Langfuse logo
AI integrationUpdated

Langfuse metrics as an Apple TV dashboard for your AI team

Langfuse is where teams building LLM features trace every model call, which makes it the best record of what an AI product costs and how fast it answers. Dance Party reads the project's aggregate metrics and puts cost by model, token volume, and p95 latency on the office display as an Apple TV dashboard, without ever reading a prompt or a response.

  • Read-only
  • API key connection
  • One connection per workspace
Get StartedHow to connect Langfuse
LLM app, last 7 daysLive
Total cost
$1,926
+11%
Observations
412,380
+8%
Total tokens
286M
+10%
p95 latency
3.4s
-0.6s

Popular Langfuse dashboards

What teams running Langfuse put on the screen, and what you would ask DJ Agent to get it.

LLM cost by model

For AI engineers and product leads

Langfuse prices every generation it traces, so it can show which model is actually eating the budget across providers. That list is where a model swap or a smaller prompt pays off first.

  • Total cost per model over the last 7 or 30 days
  • Input and output tokens per model
  • Observation count per model

Ask DJ Agent“Rank our Langfuse models by total cost over the last 7 days, with token counts for each.”

Langfuse cost by model, 7 daysLive
1claude-sonnet-4-5
$842
2gpt-5
$611
3gpt-5-mini
$268
4text-embedding-3-small
$121
5claude-haiku-4-5
$84

Daily LLM spend in production

For engineering leads and finance

A prompt change that doubles output tokens shows up as a step in the daily cost line the next morning. Filtering to the production environment keeps staging experiments out of the number.

  • Total cost per day for up to 90 days
  • Filtered to one Langfuse environment, such as production
  • Total tokens per day next to cost

Ask DJ Agent“Chart daily Langfuse total cost for the production environment over the last 30 days.”

Production LLM cost per dayLive
$7,840+16% vs prior 30 days

Model latency watch

For on-call engineers for AI features

Users feel p95 latency, not the average. One tile per model, amber or red past a threshold you choose, catches a slow provider before support tickets do.

  • p95 latency per model in milliseconds
  • Average latency per model
  • Observation count behind each figure

Ask DJ Agent“Show one status tile per Langfuse model with its p95 latency over the last 24 hours, amber above 5 seconds and red above 10.”

p95 latency by model, 24 hoursLive
claude-sonnet-4-5
p95 4.1s
gpt-5
p95 6.8s
gpt-5-mini
p95 1.9s
claude-haiku-4-5
p95 1.2s

Traced calls per hour

For the team running an LLM feature

Observation volume by the hour shows the shape of a normal day. A flat hour in the middle of the afternoon usually means the feature is down or the tracing broke, and both are worth knowing fast.

  • Observations per hour for the last 8 to 24 hours
  • Tokens per hour
  • Cost per hour

Ask DJ Agent“Show Langfuse observations per hour for the last 8 hours as bars, with the total as the headline.”

Langfuse observations per hourLive
21,640observations, last 8 hours
8a
9a
10a
11a
12p
1p
2p
3p

What Dance Party reads from Langfuse

Dance Party reads one Langfuse Cloud project through the Metrics API, which returns aggregates over observations. It reads counts, tokens, cost, and latency. It never reads prompts, outputs, individual traces, or user data.

Metrics API

Langfuse API docs ↗
  • Usage summary
    Observation count, input, output, and total tokens, input, output, and total cost in USD, and average and p95 latency, for a window of up to 90 days.
  • Usage over time
    The same metrics by hour, day, week, or month.
  • Usage by model
    The same metrics grouped by model name, sorted by cost.
  • Filters
    Narrow any read to one environment, such as production, or one model.

How to connect Langfuse

Before you start

  • An owner or admin of your Dance Party workspace makes the Langfuse connection.
  • A Langfuse Cloud project in the EU, US, Japan, or US HIPAA region. Self-hosted Langfuse is not supported
  • A project API key pair, the public key and the secret key, created by a project member allowed to manage API keys
  1. 1
    Click Connect Langfuse on the Integrations page in Dance Party. The pop-up links to Langfuse Cloud in Europe, the US, Japan, and US HIPAA, so open the one your project lives in.Open Langfuse Cloud (Europe) ↗
  2. 2
    Open your project, then Settings, then API Keys, and create a new key pair.
  3. 3
    Copy the two values into Project public key and Project secret key.
  4. 4
    Under Data region, choose the region your project lives in, then click Save. Dance Party checks it can read aggregate metrics before the card shows Connected.
The Connect Langfuse pop-up with region links, three numbered steps, the Project public key and Project secret key fields, and a Data region choice of Europe, United States, Japan, or US HIPAA
Captured from the Dance Party app, September 30, 2026.

One Langfuse connection per workspace

A workspace connects one Langfuse project, because Langfuse keys belong to a single project in a single region. Teams that trace each product in its own Langfuse project use a separate Dance Party workspace for each.

Langfuse project keys have no read-only option and can also reach other Langfuse features. Dance Party encrypts the pair and only calls the aggregate Metrics API, never trace, prompt, or dataset endpoints. Delete the key pair in project Settings, API Keys to revoke access.

From Langfuse to your office display

Once Langfuse shows Connected, the rest happens in plain English. No code, no drag-and-drop.

  1. 1. Describe it. Tell DJ Agent what you want to see.

    New dashboard+ Widget
    Describe a widget in DJ Agent, or click + Widget
    DJ Agent
    DATA SOURCES
    Langfuse logoLangfuseConnected
    Rank our Langfuse models by total cost over the last 7 days, with token counts for each.Send

  2. 2. It gets built. The widget lands on your board with real Langfuse data.

    New dashboard+ Widget
    Langfuse logoLangfuse cost by model, 7 daysLive
    1claude-sonnet-4-5
    $842
    2gpt-5
    $611
    3gpt-5-mini
    $268
    4text-embedding-3-small
    $121
    5claude-haiku-4-5
    $84
    DJ Agent
    DATA SOURCES
    Langfuse logoLangfuseConnected
    Rank our Langfuse models by total cost over the last 7 days, with token counts for each.Send

  3. 3. Put it on a screen. Pair an Apple TV with the six-character code it shows, then assign the dashboard.

    The Pair your Apple TV card in Dance Party, with six boxes for the code the Apple TV shows

Langfuse dashboard questions

Can I put Langfuse metrics on an office display?

Yes. Connect the project with its key pair, describe the board to DJ Agent, and it runs on any screen with the Dance Party Apple TV app. Boards grouped by model should refresh hourly or slower.

Does Dance Party read our prompts or model outputs?

No. It only calls Langfuse's Metrics API, which returns counts, tokens, cost, and latency. Prompts, responses, traces, and user IDs are never requested.

Does it work with self-hosted Langfuse?

No. Dance Party connects to Langfuse Cloud in the EU, US, Japan, or US HIPAA regions only.

Where does the cost number come from?

From Langfuse. It is the total cost Langfuse calculates for traced generations, in USD, so calls you do not trace or models without a price in Langfuse will not be counted.

Can I split production from staging?

Yes. Every read can be filtered to one Langfuse environment, so a board can show production only.

Can I connect more than one Langfuse project?

One project per Dance Party workspace. Use a separate workspace for each additional project.

More integrations

OpenAI UsageAnthropic UsagexAI UsageCursorDatadog

Put Langfuse on every office display

Connect Langfuse, describe the dashboard, and it runs on an Apple TV, refreshing on its own.

Get Started
Learn more: How to put a dashboard on a TV · AI dashboards vs prebuilt dashboards