys → Create Key.

Google AI Studio — aistudio.google.com/apikey. This is the API-key path; Vertex AI is separate and uses a service account.

Other providers that issue a single key work the same way: create a key in their console and paste it in.

Amazon Bedrock

Bedrock uses AWS credentials rather than a provider key.

1. Request access to the models you intend to use, under Model access in the Bedrock console. Access is per-region.
2. Create an IAM user or role whose policy allows bedrock:InvokeModel (and bedrock:InvokeModelWithResponseStream for streaming) on those models.
3. Create an access key for it, and set the region to one where your models are enabled.

The access key ID and region are stored as ordinary settings rather than secrets — they are identifiers, and keeping them readable makes a misconfigured connection diagnosable. Only the secret access key and session token go to the secret store.

Google Vertex AI

1. Enable the Vertex AI API in your Google Cloud project.
2. Create a service account with the Vertex AI User role.
3. On its Keys tab, Add Key → Create new key → JSON, and paste the whole document into the setup form.
4. Set the project ID and the region you enabled the models in.

A note on what Honeybase can see

Barry never handles this credential. When Barry offers to set up a connection in chat, it raises a form that you fill in — the value travels from your browser to the secret store without passing through the model, and Barry is told only whether the connection was created, never what you typed.