Orpheon · a visual data platform by NexPatch

Airflow, Superset and MLflow. One canvas.

Draw the whole pipeline, DAG included: sources, transforms, schedules, dashboards and models. No DAG files, no second BI tool, no separate model registry. Probe any step to see its rows.

Book a demo What it replaces

Already a customer? Sign in · Documentation (access on request)

example · BTC + volatility, hourly7 nodes · 7 edgesDAG · LIVE
TRIGGER
Hourly
cron · :05
SOURCE
Binance
BTCUSDT 1h
SOURCE
Options
Deribit DVOL
TRANSFORM
Join
python · 2 inputs
SINK
Gold hourly
parquet
CHART
24 h range
MODEL
Bands
pinned version
PROBE · Joinexample rows
time (UTC)BTC.klines.closeVOL.btc.dvol
07:0074,312.048.6
08:0074,086.049.1
09:0074,131.748.9
3 of 24 rows · both inputs present · nothing written yet
9
source adapters, from market data to live SSE and WebSocket streams
1 canvas
for the DAG, the dashboards and the models, under one login
75
MCP tools, so an AI client can build and run pipelines
your cluster
runs on Kubernetes with a Helm chart; your data stays with you
What it replaces

Three tools, three logins, one canvas.

A typical data team runs an orchestrator for the DAG, a BI tool for the dashboards and a tracking server for the models, and writes the glue between them. In Orpheon they are one graph you draw, under one login, one set of roles and one run history.

Instead ofApache Airflow

The DAG is the canvas

  • Draw the graph; there is no DAG file to keep in sync
  • Cron and manual triggers start a run
  • Retries with backoff, then a dead-letter state
  • SLA targets, overdue warnings, run history per node; rerun or cancel
Instead ofApache Superset

Dashboards on the same gold

  • Charts and maps read a sink or a chart node directly
  • Time series, candlesticks and market modules such as order-book depth
  • Share with a person, an organisation or everyone in it
  • Read-only public links with a password and an expiry date
Instead ofMLflow

Models as nodes

  • Model families with versions; production pins one
  • Training runs from the API or an AI client, with their status
  • The served version writes forecasts into an output sink, like any data
  • Charts and dashboards read the forecast like any other column

When to keep them. Orpheon replaces these tools for pipelines that turn external data into datasets, dashboards and forecasts. If your DAGs mostly orchestrate other systems, such as Spark jobs, dbt runs or file transfers between servers, keep your orchestrator next to it.

How it works

Three steps, one canvas.

The Orpheon mark is three steps: source, transform, sink. Every pipeline is built from them, wired on a canvas you can read at a glance, and each step can be probed before anything is written.

01 · Connect

Bring the data in

  • Binance and Polygon market data
  • NexPatch node feeds: Bitcoin chain, order book, macro events, options
  • Live SSE and WebSocket streams
  • Any HTTP API, in a few lines of JavaScript
02 · Shape

See every step

  • SQL or Python transforms
  • Joins with named inputs: market, options
  • Probe any node and read its rows
  • Templates for dedup, imputation, features and labels
03 · Serve

Ship the result

  • Gold as Parquet, PostgreSQL or S3
  • Dashboards with charts and maps
  • Read-only public links to a pipeline or a dashboard
  • Model families that write their forecasts back as data
Use cases

Built for decisions on live external data.

Three pipelines that run on Orpheon in production, on a schedule. Each answers one question a team asks every week, from public data sources. Open one: you see its canvas, its code and its latest rows, read-only.

Energy

When should we buy power, and when should we shift flexible load?

Data
Energy-Charts, DE-LU day-ahead price, load, wind and solar, 15-min
Out
Per quarter-hour, the price range likely to clear and the risk it goes negative
Open the live pipeline →
Financial services

How big a position, how much margin, how far a stop?

Data
Binance BTCUSDT hourly, joined with Deribit implied volatility
Out
The 24-hour range Bitcoin is likely to trade in, next to the cost of carry
Open the live pipeline →
Manufacturing

What will our inputs cost next quarter, and how sure is that?

Data
Copper, oil, gas and EUR/USD from Polygon, power from Energy-Charts
Out
An input-cost index for an example bill of materials, with a quarter-ahead range
Open the live pipeline →
Measured, not claimed

Every range keeps its own score.

A forecast pipeline on Orpheon stores, for every row, the range it issued earlier and whether the outcome landed inside. Coverage is a column you can query, not a sentence on a slide. We forecast ranges, not direction: in our own pre-registered studies, direction measured at chance.

BTC · 80 % range issued 24 h earlier vs the priceloading live data…
80 % range, issued 24 h earlierBTC priceoutside the rangenext 24 h
—

of the last 30 days' hours fell inside the BTC risk desk's 80 % range.

read live from the pipeline
—

of the last 28 days' quarter-hours fell inside the energy pipeline's 80 % price range.

read live from the pipeline
Works with your AI

Ask in plain words. Get a pipeline you can read.

  • Claude, ChatGPT and Claude Code connect by signing in, with no key to copy.
  • 75 MCP tools build, probe, run and publish, under the same roles as the app.
  • Everything lands on the canvas, so a person can check every step.
Build a pipeline with BTC hourly candles from Binance and the Deribit volatility index, joined on one grid.
pipeline_create "BTC + DVOL hourly"
node_add source · binance · BTCUSDT 1h
node_add source · nxp_options · VOL.btc.dvol
node_add transform · python, inputs market + options
node_add sink · parquet
probe join → 24 rows, 0 gaps
Done. Five nodes on the canvas. The join probe shows 24 hourly rows with both price and DVOL, no gaps. Open it to review before the first run.
Your infrastructure

Runs where your data lives.

Orpheon deploys on your own Kubernetes cluster, next to the data it reads. This page has no trackers either.

Deploy

A Helm chart, images from your own registry, staging before production.

Access

Admin, editor and viewer roles; per-dashboard sharing; organisations.

Secrets

An encrypted credential vault. Keys never sit in a pipeline's config.

Integrate

A REST API with an OpenAPI spec, MCP for AI clients, public read-only links.

Book a demo

See your data on the canvas.

Tell us which data you work with, and we show you Orpheon on that kind of data.

Or write to info@nexpatch.ai

We use these details only to answer your request. See the privacy policy.