You can generate embeddings for text using the intent-embed model. Intent Embed is a mdoel that generates embeddings for text, specifically to represent the user intent. Potential use cases include:

• User Intent classification
• Intent similarity
• Out of topic exclusion
• Intent clustering and analytics
• And more

Read the technical paper here: Phospho Intent Embeddings.

​

Requirements

Create an account on phospho.ai and get your API key. You need to have setup a billing method. You can add a it in the Settings of your dashboard here.

​

Usage

​

Using the OpenAI client

The phospho embedding endpoint is OpenAI compatible. You can use the OpenAI client to send requests to the phospho API.

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_PHOSPHO_API_KEY",
    base_url="https://api.phospho.ai/v2",
)

response = client.embeddings.create(
    model="intent-embed",
    input="I want to use the phospho intent embeddings api",
    encoding_format="float",
)

print(response)

For now, the input must be a single string. Passing more than one string will result in an error.

​

Using the API directly

To send a request, add:

• text: The text to embed, usually a user query or message.
• model: must be set to intent-embed.

Optionally, to link this embedding to one of your projects, you can specify the following optional parameters:

• project_id: The project id you want to link this embedding to.

You will get a response with the embeddings for the input text. The embeddings are a list of floats.

{
  "object": "list",
  "data": [
    {
      "object": "embedding",
      "embedding": [
        -0.045429688,
        -0.039863896,
        0.0077658836,
        ...],
      "index": 0
    }
  ],
  "model": "intent-embed",
  "usage": {
    "prompt_tokens": 3,
    "total_tokens": 3
  }
}

These embeddings can stored in vector databases like Pinecone, Milvus, Chroma, Qdrand, etc. for similarity search, clustering, and other analytics.

​

Pricing

The pricing is based on the number of tokens in the input text.

Note: You need to have a billing method setup to use the model. Acces your billing portal to add one.

Contact us for high volume pricing.