Mastering AI Resource Queries in Retool: Generate Text, Analyze Images, and Process Documents

AI resource queries allow Retool users to tap into leading AI models from providers such as OpenAI, Anthropic, Google, and Amazon Bedrock directly inside their apps and workflows. With these queries, teams can generate text, analyze images, build chat experiences, and process documents without leaving the Retool environment. This section introduces the value of AI resource queries for building intelligent internal tools, from content generation to automated data extraction. Note that this documentation applies to classic apps, which use Retool's original app editor, while the new app builder is the recommended path for new development.

Before writing queries, an AI provider resource such as OpenAI or Anthropic must be created in your Retool organization. In apps, open the Code panel, click the plus icon to add a new query, select your AI resource, and choose an action type. In workflows, add a Resource query block, select the AI resource, and choose an action type. Once configured, queries run against the chosen provider and model, giving you a consistent foundation for every AI-powered feature you build.

Text actions use language models to process and produce written content. The Generate text action creates original content for marketing copy, creative writing, code generation, and general tasks; a product description example shows how a system message and dynamic component values combine to produce tailored copy. The Summarize text action condenses long content into concise summaries, useful for analyzing customer feedback. The Classify text action assigns text to predefined categories, which suits support ticket routing and sentiment analysis. The Extract entity from text action pulls structured information such as names, dates, and amounts from unstructured input, as seen in invoice data extraction.

The Generate chat response action produces contextual replies across multi-turn conversations, maintaining flow by considering previous messages. Retool's LLM Chat component simplifies this further: add the component to the canvas, optionally define a system message to shape the assistant's persona, and the component manages message history automatically. The full conversation is accessible through the component's message history property. In workflows, developers provide the conversation history explicitly through a resource query block. Customer support chatbots are a typical use case.

Vision actions bring multimodal analysis to Retool. Generate text from image lets users ask questions about an image supplied as base64, a URL, or a File Input value. Caption image produces descriptive captions in a chosen style, useful for social media or accessibility. Classify image sorts images into predefined categories such as product types, and Extract entity from image pulls out text via OCR, objects, or structured data such as receipt fields. These actions require vision-capable models, and support varies across providers.

The Generate image action creates images from text descriptions using DALL-E models, returning base64-encoded output. Developers can combine a Textarea for the prompt and a Select component for style, reference both in the query, and display the result in an Image component by setting its source to the query data. Dimensions, style, and quality options depend on the model in use. This capability supports use cases such as generating marketing graphics on demand within internal tools. Note that image generation is available through OpenAI and Retool-managed models, not Amazon Bedrock.

The Convert document to text action extracts text from PDF files. A File Input component supplies the document, and the parsed output becomes available for downstream use. Importantly, parsing happens inside Retool and no files are sent to AI models, which helps with data handling requirements. A common workflow is to extract text first and then pass it to a Summarize or Generate text query, enabling document analysis pipelines built entirely from chained queries.

AI resource queries draw from an organization's monthly AI credits. Organization admins can review consumption in Settings under Plans and Billing, using the AI Credits tab and its Usage view to see credits broken down by product. Enterprise organizations also get an AI visibility settings page with granular insight into Retool-managed and bring-your-own-key usage, filterable by model, user, and feature. Monitoring these figures helps teams control costs as AI features expand across their apps.

Several practices improve reliability and quality. Use system messages to set consistent behavior and context for the model. Reference component values with double-brace syntax so inputs stay dynamic. Chain queries to combine actions, such as extracting text, summarizing it, and then classifying the result. Remember that AI responses stream in real time when bound to components. Validate inputs before sending them to models to avoid avoidable errors, and test several models to balance output quality, response speed, and cost for each use case.

To go further, explore the AI provider resources documentation to connect OpenAI, Anthropic, Google, and other providers; the LLM Chat component guide for building interactive chat interfaces; Retool Vectors for grounding AI responses in your own knowledge base; and Agents for building autonomous AI systems that can use tools. Together these resources let teams move from simple AI-generated text to sophisticated, context-aware internal applications.