Gemini Kit
This kit contains components that provide access to Gemini API. Currently, there are two components: the text and the nano.
The text component
Use this component to use the larger Gemini models: Flash and Pro.
The component has no required inputs, but expects one of either Context or Text input port values to be supplied. Otherwise, it will throw an error.
The component's inputs and outputs mimic Gemini generateContent API, with a few important additions:
-
The component will automatically retry the API call if it fails, unless the response error code is
404,429or400(See Gemini API error codes for reference). The Retry Count port allows specifying how many times to retry the API call. -
The component has Text input and output ports for convenience and simpler use cases that don't involve multi-modal inputs or recurring conversation context.
Input Ports
-
Context (id:
context) -- the conversation context. Expects an array of Gemini API's Content objects. If not supplied, falls back to Text. -
Text (id:
text) -- text to use as prompt for generation. This is a less sophisticated version of Context, allowing you to pass a string, rather than having to construct the Content object array. If Context input port value is supplied, the Text value will be ignored. -
Tools (id:
tools) -- the array of tool call declarations, as defined in Gemini API. -
Model (id:
model) -- the model to use. Currently, the two options aregemini-1.5-flash-latestandgemini-1.5-pro-latest. -
Response MIME Type (id:
responseMimeType) -- the output response MIME type of the generated text. Supported values aretext/planandapplication/json. See Gemini API GenerationConfig for more information. -
Retry Count (id:
retry) -- how many times to retry the API call if it fails. -
Safety Settings (id:
safetySettings) -- the safety settings for the API call, an array ofSafetySettingobjects. See Gemini API SafetySetting for more information. -
Stop Sequences (id:
stopSequences) -- the array of strings that will stop the output. -
System Instruction (id:
systemInstruction) -- the system instruction for the API call. The port value can be either a string or a Content object. SeegenerateContentdocumentation for more information.
Output Ports
-
Context (id:
context) -- the outgoing conversation context. Will be a Gemini API's Content object, containing the latest generated response, including tool calls. -
Text (id:
text) -- the same value as Context, presented as a string. Convenient when not using tool calls.
Example
The board above is how one might typically use the text node. Here's another example, using Text input and output ports:
For a more elaborate use case, let's give it a few tools and let it make a decision to call them:
When we invoke this board with a prompt like "write me a poem?", we will see it respond with text, as usual. However, if we ask it questions like "what is the square root of e?" or "what is the distance between Earth and Moon?", it will issue a tool call:
The nano component
Allows calling the on-device Gemini Nano model via the experimental Prompt API. You have to sign up for early access here.
Input Ports
- Prompt (id:
prompt) -- the text-only prompt for the model
Output Ports
- Text (id:
text) -- the generated text output
Example
The board above provides a simple example of how to use the component in a board.