Skip to main content
The Gemini endpoint generates contextual responses using Google’s Gemini 2.0 Flash model, incorporating both conversation transcripts and emotional analysis data.

​
Endpoint

POST http://localhost:5000/api/gemini

​
Request body

​
transcript
string
required
The conversation transcript text to analyze and respond to
​
emoData
object
Object containing emotion names as keys and confidence scores (0-1) as values. The top 3 emotions by score are used in the prompt.Example:
{
  "joy": 0.85,
  "contentment": 0.72,
  "interest": 0.68,
  "surprise": 0.45
}

​
Response

​
response
string
The AI-generated response text from Gemini
​
emotions
object
The emotion data that was provided in the request, echoed back in the response

​
Example request

curl -X POST http://localhost:5000/api/gemini \
  -H "Content-Type: application/json" \
  -d '{
    "transcript": "I had a really great day today. Everything went well at work.",
    "emoData": {
      "joy": 0.85,
      "contentment": 0.72,
      "interest": 0.68
    }
  }'

​
Example response

{
  "response": "That's wonderful to hear! It sounds like you had a very productive and fulfilling day.",
  "emotions": {
    "joy": 0.85,
    "contentment": 0.72,
    "interest": 0.68
  }
}

​
Error responses

​
error
string
Description of the error that occurred

​
Missing transcript (400)

{
  "error": "Transcript is required"
}

​
Server configuration error (500)

{
  "error": "Unable to process your request at this time"
}

​
Model configuration

The endpoint uses the following Gemini API parameters:
​
temperature
number
default:"0.3"
Controls randomness in the response (0.0-1.0). Lower values make output more deterministic.
​
topK
number
default:"20"
Limits the number of tokens considered at each step
​
topP
number
default:"0.8"
Nucleus sampling parameter for controlling diversity
​
maxOutputTokens
number
default:"100"
Maximum length of the generated response

​
Emotion processing

When emoData is provided:
  1. Emotions are sorted by score (highest to lowest)
  2. The top 3 emotions are selected
  3. Scores are converted to percentages with one decimal place
  4. Formatted as emotion: score% and included in the prompt
Example formatted emotion data: 
joy: 85.0%
contentment: 72.0%
interest: 68.0%

​
Implementation details

The endpoint:
  1. Validates that transcript is provided
  2. Processes emotion data (if provided) to extract top 3 emotions
  3. Formats a prompt using the GEMINI_PROMPT environment variable template
  4. Sends the request to Gemini 2.0 Flash model
  5. Extracts and returns the generated text response
The endpoint requires GEMINI_API_KEY and GEMINI_PROMPT environment variables to be configured.
If emoData is not provided, the endpoint will still process the transcript with an empty emotion context.

Build docs developers (and LLMs) love

Get started for freeTalk to us