Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.vocode.dev/llms.txt

Use this file to discover all available pages before exploring further.

When a call comes in, some applications may want to inject context (eg the caller’s first name, if the number is recognized). In order to support injecting context, the Vocode API allows prompts to be templated. When an inbound call comes in, the template variables can then be filled in at runtime.

​
How it works

At a high level, context injection works like this:
  • set your Vocode prompt to include template values
  • include a context_endpoint parameter to your bot
  • when an inbound call comes in, Vocode will query the context endpoint provided with the caller phone number
  • the queried information from the endpoint is then filled into the prompt template

​
Context injection example

In order to have your bot use injected variables, the prompt must include template variables. Let’s set up our prompt to accept a user’s first name and link it to our context endpoint. 
from vocode import AgentUpdateParams, PromptUpdateParams

name_prompt = "You are talking to ${name}"
number = vocode_client.numbers.update_number(
    phone_number="YOUR_NUMBER",
    inbound_agent=AgentUpdateParams(
        prompt=PromptUpdateParams(
            content=name_prompt,
            context_endpoint="YOUR_CONTEXT_URL",
        )
    ),
)
Now, let’s set up a simple context endpoint that will return the user’s first name: 
from fastapi import FastAPI

app = FastAPI()

@app.get("/my-context-endpoint")
def endpoint(phone_number: str):
    if phone_number == "123":
        return {"name": "Kian"}
    elif phone_number == "456":
        return {"name": "Ajay"}
    else:
        return {}
This sets up a route at http://localhost:8000/my-context-endpoint that accepts phone_number as a URL param, i.e. http://localhost:8000/my-context-endpoint?phone_number=15555555555. Now, when either Kian or Ajay call the bot – the prompt will reflect the name of the person the bot is talking to! Note: This feature is experimental… improvements and proper authentication coming.