Install
Create an API key. Install the SDK for your application.
Terminal, TypeScript
npm install @alethi/pantheon-chat react@18 react-dom@18Terminal, Python
pip install pantheon-chatReact
Render the component with an agent slug and application tool handlers.
App.tsx
import { PantheonChat } from "@alethi/pantheon-chat/react";
export default function App() {
return (
<PantheonChat
baseUrl="PANTHEON_BASE_URL"
apiKey="PANTHEON_API_KEY"
agentSlug="todo-assistant"
userId={signedInUserId}
tools={{
echo_tool: async (input) => ({ echoed: input.message }),
todo_ops: {
requireApproval: true,
execute: async (input) => ({ applied: true, operations: input.operations }),
},
}}
/>
);
}requireApproval shows an in-page Approve or Decline card.
Node client
Use the core client with Node 18 or later.
app.mjs
import { PantheonClient } from "@alethi/pantheon-chat";
const baseUrl = process.env.PANTHEON_BASE_URL;
const apiKey = process.env.PANTHEON_API_KEY;
const client = new PantheonClient({ baseUrl, apiKey });
const session = await client.createSession({
agentSlug: "todo-assistant",
userId: signedInUserId,
});
const run = await client.startRun(session.id, "Hello");
for await (const event of client.streamEvents(session.id)) {
if (event.type === "message.delta") process.stdout.write(event.delta.text || "");
if (event.type === "run.completed") break;
}Identify the person chatting
Pass the id of the tenant user this conversation belongs to. The agent keeps long-lived notes about whoever it is talking to, and this is what tells Pantheon who that is.
With it, those notes follow the person across their sessions. Without it, they stay inside the single session, so nothing the agent learns from one visitor can reach another. Read Agent memory before deciding.
Props
| Prop | Use |
|---|---|
baseUrl | Pantheon API base URL. |
apiKey | Bearer API key. |
agentSlug | Agent for a new session. |
deploymentId | Explicit deployment instead of a slug. |
userId | Which of your end users is chatting. It decides whose agent memory the conversation reads. Omit it and memory stays inside this one session. See Agent memory. |
tools | Application tool handlers and approval rules. |
theme | light, dark, or auto. |
title | Session title. |
Python
Use the async client directly.
app.py
import asyncio
import os
from pantheon_chat import PantheonClient
async def main():
async with PantheonClient(
os.environ["PANTHEON_BASE_URL"],
os.environ["PANTHEON_API_KEY"],
) as client:
session = await client.create_session(
agent_slug="todo-assistant",
user_id=signed_in_user_id,
)
run = await client.start_run(session.id, "What is 15 + 15?")
async for event in client.stream_events(session.id, run_id=run.run_id, stop_on_terminal=True):
print(event.type, event.content or event.data)
asyncio.run(main())Terminal chat
Terminal
export PANTHEON_BASE_URL="https://pantheon-todo.alethiconsulting.com/pantheon-api"
export PANTHEON_API_KEY="PANTHEON_API_KEY"
export PANTHEON_AGENT_SLUG="todo-assistant"
python -m pantheon_chatApplication tools ask for approval. Pass --auto-approve only for trusted tools.
Streamlit
Terminal
pip install "pantheon-chat[streamlit]"
export PANTHEON_BASE_URL="https://pantheon-todo.alethiconsulting.com/pantheon-api"
export PANTHEON_API_KEY="PANTHEON_API_KEY"
export PANTHEON_AGENT_SLUG="todo-assistant"
pantheon-chat-streamlitPitfalls
- Queued text can wait behind approval: Resolve the approval before expecting the queued message to run.
- A network reconnect repeats output: Persist only durable SSE ids. Never use an ephemeral delta's JSON id as the cursor.
- A Stop request fails: Keep the run active and show the error until cancel returns HTTP 202.