Plugins

For usage examples, see Usage - Plugins.

trackOpenAI / track_openai

Wraps an OpenAI client to automatically trace API calls.

TypeScript

function trackOpenAI(
  openaiClient: OpenAI,
  options: TrackOpenAIOptions
): OpenAI

Parameters:

• openaiClient: OpenAI - The OpenAI client instance to wrap
• options: TrackOpenAIOptions - Plugin configuration options

Returns: Wrapped OpenAI client with tracing enabled

TrackOpenAIOptions:

Option	Type	Required	Description
mentioraClient	MentioraClient	Yes	Mentiora client instance for sending traces
threadId	string	No	Thread/conversation ID (UUID v7) for grouping traces
tags	string[]	No	Optional tags to add to all traces
metadata	Record<string, unknown>	No	Optional metadata to add to all traces
captureContent	boolean	No	Whether to capture input/output content (default: true). Set to false for privacy.

Example:

import { trackOpenAI } from '@mentiora.ai/sdk/openai';
import OpenAI from 'openai';

const trackedClient = trackOpenAI(openaiClient, {
  mentioraClient,
  tags: ['production'],
});

Python

def track_openai(
    openai_client: OpenAI | AsyncOpenAI,
    options: TrackOpenAIOptions
) -> OpenAI | AsyncOpenAI

Parameters:

• openai_client: OpenAI | AsyncOpenAI - The OpenAI client instance to wrap (supports both sync and async clients)
• options: TrackOpenAIOptions - Plugin configuration options

Returns: Wrapped OpenAI client with tracing enabled

TrackOpenAIOptions:

Option	Type	Required	Description
mentiora_client	MentioraClient	Yes	Mentiora client instance for sending traces
thread_id	str	No	Thread/conversation ID (UUID v7) for grouping traces
tags	list[str]	No	Optional tags to add to all traces
metadata	dict[str, Any]	No	Optional metadata to add to all traces
capture_content	bool	No	Whether to capture input/output content (default: True). Set to False for privacy.

Example:

from mentiora import track_openai, TrackOpenAIOptions
from openai import AsyncOpenAI

tracked_client = track_openai(
    openai_client,
    TrackOpenAIOptions(
        mentiora_client=mentiora_client,
        tags=['production'],
    ),
)

Captured trace data:

Each traced call produces a TraceEvent with:

• input: All parameters passed to chat.completions.create (model, messages, temperature, tools, etc.), plus a prompt field extracted from the last user message. Multimodal content arrays (text + images) are supported.
• output: Full response data including id, created, system_fingerprint, service_tier, usage, and per-choice logprobs and refusal fields.
• metadata: Includes openai_id (the OpenAI response ID, e.g. chatcmpl-...) and created timestamp for correlation with OpenAI's logs.

For streaming requests, the plugin automatically injects stream_options: { include_usage: true } to capture token usage statistics. Refusal content is accumulated across stream chunks.

MentioraTracingLangChain

Callback handler for automatically tracing LangChain executions.

TypeScript

class MentioraTracingLangChain extends BaseCallbackHandler {
  constructor(options: MentioraTracingLangChainOptions)
}

MentioraTracingLangChainOptions:

Option	Type	Required	Description
mentioraClient	MentioraClient	Yes	Mentiora client instance for sending traces
threadId	string	No	Thread/conversation ID (UUID v7) for grouping traces
tags	string[]	No	Optional tags to add to all traces
metadata	Record<string, unknown>	No	Optional metadata to add to all traces
captureContent	boolean	No	Whether to capture input/output content (default: true). Set to false for privacy.

Example:

import { MentioraTracingLangChain } from '@mentiora.ai/sdk/langchain';

const callback = new MentioraTracingLangChain({
  mentioraClient,
  tags: ['production'],
});

await chain.invoke({ input: '...' }, { callbacks: [callback] });

Note: This class extends LangChain's BaseCallbackHandler and implements all required callback methods for tracing LLM calls, chain executions, tool calls, and agent operations.

Python

class MentioraTracingLangChain(AsyncCallbackHandler):
    def __init__(self, options: MentioraTracingLangChainOptions)

MentioraTracingLangChainOptions:

Option	Type	Required	Description
mentiora_client	MentioraClient	Yes	Mentiora client instance for sending traces
thread_id	str	No	Thread/conversation ID (UUID v7) for grouping traces
tags	list[str]	No	Optional tags to add to all traces
metadata	dict[str, Any]	No	Optional metadata to add to all traces
capture_content	bool	No	Whether to capture input/output content (default: True). Set to False for privacy.

Example:

from mentiora import MentioraTracingLangChain, MentioraTracingLangChainOptions

callback = MentioraTracingLangChain(MentioraTracingLangChainOptions(
    mentiora_client=mentiora_client,
    tags=['production'],
))

await chain.ainvoke({'input': '...'}, {'callbacks': [callback]})

Note: This class extends LangChain's AsyncCallbackHandler and implements all required callback methods for tracing LLM calls, chain executions, tool calls, and agent operations.

---

See also: Tracing | Streaming Helpers