Add Annotation Queue Item
Add a trace, observation, or session to a Langfuse annotation queue by queue ID and object ID. Optionally set the item's annotation status.
Open source LLM engineering platform. Traces, evals, prompt management and metrics to debug and improve your LLM application.
Add a trace, observation, or session to a Langfuse annotation queue by queue ID and object ID. Optionally set the item's annotation status.
Assign a user to a Langfuse annotation queue by queue ID and user ID. Returns the user, queue, and project IDs for the created assignment.
Check the health of the Langfuse API and database. Returns the server version and health status.
Create an annotation queue in the authenticated Langfuse project for human evaluation, using the supplied score configuration IDs. Returns the created queue and its ID.
Create a comment on an existing Langfuse trace, observation, session, or prompt in the specified project. Supports Markdown content and an optional author who belongs to the project's organization. Returns the created comment ID.
Create a dataset in the connected Langfuse project, optionally with a description, metadata, and JSON Schemas for item inputs and expected outputs. This creates the dataset container; add dataset items separately.
Create a Langfuse dataset item with input, expected output, and metadata. Providing an existing item ID updates that item in the same dataset.
Create a Langfuse evaluation rule for incoming observations, using stable evaluator IDs and their latest versions. Explicitly choose enabled: true starts automatic evaluations and may incur model costs; false creates a disabled draft. Omitted filters match all observations and omitted sampling evaluates every match. Rule names need not be unique.
Create a version-1 evaluator in the authenticated Langfuse project. Choose llm_as_judge with a prompt and structured score output, or code with source code and runtime language. Names need not be unique. An evaluator defines how to score data; evaluation rules separately determine which data to evaluate.
Obtain a presigned upload URL and media ID for Langfuse media associated with either a trace (optionally an observation) or a dataset item. Requires the media MIME type, byte length, SHA-256 hash, and containing field. Only requests the URL; does not upload a file. The URL can be null if already uploaded.
Create a custom Langfuse model definition for usage-to-USD cost calculation and optional tokenization. Supply a model name, matching regex, usage unit, and either pricing tiers (recommended) or legacy flat prices. Each tiered model needs exactly one default tier with priority 0 and no conditions.
Create a new version of a named Langfuse text or chat prompt. Chat prompts require type='chat' and support messages and named placeholders. Optionally attach arbitrary JSON configuration, deployment labels, tags, and a commit message.
Attach an evaluation score to exactly one Langfuse trace, session, or dataset run, optionally narrowing a trace score to an observation.
Create a score configuration in the authenticated Langfuse project to define numeric bounds, categorical label/value pairs, boolean scores, or text scores. Boolean categories are autogenerated; omit categories for BOOLEAN configs. Use the returned config ID when creating scores validated against this configuration.
Delete a Langfuse dataset item by its ID and all its run items. This action is irreversible.
Delete a Langfuse evaluation rule by its stable ID, including legacy trace and dataset rules. This removes the live-ingestion rule only; associated evaluators and previously produced scores are preserved.
Delete a Langfuse evaluator and all of its stored versions by its stable ID. Associated evaluation-rule assignments are also removed. Scores already produced by the evaluator are preserved.
Delete a Langfuse LLM connection by its ID. Evaluators that depend on the deleted connection are automatically paused.
Delete a project-owned Langfuse model by its ID. Models managed by Langfuse cannot be deleted; create your own definition with the same modelName to override a managed definition instead. Returns an empty response on success.
Delete Langfuse prompt versions by prompt name. Optionally filter deletion by label or version. If neither label nor version is specified, all versions of the prompt are deleted. Returns an empty response on success.
Delete a Langfuse score by its ID. Supports both trace and session scores. Returns an empty response on success.
Delete a specific Langfuse trace by its unique trace ID.
Delete multiple Langfuse traces by their trace IDs.
Retrieve one Langfuse annotation queue by its ID.
Get a specific item from a Langfuse annotation queue by queue ID and queue item ID. Returns the associated object, annotation status, and timestamps.
Retrieve one comment by its unique Langfuse ID in the authenticated project. Returns the comment content, associated object type and ID, timestamps, and author user ID when available.
Retrieve a Langfuse dataset by name, including its metadata and JSON Schemas for dataset item inputs and expected outputs.
Retrieve a single Langfuse dataset item by its ID, including its input, expected output, metadata, source IDs, and resolved media references.
Get one Langfuse evaluation rule by its stable identifier, including legacy trace or dataset rules. Returns its enabled state, sampling, stored filters, evaluator assignments, variable mappings, creator, and timestamps.
Get one evaluator by its stable identifier in the authenticated Langfuse project. Returns its latest definition, version metadata, effective runtime status, and associated evaluation-rule assignments. Use version history for older definitions.
Retrieve one media record by its unique Langfuse ID in the authenticated project. Returns the MIME type, size in bytes, upload timestamp, and download URL with its expiry. Does not download the media content.
Retrieve one Langfuse-managed or project-custom model definition by ID. Returns model matching rules, tokenizer configuration, usage unit, legacy prices, and conditional pricing tiers used for cost calculation.
Get the project associated with the connected project-scoped API key, returned in a data list with its organization, metadata, and retention configuration. No project ID is required. To get all projects using an organization-scoped API key, use the provider's organization API separately.
Retrieve a Langfuse text or chat prompt by name, optionally selecting a deployment label or exact version and resolving dependencies.
Retrieve one score configuration by ID in the authenticated Langfuse project. Returns its score data type, numeric bounds, categorical labels, description, timestamps, and archive status.
Export OpenTelemetry traces to Langfuse using OTLP HTTP/JSON with real-time v4 ingestion. Supply resourceSpans containing scopeSpans and spans. Acceptance is asynchronous. Inspect partialSuccess: HTTP 200 can include rejected spans or warnings; do not retry partially accepted batches. This action supports JSON, not binary protobuf or gzip.
List one page of items in a Langfuse annotation queue, optionally filtered by PENDING or COMPLETED status. Increment page while meta.page is less than meta.totalPages to retrieve subsequent pages.
List annotation queues configured for human evaluation in the connected Langfuse project.
List one page of comments in the authenticated Langfuse project. Optionally filter by object type, object ID, or author user ID. Filtering by object ID requires objectType. To retrieve another page, increment page while page is less than meta.totalPages.
List dataset items, optionally scoped to a dataset, source trace or observation, or historical dataset version.
List run items for a specific Langfuse dataset and run name.
List datasets available to the connected Langfuse project.
List one page of evaluation rules in the authenticated Langfuse project, newest creation first. Includes legacy trace and dataset rules for inspection and migration, with stored filters and evaluator assignments. Pass meta.cursor unchanged as cursor to retrieve the next page; an absent or null cursor means there are no more pages.
List one page of an evaluator's version history, newest first, for history and audit use cases. Returns each version's definition and creator metadata. Pass meta.cursor unchanged as cursor for the next page; an absent or null cursor means there are no more pages. For only the latest definition, use Get evaluator.
List one page of evaluators in the authenticated Langfuse project, newest creation first. Each evaluator includes its latest definition, version metadata and evaluation-rule assignments. Pass meta.cursor unchanged as cursor to retrieve the next page; an absent or null cursor means there are no more pages.
List Langfuse experiment items started on or after a required datetime, ordered by time descending. Export inputs, outputs, expected outputs, metadata, and optional item/trace scores by selecting field groups. Returns one page; pass meta.cursor with the same filters to fetch the next.
List Langfuse experiments active from a required start time, optionally specifying an end time and including metadata and directly attached scores.
List one page of LLM provider connections in the authenticated Langfuse project, including model configuration and masked secrets. To retrieve another page, increment page while page is less than meta.totalPages.
List Langfuse-managed and project-custom model pricing definitions used for usage and cost calculation.
Search Langfuse observations such as generations, spans, events, agents, and tool calls, selecting only the field groups needed.
List one page of prompt names, types, versions, labels, tags and latest matching configuration in the authenticated Langfuse project. Optionally filter by name, label, tag, update dates or advanced JSON conditions. Returns metadata rather than prompt content; use Get Prompt to retrieve a template. Increment page while meta.page is less than meta.totalPages to retrieve subsequent pages.
List one page of score configurations in the authenticated Langfuse project, including score types, numeric bounds, categorical labels, and archive status. To retrieve another page, increment page while page is less than meta.totalPages.
Search numeric, boolean, categorical, text, and correction scores with subject and annotation context.
Run an aggregate Langfuse metrics query over observations or numeric, boolean, or categorical scores.
Remove a user's assignment to a Langfuse annotation queue by queue ID and user ID. Returns whether the assignment was removed.
Remove an item from a Langfuse annotation queue using the queue ID and annotation queue item ID.
Submit feedback to Langfuse about its skills, MCP tools, CLI, docs, or public API. Only call after the user has explicitly approved the feedback content for submission; any supplied goal must also be user-approved. Do not include secrets, credentials, customer data, trace payloads, or unrelated use-case details. Not available in the HIPAA cloud region. Returns a feedback correlation ID.
Update a Langfuse annotation queue item's status to PENDING or COMPLETED using its queue ID and item ID. Returns the updated queue item.
Update a Langfuse evaluation rule by its stable ID. Supply only fields to change; explicit nulls are forwarded. Filters and evaluator assignments replace their complete lists. An empty assignment list disables the rule; enabling requires a nonempty resulting list and may incur model costs. Legacy trace and dataset rules only allow deactivation with enabled=false; their name, filters, sampling, and assignments cannot change.
Update a Langfuse evaluator by stable identifier. Supply name or description for a metadata-only change without creating a version. To replace a definition, include type and all required definition fields; definitions are replaced, not merged. The evaluator type cannot change. Null description clears it; null or omitted modelConfig selects the project default model. Definition replacement may reactivate evaluators paused for invalid or missing model configuration; other pause reasons require an explicit reactivation check.
Update an existing Langfuse media record with its upload timestamp and HTTP status, and optionally the upload error message and elapsed time. This records the upload result; it does not upload media content. Omitted optional fields are not sent; explicitly supplied null values are forwarded. Returns an empty response on success.
Update deployment labels for a specific Langfuse prompt version by name and version. Labels are unique across versions of a prompt; the latest label is reserved and managed by Langfuse. Returns the updated text or chat prompt version.
Update an existing Langfuse score configuration by config ID. Change its name, description, archive status, numerical bounds, or categorical label/value pairs. Omitted fields are not sent; explicitly supplied null values are forwarded. Boolean categories are autogenerated and cannot be passed. Returns the updated configuration.
Create or update a Langfuse LLM provider connection by provider name.
Create or replace a project-owned Langfuse model definition by ID for usage cost calculation and optional tokenization. Built-in models cannot be modified. This replaces the model definition; supply its desired configuration.
The Langfuse integration connects your Dench AI CRM directly to Langfuse, so agents can read and act on your Langfuse data as part of everyday work — answering questions in chat, keeping your CRM in sync, and running automations without anyone copying data between tools.
65 actions are available for agents to invoke on your behalf. Every call runs through Langfuse's own authorization, scoped to the account you connect.
Sign in to your Dench workspace and open Integrations.
Find Langfuse and click Connect — you'll authorize access through Langfuse's own sign-in flow. No API keys or code required.
Ask an agent to use Langfuse in chat, or call it from an automation.
Manage or disconnect the connection any time from workspace settings.
The Dench Langfuse integration connects your AI CRM to Langfuse, so AI agents can work with your Langfuse data as part of chats, automations, and CRM workflows. You connect your account once, and every agent in your workspace can use it — governed by your workspace permissions.
The Langfuse integration currently exposes 65 actions, including Add Annotation Queue Item, Assign Annotation Queue User, Check Health, Create annotation queue, Create comment, and Create Dataset. Agents invoke them on your behalf from chat or from automations.
No. You connect Langfuse from your Dench workspace using Langfuse's own sign-in and authorization flow — no API keys to copy, no glue code to maintain.
Connections are authorized through Langfuse's own authentication flow, and Dench stores only the authorization needed to act on your behalf. You can review and disconnect the Langfuse connection from your workspace settings at any time.