Chapter 150 · LLM Analytics Setup
Subchapter 150.16
references/instructor.mdMarkdown6 KBView on GitHub
Required
Setting up analytics starts with installing the PostHog SDK for your language. LLM analytics works best with our Python and Node SDKs.
PostHog AI
pip install posthognpm install @posthog/ai posthog-node2
Required
Install Instructor and the OpenAI SDK. PostHog instruments your LLM calls by wrapping the OpenAI client, which Instructor uses under the hood.
PostHog AI
pip install instructor openainpm install @instructor-ai/instructor openai zod@33
Required
Initialize PostHog with your project token and host from your project settings (opens in a new tab), then create a PostHog OpenAI wrapper and pass it to Instructor.
PostHog AI
import instructor
from pydantic import BaseModel
from posthog.ai.openai import OpenAI
from posthog import Posthog
posthog = Posthog(
"<ph_project_token>",
host="https://us.i.posthog.com"
)
openai_client = OpenAI(
api_key="your_openai_api_key",
posthog_client=posthog
)
client =import Instructor from '@instructor-ai/instructor'
import { OpenAI } from '@posthog/ai'
import { PostHog } from 'posthog-node'
import { z } from 'zod'
const phClient = new PostHog(
'<ph_project_token>',
{ host: 'https://us.i.posthog.com' }
);
const openai =
How this works
PostHog’s OpenAI wrapper is a proper subclass of openai.OpenAI, so it works directly with instructor.from_openai(). PostHog captures $ai_generation events automatically without proxying your calls.
4
Required
Now use Instructor to extract structured data from LLM responses. PostHog automatically captures an $ai_generation event for each call.
PostHog AI
class UserInfo(BaseModel):
name: str
age: int
user = client.chat.completions.create(
model="gpt-5-mini",
response_model=UserInfo,
messages=[
{"role": "user", "content": "John Doe is 30 years old."}
],
posthog_distinct_id="user_123",
posthog_trace_id="trace_123",
posthog_properties={"conversation_id": "abc123"},
)
print(f"{user.name} is {user.age} years old")const UserInfo = z.object({
name: z.string(),
age: z.number(),
})
const user = await client.chat.completions.create({
model: 'gpt-5-mini',
response_model: { schema: UserInfo, name: 'UserInfo' },
messages: [
{ role: 'user', content:
You can expect captured $ai_generation events to have the following properties:
| Property | Description |
|---|---|
| $ai_model | The specific model, like gpt-5-mini or claude-4-sonnet |
| $ai_latency | The latency of the LLM call in seconds |
| $ai_time_to_first_token | Time to first token in seconds (streaming only) |
| $ai_tools | Tools and functions available to the LLM |
| $ai_input | List of messages sent to the LLM |
| $ai_input_tokens | The number of tokens in the input (often found in response.usage) |
| $ai_output_choices | List of response choices from the LLM |
| $ai_output_tokens | The number of tokens in the output (often found in response.usage) |
| $ai_total_cost_usd | The total cost in USD (input + output) |
| […] (opens in a new tab) | See full list (opens in a new tab) of properties |
Recommended
Confirm LLM events are being sent to PostHog
Let’s make sure LLM events are being captured and sent to PostHog. Under LLM analytics, you should see rows of data appear in the Traces and Generations tabs.


5
Recommended
Now that you’re capturing AI conversations, continue with the resources below to learn what else LLM Analytics enables within the PostHog platform.
| Resource | Description |
|---|---|
| Basics (opens in a new tab) | Learn the basics of how LLM calls become events in PostHog. |
| Generations (opens in a new tab) | Read about the $ai_generation event and its properties. |
| Traces (opens in a new tab) | Explore the trace hierarchy and how to use it to debug LLM calls. |
| Spans (opens in a new tab) | Review spans and their role in representing individual operations. |
| Anaylze LLM performance (opens in a new tab) | Learn how to create dashboards to analyze LLM performance. |
Ask a question
HelpfulCould be better