Instrumentation API¶

Auto-instrumentation for popular LLM SDKs and frameworks.

Base Instrumentor¶

`prela.instrumentation.base.Instrumentor` ¶

Bases: ABC

Abstract base class for library instrumentors.

Instrumentors provide automatic tracing for external libraries by monkey-patching their functions to create spans around operations.

Example

class OpenAIInstrumentor(Instrumentor):
    def instrument(self, tracer: Tracer) -> None:
        # Wrap OpenAI API calls
        wrap_function(openai, "create", wrapper)

    def uninstrument(self) -> None:
        # Restore original functions
        unwrap_function(openai, "create")

    @property
    def is_instrumented(self) -> bool:
        return hasattr(openai, _ORIGINALS_ATTR)

Attributes¶

`is_instrumented` `abstractmethod` `property` ¶

Check if this library is currently instrumented.

Returns:

Type	Description
`bool`	True if instrumentation is active, False otherwise

Functions¶

`instrument(tracer)` `abstractmethod` ¶

Enable instrumentation for this library.

This method should wrap the library's functions to create spans automatically. It should be idempotent - calling it multiple times should not create multiple layers of wrapping.

Parameters:

Name	Type	Description	Default
`tracer`	`Tracer`	The tracer to use for creating spans	required

Raises:

Type	Description
`RuntimeError`	If instrumentation fails

`uninstrument()` `abstractmethod` ¶

Disable instrumentation and restore original functions.

This method should unwrap all previously wrapped functions and restore the library to its original state. It should be idempotent - calling it when not instrumented should be a no-op.

Raises:

Type	Description
`RuntimeError`	If uninstrumentation fails

Auto-Instrumentation¶

`prela.instrumentation.auto.auto_instrument(tracer)` ¶

Automatically instrument all detected libraries.

This function: 1. Checks which supported LLM SDKs are installed 2. Imports and initializes their instrumentors 3. Calls instrument(tracer) on each 4. Returns list of successfully instrumented libraries

The function is designed to be safe: - Missing libraries are skipped (not an error) - Instrumentation failures are logged but don't crash - Returns empty list if nothing was instrumented

Parameters:

Name	Type	Description	Default
`tracer`	`Tracer`	The tracer instance to use for instrumentation	required

Returns:

Type	Description
`list[str]`	List of library names that were successfully instrumented
`list[str]`	(e.g., ["anthropic", "openai"])

Example

from prela.core.tracer import Tracer
from prela.instrumentation.auto import auto_instrument

tracer = Tracer(service_name="my-app")
instrumented = auto_instrument(tracer)
print(f"Auto-instrumented: {instrumented}")
# Output: Auto-instrumented: ['anthropic', 'openai']

# Now all calls to these SDKs are automatically traced
from anthropic import Anthropic
client = Anthropic()
response = client.messages.create(...)  # Automatically traced!

OpenAI Instrumentor¶

`prela.instrumentation.openai.OpenAIInstrumentor` ¶

Bases: Instrumentor

Instrumentor for OpenAI SDK.

Patches the following methods: - openai.OpenAI.chat.completions.create (sync) - openai.AsyncOpenAI.chat.completions.create (async) - openai.OpenAI.completions.create (sync, legacy) - openai.OpenAI.embeddings.create (sync)

Captures detailed information about requests, responses, tool usage, and streaming events.

Attributes¶

`is_instrumented` `property` ¶

Check if OpenAI SDK is currently instrumented.

Functions¶

`init()` ¶

Initialize the OpenAI instrumentor.

`instrument(tracer)` ¶

Enable instrumentation for OpenAI SDK.

Parameters:

Name	Type	Description	Default
`tracer`	`Tracer`	The tracer to use for creating spans	required

Raises:

Type	Description
`ImportError`	If openai package is not installed
`RuntimeError`	If instrumentation fails

`uninstrument()` ¶

Disable instrumentation and restore original functions.

Anthropic Instrumentor¶

`prela.instrumentation.anthropic.AnthropicInstrumentor` ¶

Bases: Instrumentor

Instrumentor for Anthropic SDK.

Patches the following methods: - anthropic.Anthropic.messages.create (sync) - anthropic.AsyncAnthropic.messages.create (async) - anthropic.Anthropic.messages.stream (sync) - anthropic.AsyncAnthropic.messages.stream (async)

Captures detailed information about requests, responses, tool usage, and streaming events.

Attributes¶

`is_instrumented` `property` ¶

Check if Anthropic SDK is currently instrumented.

Functions¶

`init()` ¶

Initialize the Anthropic instrumentor.

`instrument(tracer)` ¶

Enable instrumentation for Anthropic SDK.

Parameters:

Name	Type	Description	Default
`tracer`	`Tracer`	The tracer to use for creating spans	required

Raises:

Type	Description
`ImportError`	If anthropic package is not installed
`RuntimeError`	If instrumentation fails

`uninstrument()` ¶

Disable instrumentation and restore original functions.

LangChain Instrumentor¶

`prela.instrumentation.langchain.LangChainInstrumentor` ¶

Bases: Instrumentor

Instrumentor for LangChain framework.

This instrumentor injects a PrelaCallbackHandler into LangChain's global callback system, which automatically traces all LangChain operations including LLM calls, chains, tools, retrievers, and agent actions.

Unlike other instrumentors that use function wrapping, this uses LangChain's built-in callback mechanism for more robust and comprehensive tracing.

Attributes¶

`is_instrumented` `property` ¶

Check if LangChain is currently instrumented.

Returns:

Type	Description
`bool`	True if instrumentation is active, False otherwise

Functions¶

`init()` ¶

Initialize the LangChain instrumentor.

`instrument(tracer)` ¶

Enable instrumentation for LangChain.

This adds a PrelaCallbackHandler to LangChain's global callback manager, which will receive events for all LangChain operations.

Parameters:

Name	Type	Description	Default
`tracer`	`Tracer`	The tracer to use for creating spans	required

Raises:

Type	Description
`ImportError`	If langchain-core package is not installed
`RuntimeError`	If instrumentation fails

`uninstrument()` ¶

Disable instrumentation and remove callback handler.

Raises:

Type	Description
`RuntimeError`	If uninstrumentation fails

PrelaCallbackHandler¶

`prela.instrumentation.langchain.PrelaCallbackHandler` ¶

LangChain callback handler that creates Prela spans.

This handler implements LangChain's BaseCallbackHandler interface and creates spans for all major LangChain operations. It maintains a mapping from run_id to span to properly handle concurrent executions and nested operations.

The handler tracks: - LLM calls: Model invocations with prompts and responses - Chains: Sequential operations and workflows - Tools: External tool invocations - Retrievers: Document retrieval operations - Agents: Agent reasoning and actions

Functions¶

`init(tracer)` ¶

Initialize the callback handler.

Parameters:

Name	Type	Description	Default
`tracer`	`Tracer`	The tracer to use for creating spans	required

`on_llm_start(serialized, prompts, *, run_id, parent_run_id=None, tags=None, metadata=None, **kwargs)` ¶

Called when an LLM starts running.

Parameters:

Name	Type	Description	Default
`serialized`	`dict[str, Any]`	Serialized LLM configuration	required
`prompts`	`list[str]`	Input prompts to the LLM	required
`run_id`	`UUID`	Unique identifier for this LLM run	required
`parent_run_id`	`UUID \| None`	ID of parent operation (if nested)	`None`
`tags`	`list[str] \| None`	Optional tags for categorization	`None`
`metadata`	`dict[str, Any] \| None`	Optional metadata	`None`
`**kwargs`	`Any`	Additional LLM parameters	`{}`

`on_llm_end(response, *, run_id, parent_run_id=None, **kwargs)` ¶

Called when an LLM finishes running.

Parameters:

Name	Type	Description	Default
`response`	`Any`	LLM response object	required
`run_id`	`UUID`	Unique identifier for this LLM run	required
`parent_run_id`	`UUID \| None`	ID of parent operation	`None`
`**kwargs`	`Any`	Additional parameters	`{}`

`on_llm_error(error, *, run_id, parent_run_id=None, **kwargs)` ¶

Called when an LLM errors.

Parameters:

Name	Type	Description	Default
`error`	`BaseException`	The error that occurred	required
`run_id`	`UUID`	Unique identifier for this LLM run	required
`parent_run_id`	`UUID \| None`	ID of parent operation	`None`
`**kwargs`	`Any`	Additional parameters	`{}`

`on_chain_start(serialized, inputs, *, run_id, parent_run_id=None, tags=None, metadata=None, **kwargs)` ¶

Called when a chain starts running.

Parameters:

Name	Type	Description	Default
`serialized`	`dict[str, Any]`	Serialized chain configuration	required
`inputs`	`dict[str, Any]`	Input values to the chain	required
`run_id`	`UUID`	Unique identifier for this chain run	required
`parent_run_id`	`UUID \| None`	ID of parent operation	`None`
`tags`	`list[str] \| None`	Optional tags	`None`
`metadata`	`dict[str, Any] \| None`	Optional metadata	`None`
`**kwargs`	`Any`	Additional parameters	`{}`

`on_chain_end(outputs, *, run_id, parent_run_id=None, **kwargs)` ¶

Called when a chain finishes running.

Parameters:

Name	Type	Description	Default
`outputs`	`dict[str, Any]`	Output values from the chain	required
`run_id`	`UUID`	Unique identifier for this chain run	required
`parent_run_id`	`UUID \| None`	ID of parent operation	`None`
`**kwargs`	`Any`	Additional parameters	`{}`

`on_chain_error(error, *, run_id, parent_run_id=None, **kwargs)` ¶

Called when a chain errors.

Parameters:

Name	Type	Description	Default
`error`	`BaseException`	The error that occurred	required
`run_id`	`UUID`	Unique identifier for this chain run	required
`parent_run_id`	`UUID \| None`	ID of parent operation	`None`
`**kwargs`	`Any`	Additional parameters	`{}`

`on_tool_start(serialized, input_str, *, run_id, parent_run_id=None, tags=None, metadata=None, **kwargs)` ¶

Called when a tool starts running.

Parameters:

Name	Type	Description	Default
`serialized`	`dict[str, Any]`	Serialized tool configuration	required
`input_str`	`str`	Input string to the tool	required
`run_id`	`UUID`	Unique identifier for this tool run	required
`parent_run_id`	`UUID \| None`	ID of parent operation	`None`
`tags`	`list[str] \| None`	Optional tags	`None`
`metadata`	`dict[str, Any] \| None`	Optional metadata	`None`
`**kwargs`	`Any`	Additional parameters	`{}`

`on_tool_end(output, *, run_id, parent_run_id=None, **kwargs)` ¶

Called when a tool finishes running.

Parameters:

Name	Type	Description	Default
`output`	`str`	Output from the tool	required
`run_id`	`UUID`	Unique identifier for this tool run	required
`parent_run_id`	`UUID \| None`	ID of parent operation	`None`
`**kwargs`	`Any`	Additional parameters	`{}`

`on_tool_error(error, *, run_id, parent_run_id=None, **kwargs)` ¶

Called when a tool errors.

Parameters:

Name	Type	Description	Default
`error`	`BaseException`	The error that occurred	required
`run_id`	`UUID`	Unique identifier for this tool run	required
`parent_run_id`	`UUID \| None`	ID of parent operation	`None`
`**kwargs`	`Any`	Additional parameters	`{}`

`on_retriever_start(serialized, query, *, run_id, parent_run_id=None, tags=None, metadata=None, **kwargs)` ¶

Called when a retriever starts.

Parameters:

Name	Type	Description	Default
`serialized`	`dict[str, Any]`	Serialized retriever configuration	required
`query`	`str`	Query string for retrieval	required
`run_id`	`UUID`	Unique identifier for this retriever run	required
`parent_run_id`	`UUID \| None`	ID of parent operation	`None`
`tags`	`list[str] \| None`	Optional tags	`None`
`metadata`	`dict[str, Any] \| None`	Optional metadata	`None`
`**kwargs`	`Any`	Additional parameters	`{}`

`on_retriever_end(documents, *, run_id, parent_run_id=None, **kwargs)` ¶

Called when a retriever finishes.

Parameters:

Name	Type	Description	Default
`documents`	`list[Any]`	Retrieved documents	required
`run_id`	`UUID`	Unique identifier for this retriever run	required
`parent_run_id`	`UUID \| None`	ID of parent operation	`None`
`**kwargs`	`Any`	Additional parameters	`{}`

LlamaIndex Instrumentor¶

`prela.instrumentation.llamaindex.LlamaIndexInstrumentor` ¶

Bases: Instrumentor

Instrumentor for LlamaIndex framework.

This instrumentor adds automatic tracing to LlamaIndex operations by injecting a callback handler into the global callback manager.

Example

from prela.instrumentation.llamaindex import LlamaIndexInstrumentor
from prela.core.tracer import Tracer

tracer = Tracer()
instrumentor = LlamaIndexInstrumentor()
instrumentor.instrument(tracer)

# All LlamaIndex operations now traced

Attributes¶

`is_instrumented` `property` ¶

Check if LlamaIndex is currently instrumented.

Returns:

Type	Description
`bool`	True if instrumented, False otherwise

Functions¶

`init()` ¶

Initialize the instrumentor.

`instrument(tracer)` ¶

Enable instrumentation for LlamaIndex.

Parameters:

Name	Type	Description	Default
`tracer`	`Tracer`	The tracer to use for creating spans	required

Raises:

Type	Description
`RuntimeError`	If llama-index-core is not installed

`uninstrument()` ¶

Disable instrumentation and remove callback handler.

PrelaHandler¶

`prela.instrumentation.llamaindex.PrelaHandler` ¶

LlamaIndex callback handler that creates Prela spans.

This handler implements LlamaIndex's BaseCallbackHandler interface and creates spans for all major LlamaIndex operations. It maintains a mapping from event_id to span to properly handle concurrent executions and nested operations.

The handler tracks: - LLM calls: Model invocations with prompts and responses - Embeddings: Vector generation operations - Retrieval: Document retrieval with similarity scores - Query: Query engine operations - Synthesis: Response synthesis from retrieved documents

Functions¶

`init(tracer)` ¶

Initialize the callback handler.

Parameters:

Name	Type	Description	Default
`tracer`	`Tracer`	The tracer to use for creating spans	required

`on_event_start(event_type, payload=None, event_id='', parent_id='', **kwargs)` ¶

Called when an event starts.

Parameters:

Name	Type	Description	Default
`event_type`	`str`	Type of event (LLM, EMBEDDING, RETRIEVE, etc.)	required
`payload`	`Optional[dict[str, Any]]`	Event-specific data	`None`
`event_id`	`str`	Unique identifier for this event	`''`
`parent_id`	`str`	ID of parent event (if nested)	`''`
`**kwargs`	`Any`	Additional arguments	`{}`

Returns:

Type	Description
`str`	The event_id for tracking

`on_event_end(event_type, payload=None, event_id='', **kwargs)` ¶

Called when an event ends.

Parameters:

Name	Type	Description	Default
`event_type`	`str`	Type of event (LLM, EMBEDDING, RETRIEVE, etc.)	required
`payload`	`Optional[dict[str, Any]]`	Event-specific response data	`None`
`event_id`	`str`	Unique identifier for this event	`''`
`**kwargs`	`Any`	Additional arguments	`{}`

N8N Webhook Handler¶

`prela.instrumentation.n8n.webhook.N8nWebhookHandler` ¶

HTTP server for receiving n8n webhook traces locally.

This handler runs a lightweight HTTP server that receives webhook POST requests from n8n workflows and automatically converts them into Prela spans.

Example

from prela import init
from prela.instrumentation.n8n.webhook import N8nWebhookHandler

tracer = init(service_name="n8n-workflows")
handler = N8nWebhookHandler(tracer, port=8787)
handler.start()

# Configure n8n webhook node to POST to http://localhost:8787/webhook
# Handler will automatically trace all workflow executions

Functions¶

`init(tracer, port=8787, host='0.0.0.0')` ¶

Initialize the webhook handler.

Parameters:

Name	Type	Description	Default
`tracer`	`Tracer`	Prela tracer instance for creating spans	required
`port`	`int`	Port to listen on (default: 8787)	`8787`
`host`	`str`	Host to bind to (default: 0.0.0.0)	`'0.0.0.0'`

`start()` ¶

Start the HTTP server.

This method starts an aiohttp server on the configured host and port. It runs in the current event loop, so it should be called from an async context or run in a separate thread.

`start_background()` ¶

Start the HTTP server in a background thread.

This method creates a new event loop and runs the server in it. Designed to be called from a background thread via threading.Thread.

`stop()` ¶

Stop the HTTP server.

N8N Code Node Context¶

`prela.instrumentation.n8n.code_node.PrelaN8nContext` ¶

Context manager for tracing custom Python code within n8n Code nodes.

This class creates a workflow-level span and node-level span that properly integrate with Prela's tracing infrastructure. It provides helper methods for logging LLM calls, tool calls, and retrieval operations.

Example

# Inside n8n Code node
from prela.instrumentation.n8n import PrelaN8nContext

ctx = PrelaN8nContext(
    workflow_id=$workflow.id,
    workflow_name=$workflow.name,
    execution_id=$execution.id,
    node_name=$node.name
)

with ctx:
    # Your custom code
    response = call_my_llm(prompt)
    ctx.log_llm_call(
        model="gpt-4",
        prompt=prompt,
        response=response,
        tokens={"prompt": 100, "completion": 50}
    )

    return [{"json": {"result": response}}]

Functions¶

`init(workflow_id, workflow_name, execution_id, node_name, node_type='n8n-nodes-base.code', tracer=None, api_key=None, endpoint=None)` ¶

Initialize n8n Code node tracing context.

Parameters:

Name	Type	Description	Default
`workflow_id`	`str`	Unique workflow identifier	required
`workflow_name`	`str`	Human-readable workflow name	required
`execution_id`	`str`	Unique execution identifier	required
`node_name`	`str`	Name of the Code node	required
`node_type`	`str`	Node type identifier (default: n8n-nodes-base.code)	`'n8n-nodes-base.code'`
`tracer`	`Optional[Tracer]`	Prela tracer instance (defaults to global tracer)	`None`
`api_key`	`Optional[str]`	Optional API key for remote export	`None`
`endpoint`	`Optional[str]`	Optional endpoint URL for remote export	`None`

`enter()` ¶

Start tracing context.

`exit(exc_type, exc_val, exc_tb)` ¶

End tracing context and export spans.

`log_llm_call(model, prompt, response, tokens=None, provider=None, temperature=None, **kwargs)` ¶

Log an LLM call within the Code node.

Parameters:

Name	Type	Description	Default
`model`	`str`	Model identifier (e.g., "gpt-4", "claude-3-opus")	required
`prompt`	`str`	Input prompt text	required
`response`	`str`	Model response text	required
`tokens`	`Optional[dict]`	Token usage dict with "prompt", "completion", "total" keys	`None`
`provider`	`Optional[str]`	AI provider (openai, anthropic, etc.)	`None`
`temperature`	`Optional[float]`	Temperature parameter used	`None`
`**kwargs`		Additional attributes to attach to the span	`{}`

`log_tool_call(tool_name, input, output, error=None, **kwargs)` ¶

Log a tool call within the Code node.

Parameters:

Name	Type	Description	Default
`tool_name`	`str`	Name of the tool/function called	required
`input`	`Any`	Input parameters to the tool	required
`output`	`Any`	Output result from the tool	required
`error`	`Optional[str]`	Optional error message if tool failed	`None`
`**kwargs`		Additional attributes to attach to the span	`{}`

`log_retrieval(query, documents, retriever_type=None, similarity_top_k=None, **kwargs)` ¶

Log a retrieval/search operation within the Code node.

Parameters:

Name	Type	Description	Default
`query`	`str`	Search query text	required
`documents`	`list[dict]`	Retrieved documents (list of dicts with text/score/metadata)	required
`retriever_type`	`Optional[str]`	Type of retriever used (vector, keyword, hybrid)	`None`
`similarity_top_k`	`Optional[int]`	Number of documents requested	`None`
`**kwargs`		Additional attributes to attach to the span	`{}`

N8N Models¶

N8nWorkflowExecution¶

`prela.instrumentation.n8n.models.N8nWorkflowExecution` ¶

Bases: BaseModel

Represents a complete n8n workflow execution.

Captures high-level metadata about a workflow run, including timing, status, node counts, and aggregate token/cost metrics.

Functions¶

`validate_non_negative(v)` `classmethod` ¶

Ensure counts are non-negative.

`validate_cost(v)` `classmethod` ¶

Ensure cost is non-negative.

`validate_completed_at(v, info)` `classmethod` ¶

Ensure completed_at is after started_at if both exist.

`duration_ms()` ¶

Calculate execution duration in milliseconds.

`to_span_attributes()` ¶

Convert to Prela span attributes.

N8nNodeExecution¶

`prela.instrumentation.n8n.models.N8nNodeExecution` ¶

Bases: BaseModel

Represents execution of a single node within a workflow.

Captures node-level telemetry including input/output data, timing, status, and error information.

Functions¶

`validate_non_negative(v)` `classmethod` ¶

Ensure counts are non-negative.

`validate_completed_at(v, info)` `classmethod` ¶

Ensure completed_at is after started_at if both exist.

`duration_ms()` ¶

Calculate node execution duration in milliseconds.

`to_span_attributes()` ¶

Convert to Prela span attributes.

N8nAINodeExecution¶

`prela.instrumentation.n8n.models.N8nAINodeExecution` ¶

Bases: N8nNodeExecution

Represents execution of an AI-specific node (LLM, agent, vector store, etc.).

Extends N8nNodeExecution with AI-specific telemetry including model info, token usage, costs, prompts, and retrieval data.

Functions¶

`validate_non_negative_tokens(v)` `classmethod` ¶

Ensure token counts are non-negative.

`validate_cost(v)` `classmethod` ¶

Ensure cost is non-negative.

`validate_temperature(v)` `classmethod` ¶

Ensure temperature is in valid range.

`to_span_attributes()` ¶

Convert to Prela span attributes with AI-specific fields.

`infer_span_type()` ¶

Infer the appropriate span type based on node characteristics.

Returns:

Type	Description
`N8nSpanType`	N8nSpanType appropriate for this node's function

Instrumentation API¶

Base Instrumentor¶

prela.instrumentation.base.Instrumentor ¶

Attributes¶

is_instrumented abstractmethod property ¶

Functions¶

instrument(tracer) abstractmethod ¶

uninstrument() abstractmethod ¶

Auto-Instrumentation¶

prela.instrumentation.auto.auto_instrument(tracer) ¶

OpenAI Instrumentor¶

prela.instrumentation.openai.OpenAIInstrumentor ¶

Attributes¶

is_instrumented property ¶

Functions¶

__init__() ¶

instrument(tracer) ¶

uninstrument() ¶

Anthropic Instrumentor¶

prela.instrumentation.anthropic.AnthropicInstrumentor ¶

Attributes¶

is_instrumented property ¶

Functions¶

__init__() ¶

instrument(tracer) ¶

uninstrument() ¶

LangChain Instrumentor¶

prela.instrumentation.langchain.LangChainInstrumentor ¶

Attributes¶

is_instrumented property ¶

Functions¶

__init__() ¶

instrument(tracer) ¶

uninstrument() ¶

PrelaCallbackHandler¶

prela.instrumentation.langchain.PrelaCallbackHandler ¶

Functions¶

__init__(tracer) ¶

on_llm_start(serialized, prompts, *, run_id, parent_run_id=None, tags=None, metadata=None, **kwargs) ¶

on_llm_end(response, *, run_id, parent_run_id=None, **kwargs) ¶

on_llm_error(error, *, run_id, parent_run_id=None, **kwargs) ¶

on_chain_start(serialized, inputs, *, run_id, parent_run_id=None, tags=None, metadata=None, **kwargs) ¶

on_chain_end(outputs, *, run_id, parent_run_id=None, **kwargs) ¶

on_chain_error(error, *, run_id, parent_run_id=None, **kwargs) ¶

on_tool_start(serialized, input_str, *, run_id, parent_run_id=None, tags=None, metadata=None, **kwargs) ¶

on_tool_end(output, *, run_id, parent_run_id=None, **kwargs) ¶

on_tool_error(error, *, run_id, parent_run_id=None, **kwargs) ¶

on_retriever_start(serialized, query, *, run_id, parent_run_id=None, tags=None, metadata=None, **kwargs) ¶

on_retriever_end(documents, *, run_id, parent_run_id=None, **kwargs) ¶

LlamaIndex Instrumentor¶

prela.instrumentation.llamaindex.LlamaIndexInstrumentor ¶

Attributes¶

is_instrumented property ¶

Functions¶

__init__() ¶

instrument(tracer) ¶

uninstrument() ¶

PrelaHandler¶

prela.instrumentation.llamaindex.PrelaHandler ¶

Functions¶

__init__(tracer) ¶

on_event_start(event_type, payload=None, event_id='', parent_id='', **kwargs) ¶

on_event_end(event_type, payload=None, event_id='', **kwargs) ¶

N8N Webhook Handler¶

prela.instrumentation.n8n.webhook.N8nWebhookHandler ¶

Functions¶

__init__(tracer, port=8787, host='0.0.0.0') ¶

start() ¶

start_background() ¶

stop() ¶

N8N Code Node Context¶

prela.instrumentation.n8n.code_node.PrelaN8nContext ¶

Functions¶

__init__(workflow_id, workflow_name, execution_id, node_name, node_type='n8n-nodes-base.code', tracer=None, api_key=None, endpoint=None) ¶

__enter__() ¶

__exit__(exc_type, exc_val, exc_tb) ¶

log_llm_call(model, prompt, response, tokens=None, provider=None, temperature=None, **kwargs) ¶

log_tool_call(tool_name, input, output, error=None, **kwargs) ¶

log_retrieval(query, documents, retriever_type=None, similarity_top_k=None, **kwargs) ¶

N8N Models¶

`prela.instrumentation.base.Instrumentor` ¶

`is_instrumented` `abstractmethod` `property` ¶

`instrument(tracer)` `abstractmethod` ¶

`uninstrument()` `abstractmethod` ¶

`prela.instrumentation.auto.auto_instrument(tracer)` ¶

`prela.instrumentation.openai.OpenAIInstrumentor` ¶

`is_instrumented` `property` ¶

`init()` ¶

`instrument(tracer)` ¶

`uninstrument()` ¶

`prela.instrumentation.anthropic.AnthropicInstrumentor` ¶

`is_instrumented` `property` ¶

`init()` ¶

`instrument(tracer)` ¶

`uninstrument()` ¶

`prela.instrumentation.langchain.LangChainInstrumentor` ¶

`is_instrumented` `property` ¶

`init()` ¶

`instrument(tracer)` ¶

`uninstrument()` ¶

`prela.instrumentation.langchain.PrelaCallbackHandler` ¶

`init(tracer)` ¶

`on_llm_start(serialized, prompts, *, run_id, parent_run_id=None, tags=None, metadata=None, **kwargs)` ¶

`on_llm_end(response, *, run_id, parent_run_id=None, **kwargs)` ¶

`on_llm_error(error, *, run_id, parent_run_id=None, **kwargs)` ¶

`on_chain_start(serialized, inputs, *, run_id, parent_run_id=None, tags=None, metadata=None, **kwargs)` ¶

`on_chain_end(outputs, *, run_id, parent_run_id=None, **kwargs)` ¶

`on_chain_error(error, *, run_id, parent_run_id=None, **kwargs)` ¶

`on_tool_start(serialized, input_str, *, run_id, parent_run_id=None, tags=None, metadata=None, **kwargs)` ¶

`on_tool_end(output, *, run_id, parent_run_id=None, **kwargs)` ¶

`on_tool_error(error, *, run_id, parent_run_id=None, **kwargs)` ¶

`on_retriever_start(serialized, query, *, run_id, parent_run_id=None, tags=None, metadata=None, **kwargs)` ¶

`on_retriever_end(documents, *, run_id, parent_run_id=None, **kwargs)` ¶

`prela.instrumentation.llamaindex.LlamaIndexInstrumentor` ¶

`is_instrumented` `property` ¶

`init()` ¶

`instrument(tracer)` ¶

`uninstrument()` ¶

`prela.instrumentation.llamaindex.PrelaHandler` ¶

`init(tracer)` ¶

`on_event_start(event_type, payload=None, event_id='', parent_id='', **kwargs)` ¶

`on_event_end(event_type, payload=None, event_id='', **kwargs)` ¶

`prela.instrumentation.n8n.webhook.N8nWebhookHandler` ¶

`init(tracer, port=8787, host='0.0.0.0')` ¶

`start()` ¶

`start_background()` ¶

`stop()` ¶

`prela.instrumentation.n8n.code_node.PrelaN8nContext` ¶

`init(workflow_id, workflow_name, execution_id, node_name, node_type='n8n-nodes-base.code', tracer=None, api_key=None, endpoint=None)` ¶

`enter()` ¶

`exit(exc_type, exc_val, exc_tb)` ¶

`log_llm_call(model, prompt, response, tokens=None, provider=None, temperature=None, **kwargs)` ¶

`log_tool_call(tool_name, input, output, error=None, **kwargs)` ¶

`log_retrieval(query, documents, retriever_type=None, similarity_top_k=None, **kwargs)` ¶

`prela.instrumentation.n8n.models.N8nWorkflowExecution` ¶

`validate_non_negative(v)` `classmethod` ¶

`validate_cost(v)` `classmethod` ¶

`validate_completed_at(v, info)` `classmethod` ¶

`duration_ms()` ¶

`to_span_attributes()` ¶

`prela.instrumentation.n8n.models.N8nNodeExecution` ¶

`validate_non_negative(v)` `classmethod` ¶

`validate_completed_at(v, info)` `classmethod` ¶

`duration_ms()` ¶

`to_span_attributes()` ¶

`prela.instrumentation.n8n.models.N8nAINodeExecution` ¶

`validate_non_negative_tokens(v)` `classmethod` ¶

`validate_cost(v)` `classmethod` ¶

`validate_temperature(v)` `classmethod` ¶

`to_span_attributes()` ¶

`infer_span_type()` ¶