Migrate from Appsmith AI
Appsmith AI is deprecated and stops working on September 30, 2026. Creating new Appsmith AI datasources, or new queries on existing ones, is no longer possible. Existing queries continue to work until that date; files uploaded for context are not available afterwards.
This page shows how to move your existing Appsmith AI queries to the OpenAI, Anthropic, or Google AI datasources using your own API key.
Prerequisites
- An API key from the provider you want to use:
- A list of the queries in your app that use an Appsmith AI datasource. Each query uses one of the seven commands covered below.
With your own API key, the provider bills you directly for token usage. Appsmith does not proxy or pay for these requests.
Create the new datasource
- Click the (+) sign in the Datasources pane.
- Under AI Integrations, select OpenAI, Anthropic, or Google AI.
- Enter your API key and save the datasource.
Recreate each query
Every Appsmith AI command maps to a command on the new datasource. The Appsmith AI commands were purpose-built prompts over a large language model, so migrating a query mostly means moving your input into a prompt.
| Appsmith AI command | OpenAI | Anthropic | Google AI |
|---|---|---|---|
| Generate text | Chat | Chat | Generate Content |
| Summarise text | Chat | Chat | Generate Content |
| Classify text | Chat | Chat | Generate Content |
| Extract entities from text | Chat | Chat | Generate Content |
| Describe Image | Vision | Vision | Generate Content |
| Classify Image | Vision | Vision | Generate Content |
| Extract entities from image | Vision | Vision | Generate Content |
Generate text
Move your Prompt into the message of the new Chat or Generate Content query. If you used standing instructions (tone, role, output format), put them in a system message.
Summarise, classify, and extract entities from text
These commands wrapped your input in a task-specific prompt. Recreate the task as an explicit instruction in a Chat or Generate Content query:
- Summarise text:
Summarise the following text in about 5 sentences: {{Input1.text}} - Classify text:
Classify the following text into exactly one of these labels: Urgent, High priority, Low priority. Respond with only the label. Text: {{Input1.text}} - Extract entities from text:
Extract the following entities from the text: Name, Email, Address. Respond as JSON with one key per entity. If an entity is not found, use "Not found". Text: {{Input1.text}}
Anything you had in Additional Instructions can be appended to the prompt. For classification and extraction, instructing the model to respond with only the label or only JSON keeps widget bindings simple.
Image commands
Use the Vision command (OpenAI, Anthropic) or Generate Content with an image (Google AI). Pass the image as a URL or base64 string, the same formats Appsmith AI accepted, together with an instruction:
- Describe Image:
Describe this image. - Classify Image:
Classify this image into exactly one of: Jacket, Shirt, Pant, T-Shirt. Respond with only the label. - Extract entities from image:
Extract the following from this image: name, date of birth, license number. Respond as JSON.
Update widget bindings
The response structure differs between providers and from Appsmith AI. After recreating a query, run it once and re-bind your widgets to the new response structure. See the response details in the OpenAI, Anthropic, and Google AI references.
Migrate file context
Appsmith AI's Generate text command could answer questions using files uploaded to the datasource (file context). Uploaded files and their indexes are not migrated and are not available after September 30, 2026. There is no drop-in replacement; choose the option that fits your data size:
- Small documents: include the document text directly in the prompt of a Chat or Generate Content query.
- Larger document sets: build a retrieval flow. For example, generate embeddings with the OpenAI datasource's Embeddings command, store them in a vector-capable database such as PostgreSQL with pgvector, and pass the retrieved passages into your prompt.
- Provider-hosted retrieval: upload documents to your provider's own file or retrieval tooling and call it with the REST API datasource.
If you rely on file context, download your original files and migrate before the shutdown date while they are still accessible.
Clean up
Once all queries in an app are migrated and verified, delete the old Appsmith AI queries and datasource.