STAMM
What it is
STAMM is a free, open-source MLOps framework for deploying, monitoring, and maintaining machine-learning soft sensors in industrial processes. It watches live process data for concept drift and operating-regime changes, while its dashboard displays measurements, soft-sensor outputs, drift signals, historical context, and human-entered labels. A model registry records versions, configurations, artifacts, validation results, and metadata. Python and R soft sensors are served through REST inference. The event-driven orchestrator calls the registry over REST and links each prediction to the data snapshot that produced it. STAMM’s drift-detector package offers 10 detectors through one interface and can also operate outside the framework. Data can arrive through equipment REST hooks, MQTT applications, or frameworks such as LEAF. InfluxDB is the reference time-series store; a PostgreSQL adapter is described as in progress. Docker Compose deployment supports Linux, macOS, or Windows with Docker 24 or later, with at least 8 GB RAM recommended. The project uses Apache License 2.0 and supports self-hosted deployment. It identifies process modelers, ML engineers, operators, and project or production managers as users.
For and against
For
- Monitors concept drift and regime changes
- Tracks model versions and validation metadata
- Supports Python and R soft sensors
- Offers 10 drift detectors
- Self-hosted Docker Compose deployment
Against
- Does not prescribe how to rebuild models
- PostgreSQL adapter is still in progress
- Requires Docker 24 or later
- At least 8 GB RAM recommended
The verdict
STAMM provides an open framework for running soft sensors and identifying when their behavior changes. It supplies monitoring and maintenance signals, but teams must determine how models are rebuilt.
Who it suits
It suits industrial teams operating soft sensors who need model registries, drift monitoring, dashboards, and REST-based deployment. The intended audience includes modelers, ML engineers, operators, and process managers.
On the record
- Audience
- The maker identifies process modelers, ML engineers, operators, and project or production managers as intended users.stamm.inrae.fr · read 4 Oct 2026
- Dashboard
- The dashboard shows live measurements, soft-sensor outputs, drift signals, and historical context, and supports human-in-the-loop labelling.stamm.inrae.fr · read 7 Oct 2026
- Data storage
- The reference time-series store uses InfluxDB; a PostgreSQL adapter is described as in progress.stamm.inrae.fr · read 7 Oct 2026
- Demo
- The reference demo applies STAMM to an industrial-scale penicillin fermentation simulator and includes a curated dataset, Node-RED bioreactor, and working model registry.stamm.inrae.fr · read 7 Oct 2026
- Deployment
- It integrates existing soft sensors into live systems alongside physical instruments without requiring rewrites.stamm.inrae.fr · read 7 Oct 2026
- Deployment requirements
- The documented Docker Compose installation supports Linux, macOS, or Windows with Docker 24 or later; 8 GB RAM or more is recommended.github.com · read 4 Oct 2026
- Drift detection
- It detects regime shifts and concept drift, including through a Python package with 10 detectors behind a single API.stamm.inrae.fr · read 7 Oct 2026
- Drift detectors
- The drift detector package provides 10 detectors through a common interface and can be used inside or outside STAMM.stamm.inrae.fr · read 4 Oct 2026
- Installation
- The reference deployment uses Docker Compose and lists Linux, macOS, or Windows with Docker 24 or later and Docker Compose as requirements.github.com · read 7 Oct 2026
- Integrations
- The workflow orchestrator calls the model registry over REST, and the project describes data ingestion through equipment REST hooks, MQTT applications, or frameworks such as LEAF.github.com · read 7 Oct 2026
- Intended users
- The site identifies process modelers, ML engineers, operators, and project leaders or process and production managers as intended users.stamm.inrae.fr · read 7 Oct 2026
- Language support
- The maker describes support for Python and R soft sensors, served through REST inference.stamm.inrae.fr · read 4 Oct 2026
- License
- STAMM is released under the Apache License 2.0.github.com · read 7 Oct 2026
- Model registry
- The registry tracks model versions, configuration, artifacts, validation results, and metadata; it supports Python and R models and REST inference.stamm.inrae.fr · read 7 Oct 2026
- Notable limitation
- STAMM surfaces when maintenance may be needed but does not prescribe how the model should be rebuilt.stamm.inrae.fr · read 7 Oct 2026
- Purpose
- STAMM is an open-source MLOps framework for deploying, monitoring, and maintaining machine-learning soft sensors in industrial processes.stamm.inrae.fr · read 7 Oct 2026
- Real-time monitoring
- It monitors live process data and detects concept drift and changes in operating regimes.stamm.inrae.fr · read 4 Oct 2026
- Security and compliance
- The project describes a FAIR-aligned YAML metadata schema for soft-sensor models that can link to FAIRDOM-SEEK catalogues such as the IBISBA Knowledge Hub.github.com · read 7 Oct 2026
