Implementation:BerriAI Litellm Literal AI Logger

Attribute	Value
Sources	`litellm/integrations/literal_ai.py`
Domains	Logging, Observability, LLM Monitoring, Integrations
Last Updated	2026-02-15 16:00 GMT

Overview

The LiteralAILogger is a batch logging integration that sends LLM step data to the Literal AI observability platform via its GraphQL API.

Description

LiteralAILogger extends CustomBatchLogger to collect LLM call metadata -- including messages, token counts, model parameters, thread and parent IDs, prompt management data, and timing -- and sends them as GraphQL ingestStep mutations to the Literal AI API. It supports both sync and async logging, batched GraphQL operations (multiple steps per mutation), configurable batch sizes, and integration with Literal AI's threading and prompt tracking features.

Usage

Import and register LiteralAILogger when you need to send LLM call traces to Literal AI. Requires a LITERAL_API_KEY environment variable or passing the key directly to the constructor.

Code Reference

Source Location

litellm/integrations/literal_ai.py

Signature

class LiteralAILogger(CustomBatchLogger):
    def __init__(
        self,
        literalai_api_key=None,
        literalai_api_url="https://cloud.getliteral.ai",
        env=None,
        **kwargs,
    )

Import

from litellm.integrations.literal_ai import LiteralAILogger

I/O Contract

Inputs

Parameter	Type	Required	Description
`literalai_api_key`	`str`	No	API key. Falls back to `LITERAL_API_KEY` env var.
`literalai_api_url`	`str`	No	API base URL. Defaults to `"https://cloud.getliteral.ai"`. Overridden by `LITERAL_API_URL` env var.
`env`	`str`	No	Optional environment header value sent as `x-env`.
`kwargs`	`dict`	No	Additional arguments passed to `CustomBatchLogger`.

Key Methods

Method	Returns	Description
`log_success_event(kwargs, response_obj, start_time, end_time)`	`None`	Sync success logging. Queues data and flushes when batch size is reached.
`log_failure_event(kwargs, response_obj, start_time, end_time)`	`None`	Sync failure logging.
`async_log_success_event(kwargs, response_obj, start_time, end_time)`	`None`	Async success logging.
`async_log_failure_event(kwargs, response_obj, start_time, end_time)`	`None`	Async failure logging.
`async_send_batch()`	`None`	Sends accumulated steps as a batched GraphQL mutation.

Outputs

Output	Type	Description
Side effect	HTTP POST (GraphQL)	Posts batched `ingestStep` mutations to `{api_url}/api/graphql`.

Usage Examples

import litellm

# Register the callback
litellm.success_callback = ["literalai"]

response = litellm.completion(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello!"}],
    metadata={
        "literalai_thread_id": "thread-abc-123",
        "literalai_parent_id": "parent-step-id",
        "literalai_tags": ["production", "v2"],
    },
)

# Direct instantiation
from litellm.integrations.literal_ai import LiteralAILogger

logger = LiteralAILogger(
    literalai_api_key="my-api-key",
    literalai_api_url="https://cloud.getliteral.ai",
    env="production",
)

Related Pages

BerriAI_Litellm_Argilla_Logger - another batch logging integration
BerriAI_Litellm_Braintrust_Logger - another observability integration
BerriAI_Litellm_MLflow_Logger - another tracing integration

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment