@zerodot/ai

Generate text, embeddings, images, and audio with OpenAI, Anthropic, Google Gemini, Mistral, Groq, xAI, DeepSeek, OpenRouter, Ollama, Azure OpenAI, Amazon Bedrock, Google Vertex AI, or any OpenAI-compatible service, in one node. Use it to add AI steps to a workflow.

Install

In zerodot, go to System > Nodes and install AI.

Manual: install @zerodot/ai where the web app and the worker run, and add it to ZERODOT_NODE_PACKAGES.

Nodes

AI (ai.api)

Choose a Provider. The provider decides which fields and operations the node shows. The Model list shows only the models of that provider. Choose Custom to type any model ID.

| Operation | OpenAI | Anthropic | Gemini | Compatible | | --------------------------------------------------------------------------- | ------ | ------------- | ------------- | ---------- | | Text: generate (system prompt, max tokens, JSON schema, image input, tools) | yes | yes | yes | yes | | Text: count tokens | - | yes | yes | - | | Embedding: create | yes | - | yes | yes | | Image: generate, edit | yes | - | yes | - | | Moderation, audio (transcribe, translate, speak) | yes | - | - | - | | File: upload, list, delete | yes | yes (and get) | yes (and get) | - | | Conversation and conversation item | yes | - | - | - | | Model: list | yes | yes | yes | yes |

The other providers have these operations:

| Provider | Text: generate | Embedding: create | Model: list | Model | | ---------------- | -------------- | ----------------- | ----------- | ------------------------ | | Mistral AI | yes | yes | yes | list, or Custom | | Groq | yes | - | yes | list, or Custom | | xAI | yes | - | yes | list, or Custom | | DeepSeek | yes | - | yes | list, or Custom | | OpenRouter | yes | - | yes | type the model ID | | Ollama | yes | yes | yes | type the model name | | Azure OpenAI | yes | yes | - | type the deployment name | | Amazon Bedrock | yes | yes | - | list, or Custom | | Google Vertex AI | yes | yes | - | list, or Custom |

Provider notes:

• OpenAI: models gpt-6-astra, gpt-6.1-sol, gpt-6.1-terra, gpt-6-luna, gpt-5.5, gpt-5.4-mini. Embeddings, image, moderation, audio, and file models have their own lists.
• Anthropic: models claude-fable-5-1, claude-opus-5-5, claude-sonnet-5-5, claude-haiku-5-5. The node can read PDF files and uploaded file IDs, set the Effort, and continue after a tool call with Previous Answer Content and Tool Results. The node does not send temperature to Anthropic. Some models, such as claude-sonnet-5-5, do not take the tool choices Any Tool and One Tool. For these models the node sends Auto, and with One Tool it sends only the chosen tool.
• Gemini: models gemini-4-argon, gemini-3.8-flash, gemini-3.6-flash, gemini-3.5-flash-lite, gemini-3.1-flash-lite, gemini-3.1-pro-preview. Embeddings gemini-embedding-2 and gemini-embedding-001. Image models gemini-nano-banana-2.1, gemini-3.1-flash-image, gemini-3.1-flash-lite-image, gemini-3-pro-image. Gemini reads images from data URLs, from the binary field, and from File URIs. Use File Upload to get a uri, and put it in File URIs to send a PDF or other media. The node reads the type of each file from Gemini before it sends the prompt.
• OpenAI-compatible: type the model ID in Custom Model ID. The node uses /chat/completions, /embeddings, and /models.

Output of Text: generate, the same for all providers: text, json (when the format is JSON), toolCalls (id, name, input), finishReason (stop, length, tool_calls, content_filter, or other), stopReason (the value of the provider), usage (inputTokens, outputTokens, totalTokens), model, and id. Turn on Include Raw Answer to add the full answer as raw. Anthropic adds assistantContent, to continue after a tool call.

Tools are a JSON array. Each tool has name, description, and input_schema (a JSON Schema with type object). The node changes the format for each provider. A tool choice other than Auto needs tools.

Text: generate and Embedding: create use the AI SDK. The other operations call the API of the provider. The node does not download the URLs of the prompt. The provider reads them.

A list can return all items (at most 5000) or a limit. Lists of URLs and IDs have one value on each line. When a provider gives an error, the node fails with the message, the error code, and a short help text. A rate limit (HTTP 429) gives the time to wait from Retry-After.

Credentials

The node uses the credential that matches the Provider.

• OpenAI API (ai.openai): apiKey, and optional organization, project, and baseUrl.
• Anthropic API key (ai.anthropic): apiKey, and optional baseUrl.
• Google Gemini API key (ai.gemini): apiKey from Google AI Studio, and optional baseUrl.
• OpenAI-compatible service (ai.compatible): baseUrl up to the version, and an optional apiKey.
• Mistral AI, Groq, xAI, DeepSeek, and OpenRouter (ai.mistral, ai.groq, ai.xai, ai.deepseek, ai.openrouter): apiKey, and an optional baseUrl for a proxy.
• Ollama (ai.ollama): an optional baseUrl (default http://localhost:11434/v1) and an optional apiKey.
• Azure OpenAI (ai.azure): apiKey, and resourceName or the full baseUrl up to /openai. apiVersion is optional.
• Amazon Bedrock (ai.bedrock): region, and accessKeyId with secretAccessKey (and sessionToken for temporary keys), or a Bedrock apiKey.
• Google Vertex AI (ai.vertex): project, an optional location (default us-central1), and serviceAccountJson (the JSON key of a service account) or an express mode apiKey.

A baseUrl must start with https://. Plain http:// works only for localhost, 127.0.0.1, and [::1]. The keys are secret. The Test button reads one model. Azure, Bedrock, and Vertex have no Test button.

License

AGPL-3.0-only