Pydantic Logfire

https://pydantic.dev/docs/logfire/
Admin
Pydantic Logfire is an observability platform built on OpenTelemetry that provides end-to-end...
Tokens:522,011
Snippets:4,127
Trust Score:9.5
Update:1 month ago
Show doc for...
Tokens:
Raw
### Install Uvicorn and start the app

Source: https://pydantic.dev/docs/logfire/integrations/web-frameworks/litestar

Install Uvicorn, the server used by the example, and start the app with `python main.py`.

```bash
pip install uvicorn
python main.py
```

--------------------------------

### OpenCode: Interactive setup command

Source: https://pydantic.dev/docs/logfire/guides/mcp-server

Run this command to start the interactive setup for adding the MCP server. Answer the prompts as described in the guide.

```bash
opencode mcp add
```

--------------------------------

### Install Uvicorn server

Source: https://pydantic.dev/docs/logfire/integrations/web-frameworks/asgi

Install Uvicorn, the ASGI server, required to run the example application.

```bash
pip install uvicorn
```

--------------------------------

### Set up environment and run Node.js OpenTelemetry example

Source: https://pydantic.dev/docs/logfire/guides/alternative-clients

Terminal commands to set the OTLP endpoint and authorization header, initialize a package.json with type module, install the @opentelemetry/sdk-node package, and run main.js.

```sh
export OTEL_EXPORTER_OTLP_ENDPOINT=https://logfire-us.pydantic.dev
export OTEL_EXPORTER_OTLP_HEADERS='Authorization=your-write-token'

npm init es6 -y # creates package.json with type module
npm install @opentelemetry/sdk-node
node main.js

```

--------------------------------

### Start local PostgreSQL with Docker for asyncpg example

Source: https://pydantic.dev/docs/logfire/integrations/databases/asyncpg

Start a local PostgreSQL container for testing. Creates a database reachable at postgres://user:secret@127.0.0.1:5432/database.

```bash
docker run --name postgres \
    -e POSTGRES_USER=user \
    -e POSTGRES_PASSWORD=secret \
    -e POSTGRES_DB=database \
    -p 5432:5432 \
    -d postgres
```

--------------------------------

### Install requests for admin automation

Source: https://pydantic.dev/docs/logfire/deploy/self-hosted-deployment/examples

Install the requests library if it is not already available. Required for the instance admin automation example.

```bash
python -m pip install requests
```

--------------------------------

### Start local MongoDB with Docker

Source: https://pydantic.dev/docs/logfire/integrations/databases/pymongo

Start a local MongoDB instance using Docker, which is required for the usage example.

```bash
docker run --name mongo -p 127.0.0.1:27017:27017 -d mongo:latest
```

--------------------------------

### Start Redis broker with Docker

Source: https://pydantic.dev/docs/logfire/integrations/event-streams/celery

Start a Redis broker for the Celery integration example using Docker.

```bash
docker run --rm -d -p 127.0.0.1:6379:6379 redis
```

--------------------------------

### Install and run FastAPI app with Logfire (Terminal)

Source: https://pydantic.dev/docs/logfire/get-started/why

These commands install the required packages and run the FastAPI app. The first command installs `logfire[fastapi]`, `fastapi`, and `uvicorn`. The second command runs the app with uvicorn. This setup is necessary to run the instrumented app.

```bash
pip install 'logfire[fastapi]' fastapi uvicorn  # (1)
uvicorn main:app # (2)
```

--------------------------------

### Python: Add Logfire to your app (hello.py)

Source: https://pydantic.dev/docs/logfire/get-started/first-trace

Python example that configures Logfire, starts a span named 'greeting', and logs an info message inside it. `configure()` connects your app to Logfire, `span()` records one operation, and `info()` writes a log inside it, together forming your first trace.

```py
import logfire

logfire.configure()

with logfire.span('greeting'):
    logfire.info('Hello, {name}!', name='world')
```

--------------------------------

### Full example: initialize OFREP provider, set context, and evaluate flags

Source: https://pydantic.dev/docs/logfire/manage/client-side-feature-flags

Complete setup combining initialization, context, and evaluation. Initializes the OFREP web provider with a custom fetch implementation that adds Authorization and Content-Type headers, sets user context with a targeting key, and evaluates boolean, string, and number flags.

```typescript
import { OFREPWebProvider } from '@openfeature/ofrep-web-provider'
import { OpenFeature } from '@openfeature/web-sdk'

// Initialize once at app startup
function initFeatureFlags(apiKey: string, apiHost: string) {
  const provider = new OFREPWebProvider({
    baseUrl: `https://${apiHost}/v1`,
    fetchImplementation: (input, init) => {
      const headers = new Headers(input instanceof Request ? input.headers : undefined)
      new Headers(init?.headers).forEach((value, key) => headers.set(key, value))
      headers.set('Authorization', `Bearer ${apiKey}`)
      headers.set('Content-Type', 'application/json')
      return fetch(input, { ...init, headers })
    },
  })
  OpenFeature.setProvider(provider)
}

// Set context when user authenticates
async function setUserContext(userId: string, attributes: Record<string, string> = {}) {
  await OpenFeature.setContext({
    targetingKey: userId,
    ...attributes,
  })
}

// Evaluate flags anywhere in your app
function getFeatureFlags() {
  const client = OpenFeature.getClient()
  return {
    showNewDashboard: client.getBooleanValue('show_new_dashboard', false),
    pricingTier: client.getStringValue('pricing_tier_config', 'standard'),
    maxUploadSize: client.getNumberValue('max_upload_size_mb', 10),
  }
}
```

--------------------------------

### Install Letta and letta-client

Source: https://pydantic.dev/docs/logfire/integrations/llms/letta

Install the Letta server and client packages via pip.

```bash
pip install letta letta-client
```

--------------------------------

### logfire projects use backend example

Source: https://pydantic.dev/docs/logfire/reference/cli

Example that uses the `backend` project.

```bash
logfire projects use backend
```

--------------------------------

### Install LlamaIndex instrumentation and dependencies

Source: https://pydantic.dev/docs/logfire/integrations/llms/llamaindex

Installs the OpenTelemetry LlamaIndex instrumentation and required LlamaIndex packages for the example, including OpenAI LLM and web reader support.

```bash
pip install opentelemetry-instrumentation-llamaindex \
    llama-index-core llama-index-llms-openai llama-index-readers-web html2text
```

--------------------------------

### Run Python example

Source: https://pydantic.dev/docs/logfire/get-started/first-trace

Run the Python example. The first run asks you to pick or create a Logfire project, then sends your trace.

```bash
uv run hello.py
```

--------------------------------

### Install and configure OpenAI instrumentation

Source: https://pydantic.dev/docs/logfire/observe/llms

Install the required packages and set the LOGFIRE_TOKEN environment variable. This is the first step to get your first LLM row to appear.

```bash
pip install logfire openai
export LOGFIRE_TOKEN=<your write token from project Settings → Write tokens>
```

--------------------------------

### main.rs: create tracer and export span

Source: https://pydantic.dev/docs/logfire/guides/alternative-clients

Sets up an OTLP HTTP exporter, installs a simple tracer provider, and creates a span named "Hello World". Includes commented-out code for programmatic endpoint and header configuration.

```rust
use opentelemetry::{
    global::ObjectSafeSpan,
    trace::{Tracer, TracerProvider},
};

fn main() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    let otlp_exporter = opentelemetry_otlp::new_exporter()
        .http()
        .with_protocol(opentelemetry_otlp::Protocol::HttpBinary)
        // If you don't want to export environment variables, you can also configure
        // programmatically like so:
        //
        // (You'll need to add `use opentelemetry_otlp::WithExportConfig;` to the top of the
        // file to access the `.with_endpoint` method.)
        //
        // .with_endpoint("https://logfire-us.pydantic.dev/v1/traces")
        // .with_headers({
        //     let mut headers = std::collections::HashMap::new();
        //     headers.insert(
        //         "Authorization".into(),
        //         "your-write-token".into(),
        //     );
        //     headers
        // })
        ;

    let tracer_provider = opentelemetry_otlp::new_pipeline()
        .tracing()
        .with_exporter(otlp_exporter)
        .install_simple()?;
    let tracer = tracer_provider.tracer("my_tracer");

    tracer.span_builder("Hello World").start(&tracer).end();

    Ok(())
}
```

--------------------------------

### Initialize and run the Go module

Source: https://pydantic.dev/docs/logfire/instrument/go

Commands to initialize the Go module, fetch dependencies, and run the program. This is the final step to send the trace to Logfire.

```sh
go mod init hello-go
go mod tidy
go run .

```

--------------------------------

### Example response for GET /proxy/models

Source: https://pydantic.dev/docs/logfire/manage/ai-gateway/embeddings

Example JSON response showing chat models under 'models' and embedding models under 'embedding_models'. The response is one entry per route.

```json
[
  {
    "route": "openai",
    "provider": "openai",
    "models": [{ "id": "gpt-5.2", "name": "gpt-5.2", "context_window": 400000 }],
    "embedding_models": [
      { "id": "text-embedding-3-small", "name": "text-embedding-3-small", "context_window": 8192 },
      { "id": "text-embedding-3-large", "name": "text-embedding-3-large", "context_window": 8192 }
    ]
  }
]
```

--------------------------------

### Install and configure OpenInference LangChain instrumentor

Source: https://pydantic.dev/docs/logfire/integrations/llms/langchain

Install the OpenInference LangChain instrumentation package and configure it with Logfire. This setup marks agent root spans so agents appear on the Agents page, unlike the LangSmith export which only lands traces in the Live view.

```bash
pip install openinference-instrumentation-langchain
```

```python
from openinference.instrumentation.langchain import LangChainInstrumentor

import logfire

logfire.configure()
LangChainInstrumentor().instrument()
```

--------------------------------

### Install OpenFeature Web SDK and OFREP provider

Source: https://pydantic.dev/docs/logfire/manage/client-side-feature-flags

Install the OpenFeature Web SDK and OFREP provider using npm, pnpm, or yarn. Choose the command that matches your package manager.

```bash
npm install @openfeature/web-sdk @openfeature/ofrep-web-provider
```

```bash
pnpm add @openfeature/web-sdk @openfeature/ofrep-web-provider
```

```bash
yarn add @openfeature/web-sdk @openfeature/ofrep-web-provider
```

--------------------------------

### Refresh token with curl

Source: https://pydantic.dev/docs/logfire/manage/oauth-apps

Uses curl to POST to the token endpoint with refresh_token grant, authenticating with client credentials. Requires the environment variables from the setup example.

```bash
curl --fail-with-body --silent --show-error \
  --user "$CLIENT_ID:$CLIENT_SECRET" \
  --data-urlencode 'grant_type=refresh_token' \
  --data-urlencode "refresh_token=$REFRESH_TOKEN" \
  "$LOGFIRE_BASE_URL/api/oauth/token"

```

--------------------------------

### Example console output

Source: https://pydantic.dev/docs/logfire/instrument/python/add-manual-tracing

Example output when running the span and log example, showing the project URL and the log lines with timestamps and indentation.

```text
Logfire project URL: https://logfire.pydantic.dev/my_username/my_project_name
21:02:55.078 This is a span
21:02:56.084   This is an info log

```

--------------------------------

### Install logfire variables extra and fetch a managed prompt for a CrewAI agent

Source: https://pydantic.dev/docs/logfire/integrations/llms/crewai

Install the variables extra for the Logfire SDK, then fetch a versioned prompt and pass it into a CrewAI agent. The Python example defines a BackstoryInputs model, retrieves the prompt with logfire.template_var, and uses the resolved value as the agent's backstory.

```bash
pip install 'logfire[variables]'
```

```python
from crewai import Agent
from pydantic import BaseModel

import logfire

logfire.configure()


class BackstoryInputs(BaseModel):
    domain: str


backstory_var = logfire.template_var(
    name='prompt__researcher_backstory',
    type=str,
    default='You are a knowledgeable analyst who values brevity.',
    inputs_type=BackstoryInputs,
)

with backstory_var.get(BackstoryInputs(domain='observability'), label='production') as resolved:
    backstory = resolved.value

researcher = Agent(
    role='Researcher',
    goal='Explain a topic clearly and concisely',
    backstory=backstory,
    llm='openai/gpt-4o-mini',
)
```

--------------------------------

### Capture specific HTTP request headers

Source: https://pydantic.dev/docs/logfire/integrations/web-frameworks

Set `OTEL_INSTRUMENTATION_HTTP_CAPTURE_HEADERS_SERVER_REQUEST` to a comma-separated list of regexes to capture only matching headers. This example captures the `content-type` header and any header starting with `X-`.

```bash
OTEL_INSTRUMENTATION_HTTP_CAPTURE_HEADERS_SERVER_REQUEST="content-type,X-.*"
```

--------------------------------

### Environment Setup for Verification

Source: https://pydantic.dev/docs/logfire/manage/oauth-apps

Set the API base URL and access token environment variables before listing projects.

```bash
export LOGFIRE_API_BASE_URL='https://api-us.pydantic.dev/api/v1'
export ACCESS_TOKEN='ACCESS_TOKEN_FROM_THE_TOKEN_RESPONSE'
```

--------------------------------

### Add tracing code to hello.php

Source: https://pydantic.dev/docs/logfire/instrument/php

Initialize the tracer and create a manual span named 'Hello World'.

```php
<?php
require __DIR__ . '/vendor/autoload.php';

\OpenTelemetry\API\Globals::tracerProvider()
    ->getTracer('hello-php')
    ->spanBuilder('Hello World')
    ->startSpan()
    ->end();
```

--------------------------------

### Basic auto-tracing setup with main.py

Source: https://pydantic.dev/docs/logfire/instrument/python/add-auto-tracing

Call logfire.install_auto_tracing() before importing the modules to trace. This example traces all functions in the 'app' package that take longer than 0.01 seconds. Generator functions are not traced.

```python
import logfire

logfire.configure()
logfire.install_auto_tracing(modules=['app'], min_duration=0.01)

from app.main import main

main()
```

--------------------------------

### Start Gunicorn with configuration file

Source: https://pydantic.dev/docs/logfire/integrations/web-frameworks/gunicorn

Start Gunicorn using the configuration file that sets up Logfire. Replace myapp:app with your WSGI application.

```bash
gunicorn myapp:app --config gunicorn_config.py
```

--------------------------------

### Sync Logfire Query Client Usage

Source: https://pydantic.dev/docs/logfire/manage/query-api

Example of using the LogfireQueryClient to query data in JSON, Arrow, and CSV formats, and to get read token info. Requires pyarrow for Arrow format.

```python
from datetime import UTC, datetime, timedelta
from io import StringIO

import polars as pl

from logfire.query_client import LogfireQueryClient


def main():
    query = """
    SELECT start_timestamp
    FROM records
    LIMIT 1
    """
    min_timestamp = datetime.now(tz=UTC) - timedelta(hours=2)

    with LogfireQueryClient(read_token='<your_read_token>') as client:
        # Load data as JSON
        json_rows = client.query_json_rows(sql=query, min_timestamp=min_timestamp)
        print(json_rows)
        """
        {
          "columns": [{'name': 'start_timestamp', 'datatype': {'Timestamp': ['Microsecond', 'UTC']}, 'nullable': False}],
          "rows":  [{'start_timestamp': '2026-05-27T13:16:36.517321Z'}]
        }
        """

        # Retrieve data in arrow format, and load into a polars DataFrame
        # Note that JSON columns such as `attributes` will be returned as
        # JSON-serialized strings
        df_from_arrow = pl.from_arrow(client.query_arrow(sql=query, min_timestamp=min_timestamp))
        print(df_from_arrow)

        # Retrieve data in CSV format, and load into a polars DataFrame
        # Note that JSON columns such as `attributes` will be returned as
        # JSON-serialized strings
        df_from_csv = pl.read_csv(StringIO(client.query_csv(sql=query, min_timestamp=min_timestamp)))
        print(df_from_csv)

        # Get read token info
        read_token_info = client.info()
        print(read_token_info)


if __name__ == '__main__':
    main()
```

--------------------------------

### Async Logfire Query Client Usage

Source: https://pydantic.dev/docs/logfire/manage/query-api

Example of using the AsyncLogfireQueryClient to query data in JSON, Arrow, and CSV formats, and to get read token info. Requires pyarrow for Arrow format.

```python
from datetime import UTC, datetime, timedelta
from io import StringIO

import polars as pl

from logfire.query_client import AsyncLogfireQueryClient


async def main():
    query = """
    SELECT start_timestamp
    FROM records
    LIMIT 1
    """
    min_timestamp = datetime.now(tz=UTC) - timedelta(hours=2)

    async with AsyncLogfireQueryClient(read_token='<your_read_token>') as client:
        # Load data as JSON
        json_rows = await client.query_json_rows(sql=query, min_timestamp=min_timestamp)
        print(json_rows)
        """
        {
            "columns": [{'name': 'start_timestamp', 'datatype': {'Timestamp': ['Microsecond', 'UTC']}, 'nullable': False}],
            "rows":  [{'start_timestamp': '2026-05-27T13:16:36.517321Z'}]
        }
        """

        # Retrieve data in arrow format, and load into a polars DataFrame
        # Note that JSON columns such as `attributes` will be returned as
        # JSON-serialized strings
        df_from_arrow = pl.from_arrow(await client.query_arrow(sql=query, min_timestamp=min_timestamp))
        print(df_from_arrow)

        # Retrieve data in CSV format, and load into a polars DataFrame
        # Note that JSON columns such as `attributes` will be returned as
        # JSON-serialized strings
        df_from_csv = pl.read_csv(StringIO(await client.query_csv(sql=query, min_timestamp=min_timestamp)))
        print(df_from_csv)

        # Get read token info
        read_token_info = await client.info()
        print(read_token_info)


if __name__ == '__main__':
    import asyncio

    asyncio.run(main())
```

--------------------------------

### Switch to pydantic-skills marketplace for Claude Code

Source: https://pydantic.dev/docs/logfire/guides/mcp-server

Removes the official plugin and installs from the pydantic/skills marketplace to get a version tracking the latest commit. Run `/reload-plugins` or restart Claude Code after switching.

```bash
claude plugin remove logfire@claude-plugins-official
claude plugin marketplace add pydantic/skills
claude plugin install logfire@pydantic-skills
```

--------------------------------

### Instrument requests library in main.py

Source: https://pydantic.dev/docs/logfire/integrations/http-clients/requests

Add `logfire.configure()` and `logfire.instrument_requests()` to your app to record every outgoing request. The example makes a GET request to `https://httpbin.org/get` and the span appears in the Logfire Live view.

```python
import requests

import logfire

logfire.configure()
logfire.instrument_requests()

requests.get('https://httpbin.org/get')
```

--------------------------------

### Set up local project credentials with CLI

Source: https://pydantic.dev/docs/logfire/instrument/typescript/configuration

Use these commands to authenticate and select a project, which writes local credentials to `.logfire/logfire_credentials.json` for Node.js development.

```bash
npx logfire auth
npx logfire projects use my-project
```
Tokens:
Raw