Skip to main content

​
Set up an account

Create an account here.

​
Create a project

Start by creating AgentView project: 
npm create agentview@latest my-agentview-project
This will install example project in my-agentview-project/ dir. Go to the project: 
cd my-agentview-project
It’s a simple Vite app with AgentView Studio and demo Chat Endpoint in agent.ts.

​
Set up environment variables

Before you start, you need to set up environment variables. Go to your org Dashboard to the API Keys tab and generate a development API key. Here’s what your .env file should look like:
  1. VITE_AGENTVIEW_ORGANIZATION_ID - your organization id, you can find it in Organization Details tab.
  2. AGENTVIEW_API_KEY - paste your development API key.
  3. OPENAI_API_KEY - your OpenAI API key (it’s needed since demo is using Responses API)

​
Run Chat Endpoint

Let’s run the chat endpoint: 
npm run dev:agent
It should be running at localhost:3000. It’s a simple Hono HTTP server computing AI responses using OpenAI Responses API and AgentView SDK.

​
Test

Let’s test the chat endpoint: 
curl -X POST --json '{ "input": { "type": "message", "role": "user", "content": "Hey, my name is Bob." } }' http://localhost:3000/simple_chat
You should get a response in a following JSON format: 
{
  "id": "SESSION_ID",
  "output": [ /* output items */],
  "token": "USER_TOKEN"
}
Let’s continue the thread by sending new message (insert SESSION_ID and USER_TOKEN from the previous response): 
curl -X POST --json '{
    "id": "SESSION_ID",
    "input": {"type": "message", "role": "user", "content": "Hello, what was my name again?"},
    "token": "USER_TOKEN"
  }' http://localhost:3000/simple_chat
You should get a response which shows that the agent remembers your name. AgentView correctly stores the conversation history and remembers the context of the conversation. Let’s send an incorrect input: 
curl -i -X POST --json '{ "input": { "type": "message", "role": "user", "content": 100 } }' http://localhost:3000/simple_chat
This triggers 422 error, as the input is not a string. AgentView handles the validation and provides a correct HTTP status code.

​
Run Studio

Let’s run the studio: 
npm install
npm run dev:studio
Studio should be running at localhost:1989. After you log in you you can:
  1. browse sessions, you should see the sessions we just created via curl
  2. create new playground sessions
  3. invite teammates
  4. comment on sessions, mention others, leave like/dislike default scores.