Custom API Endpoints
Custom API endpoints allow you to extend Chainlit’s Fast API backend with your own endpoints. One good use case for this is to serve an agent to external services.
If you create or update a custom endpoint, you will have to restart your Chainlit process. The file watcher will not pick up changes.
How it works
To add a custom endpoint, import the FastAPI app
instance from chainlit.server
and declare your endpoint as you would in FastAPI.
from chainlit.server import app
from fastapi import Request
from fastapi.responses import (
HTMLResponse,
)
@app.get("/hello")
def hello(request: Request):
print(request.headers)
return HTMLResponse("Hello World")
Use Chainlit APIs
To use Chainlit APIs, you have to init a Chainlit context.
HTTP context
In HTTP context, the Chainlit APIs only logs data (if data persistence is enabled) to the platform. These intermediary Chainlit APIs (Message, Image, etc.) will not send any data to the client. The data sent back to the client is defined by the HTTP response.
from chainlit.server import app
from fastapi import Request
from fastapi.responses import (
HTMLResponse,
)
from chainlit.context import init_http_context
import chainlit as cl
@app.get("/hello")
async def hello(
request: Request,
):
init_http_context()
await cl.Message(content="Hello World").send()
return HTMLResponse("Hello World")
Websocket context
Websocket context is the same context used by your regular Chainlit app. The only use case that requires to use the Websocket context within an HTTP request is to send data to a websocket client (which you know the session ID for) based on some arbitrary HTTP request.
from fastapi import Request
from fastapi.responses import HTMLResponse
from chainlit.server import app
from chainlit.context import init_ws_context
from chainlit.session import WebsocketSession
import chainlit as cl
@cl.on_chat_start
def main():
print("Session id:", cl.user_session.get("id"))
@app.get("/hello/{session_id}")
async def hello(
request: Request,
session_id: str,
):
ws_session = WebsocketSession.get_by_id(session_id=session_id)
init_ws_context(ws_session)
await cl.Message(content="Hello World").send()
return HTMLResponse("Data sent to the websocket client")
Authentication
You can use any authentication system since the request is accessible. However Chainlit authentication is fully compatible with custom endpoints.
from typing_extensions import Annotated
from fastapi import Request, Depends
from fastapi.responses import (
HTMLResponse,
)
from chainlit.server import app
from chainlit.context import init_http_context
from chainlit.auth import authenticate_user
import chainlit as cl
@app.get("/hello")
async def hello(
request: Request,
current_user: Annotated[
Union[cl.AppUser, cl.PersistedAppUser], Depends(authenticate_user)
],
):
init_http_context(user=current_user)
await cl.Message(content="Hello World").send()
return HTMLResponse("Hello World")
Once an endpoint is protected, it will require to have a valid token in the Authorization
header.
{
"Authorization": "Bearer TOKEN"
}
Generate a token
The token is the same token generated when you login in the Chainlit app. You can generate a token manually for a given user with this python script (not a Chainlit app).
This will require to have a CHAINLIT_AUTH_SECRET. You can run chainlit create-secret
to create one.
from chainlit.auth import create_jwt
from chainlit.client.base import AppUser
print(create_jwt(AppUser(username="USERNAME")))
If data persistence is enabled, authentication will fail if the user your signed a token for does not exist. Here is a python script (not a Chainlit app) to create an app user and then sign a token for it.
import os
import asyncio
from chainlit.client.cloud import ChainlitCloudClient, AppUser
from chainlit.auth import create_jwt
CHAINLIT_API_KEY = os.environ.get("CHAINLIT_API_KEY")
client = ChainlitCloudClient(CHAINLIT_API_KEY)
async def main():
# Create the user
app_user = AppUser(username="foobar")
persisted_app_user = await client.create_app_user(app_user)
# Sign the token
print(create_jwt(persisted_app_user))
loop = asyncio.get_event_loop()
loop.run_until_complete(main())
Was this page helpful?