Data inputs
- OCEL 2.0
- XES
- CSV & Parquet
Bring any event or object-centric data.
Welcome to Promenade
Powering your
Process Intelligence
Based on decades of research. Built with the latest web technologies. Designed to be open, scalable, and accessible for everyone who wants to turn data into insights.
- Research-driven
Built on the foundations of process mining research
- Modern & Open
State-of-the-art web technologies and open standards
- Scalable
From classroom examples to real-world data
- For Everyone
Researchers, students and practitioners
- No installation
- Local-first
- XES, CSV & OCEL 2.0
- Tab → container → cloud
The whole pipeline
One workspace. Every layer of process intelligence.
Open compute
- Rust · WebAssembly
- Python · Pyodide
- SQL · relational
- TypeScript views
Open languages, high performance, local by default.
Interactive analysis
- Process graphs
- Variants
- Conformance
- Object-centric
Explore, validate, and improve with confidence.
- €0forever
- MB → TBone workspace, any size
- 35installable plugins
- 4plugin runtimes
See it run
A discovered model, replayed against the log.
This is the workspace layout — artifacts on the left, a tabbed analysis pane in the center, and properties on the right.
Artifactsorder_to_cash.xesDirectly-Follows GraphInductive modelAlignment replayVariantsorder_management.ocelObject-Centric Petri NetMetro mapLocal storage 18.4 MB
Alignment replayInductive modelVariantsDotted chartwasm · 840 ms
InspectStatistics
Check credit
- Type
- Task
- Occurrences
- 2,674
- Coverage
- 95.1%
- Median duration
- 1h 42m
- Mean duration
- 2h 36m
- Log moves
- 118
- Classifier
- Activity name
- Noise threshold
- 0.20
- Aggregation
- Count
import xes└ inductive.discover└ alignment.replay
Scale
A tutorial log and a terabyte warehouse, in the same window.
Start on a five-megabyte sample in a browser tab. Point the same workspace at a container on your laptop when the log reaches gigabytes, or at an engine on your own infrastructure when it reaches terabytes. Same artifact tree, same plugins, same clicks — you change one dropdown.
- In the tab
Browser engine
DuckDB-wasm and WebAssembly plugins. Nothing is uploaded, because there is no server.
- On your machine
Local Docker engine
One container, one command. Native plugin runners, real threads, spill-to-disk.
- On your infrastructure
On-prem or cloud engine
The identical image, next to the warehouse. Your network, your keys, your audit trail.
Same discovery run, three engines — more data, and more throughput to match.
Browser · WASM5 MB log0.8 s15 K events/s
Local Docker · 24 cores4 GB log6 s6 M events/s
On-prem engine2 TB log41 s115 M events/s
- The same artifact tree
- The same plugins, unchanged
- The same actions and views
- One provenance graph across all of them
Data locality follows execution.
Move an artifact to an engine and the heavy data stays there. Only metadata, previews and the paged results you actually look at cross the network — so a terabyte log never has to become a terabyte download. Artifacts on an engine stay in the same tree, with a badge rather than a separate list.
How engines workEnginesBrowser (WASM)readyLocal Docker · 24 coresreadywarehouse-eu · on-premreadyAdd an engine
One workspace, four ways in
Whether this is your first event log or your fiftieth paper
Learn
I'm new to process mining
A demo log is already loaded. Fifteen minutes from a table of events to a model you can explain out loud.
Start the beginner trackAnalyze
I run process work in a company
One export, one laptop, no procurement cycle. Rework, waiting time and conformance on your own data this week.
Open the practitioner trackResearch
I work on process mining itself
Object-centric techniques, exact alignments and runs a reviewer can retrace. Publish your technique as a plugin others install in one click.
Open the expert trackDevelop
I'm building with Promenade
Ship your algorithm as a plugin in Rust, Python, SQL or TypeScript. One manifest, one data door, provenance for free.
Plugin quickstart
Capabilities
Built for real event data, not just demos
Every capability below is in the core open-source application today. Nothing here is a roadmap item or a paid add-on.
Runs in the browser
A streaming parser feeds DuckDB-wasm; artifacts persist as compressed Parquet in the browser's own file system and survive a reload.
XES and CSV event data
Import the formats you already have, map columns to case, activity and timestamp, and get a first model in the same session.
OCEL 2.0, first class
Object-centric Petri nets, OC-DFGs, TOTeM, an object-centric Inductive Miner, replay and a metro-map view — not a bolt-on mode.
Conformance, exactly
Alignment-based replay computes the cheapest alignment per trace variant. Exact, computed locally, and animated over the model.
Extensible to the core
Algorithms, views, importers and exporters are all plugins — ours included. Yours installs exactly the same way.
Grows with the data
When a log outgrows the tab, point the workspace at a Docker engine on your machine or one on your own infrastructure.
Extensibility
Every algorithm is a plugin. Including ours.
Nothing is welded into the core: miners, conformance checkers, renderers, importers and exporters all arrive through the same plugin interface, and everything shipped with Promenade goes through it too. Write one, and it sits beside ours in the action menu.
Write it in your language
Rust compiled to WebAssembly, Python through Pyodide, SQL against the relational API, or TypeScript for views. One manifest format covers all four.
Write once, run at any size
The same bundle executes in the browser tab, in a Docker engine on your laptop and on an engine beside your warehouse. You do not port anything.
One data door, enforced
A plugin reads the log through a read-only SQL interface and renders into a sandboxed iframe. It cannot quietly reach past either.
- Rust · WebAssembly0.1.2Inductive MinerLeemans' Inductive Miner (IM and IMf) in Rust: recursively cuts the directly-follows graph into a block-structured process tree.
- Rust · WebAssembly+10.2.1Alignment-based ConformanceReplays a log against a Petri net: for every distinct trace variant, the cheapest alignment of sync/log/model moves, plus fitness and escaping-edges precision…
- Python · Pyodide+10.2.3Inductive Miner (pm4py)Discovers a process tree with pm4py's inductive miner, and converts it to an accepting Petri net.
- Rust · WebAssembly+10.17.5Metro MapRenders an object-centric process as a transit-schematic 'metro map': activities become stations shared across object types, object types become color-coded lines…
- Rust · WebAssembly+10.3.11Inductive Visual MinerBrowser-native alignment-backed process-model replay: aligns an Accepting Petri Net with a Traditional Event Log in Rust/WASM, once per distinct trace variant, then…
- SQL · relational+20.1.10Object-Centric Inductive MinerNiklas van Detten's Object-Centric Inductive Miner in Rust.
- View · TypeScript0.1.9Performance SpectrumFine-grained performance analysis after Denisov, Fahland, and van der Aalst: each observed directly-follows flow is rendered over calendar time and colored by…
- + 28 more
Alpha Miner (Classic) · BPMN 2.0 · Cardinality Impact Explorer · Cases & Variants (OCEL) · Directly-Follows Graph (Rust) · Dotted Chart · Event Log Transformations…
Develop
Your algorithm, installable.
Declare the artifact types your action consumes and produces; the host handles data access, color, selection and provenance. Ship the bundle next to the paper and a reviewer runs it in one click.
- Rust → WebAssembly, Python → Pyodide, SQL → the relational API, or TypeScript for views
- One data door — host.sql() — and nothing reaches past it
- Views render in sandboxed iframes with their own CSP
{
"id": "org.acme.my-miner",
"runtime": "wasm",
"actions": [{
"id": "discover",
"inputs": { "log": "EventLog" },
"outputs": [ "AcceptingPetriNet" ]
}]
}Research
Built so a result can be retraced.
Every artifact records the action, the parameters and the runtime version that produced it. A workspace exports as a bundle — log, models and provenance together — so the figure in a paper can be reopened rather than described.
Reproducible runs
Provenance is a DAG, not a filename convention. Re-run a chain with one parameter changed and keep both results side by side.
Standard formats
XES, CSV, OCEL 2.0 (JSON and SQLite) and PNML in and out — no proprietary container in the middle.
Extensible for new techniques
A new discovery algorithm or conformance technique is a plugin, not a fork — against the same interface the built-ins use.
Citable
Canonical citation metadata lives with the source, so a paper can point at a version rather than at a website.
Open source
Free software, not a free tier.
There is no upgrade path, because there is nothing to upgrade to. Promenade is open source — fork it, cite it, teach with it, or ship a plugin that becomes part of it.
No account, no lock-in
Open the application and start working. Your data stays in your browser — or on your own engine — unless you export it.
Source available today
The application, the plugin runtime and every algorithm implementation are on GitHub under an open-source license.
Built with the community
Algorithms arrive as plugins from the people who published them. Discussion, issues and roadmap are all in the open.
Open a tab. Drop in a log.
Free, open source, and running before you finish reading this sentence. When the log outgrows the tab, the same workspace points at an engine — nothing about your analysis changes.