Identify your users, track their cost, conversations and more.
The strict minimum to enable user tracking is to report a userId, however you can report any property you’d like such as an email or name using an userProps object.
Tracking users with the backend SDK
Identify OpenAI calls
The easiest way to get started tracking users is to send user data with your OpenAI API call.
const res = await openai.chat.completions.create({
model: "gpt-4o",
messages: [{ role: "user", content: "Hello" }],
user: "user123",
userProps: { name: "John" },
})
const res = await openai.chat.completions.create({
model: "gpt-4o",
messages: [{ role: "user", content: "Hello" }],
user: "user123",
userProps: { name: "John" },
})
chat_completion = client.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": "Hello"}],
user_id="user123",
user_props={ "name": "John" }
)
If you’re using LangChain, you can similarly pass user data as metadata.
const chat = new ChatOpenAI({
callbacks: [new LunaryHandler()],
});
const res = await chat.call([new HumanMessage("Hello!")], {
metadata: {
userId: "123",
userProps: { name: "John" },
},
});
const chat = new ChatOpenAI({
callbacks: [new LunaryHandler()],
});
const res = await chat.call([new HumanMessage("Hello!")], {
metadata: {
userId: "123",
userProps: { name: "John" },
},
});
handler = LunaryCallbackHandler()
chat = ChatOpenAI(
callbacks=[handler],
metadata={
"user_id": "user123"
}, # Assigning user ids to models in the metadata
)
Advanced: Inject user into context
When tracking traces, you can inject user data into the context using the identify methods. This will cascade down to all the child runs.
async function TranslatorAgent(input) {
// Some AI queries
// Everything done in this context will be tracked with the user
}
// Wrap the agent with the monitor
const translate = lunary.wrapAgent(TranslatorAgent)
// Using identify to inject the user into the context
const res = await translate(`Hello, what's your name?`)
.identify("user123", { email: "email@example.org" })
async function TranslatorAgent(input) {
// Some AI queries
// Everything done in this context will be tracked with the user
}
// Wrap the agent with the monitor
const translate = lunary.wrapAgent(TranslatorAgent)
// Using identify to inject the user into the context
const res = await translate(`Hello, what's your name?`)
.identify("user123", { email: "email@example.org" })
import lunary
def my_agent():
# Some AI queries
# Everything done in this context will be tracked with the user
def main():
# Using identify to inject the user into the context
with lunary.identify('user123', user_props={"email": "email@example.org"}):
my_agent()
Identifying users on the frontend
If you are tracking chat messages or feedback on the frontend, you can use the identify method to identify the user there.
lunary.identify("user123", {
email: "test@example.org",
});
Identifying Threads
If you are using threads to track conversations, you can pass userId and userProps to the openThread method.
const thread = await lunary.openThread({
userId: "user123",
userProps: { name: "John" },
});
User Properties
While you can track any property you’d like, we recommend using the following ones:
| Property | Description |
|---|
name | Name of the user |
email | Email of the user |
avatar | URL to an avatar |
group | Group or company ID the user belongs to |
