acquirium 0.1.1

Acquirium: A data-metadata management platform for water treatment systems

Acquirium

A Data-Metadata Framework for Water Treatment Plants

Acquirium is a framework for storing, managing, querying, and integrating data and metadata for water treatment systems. It combines knowledge graphs and time series data to support analysis, monitoring, and experimentation.

Installation

From PyPI:

pip install acquirium

Optional extras for specific drivers:

pip install "acquirium[mqtt]"       # MQTT ingestion driver
pip install "acquirium[xlsx]"       # Excel ingestion driver
pip install "acquirium[watertap]"   # WaterTAP simulation driver

Or with uv:

uv pip install acquirium

For development from a clone:

git clone https://github.com/DataDrivenCPS/acquirium.git
cd acquirium
python -m venv .venv && source .venv/bin/activate
pip install -e .
# or: uv sync

Quickstart

Acquirium ships a single CLI entry point. Start the server and any configured drivers with:

acquirium server --config acquirium.toml

A sample acquirium.toml is included at the repository root. Key sections:

• [server] — bind host/port, choice of timeseries backend (DuckDB or TimescaleDB), data directory.
• [driver] — connection defaults applied to all drivers (server URL, port, tick interval).
• [[drivers]] — drivers to start alongside the server.

By default the server stores data on local disk — an embedded Oxigraph RDF store and a single DuckDB file under data_dir. No external services are required for a fresh install. For multi-worker or production deployments, switch the config to timeseries_backend = "timescale" and point pg_dsn at a Postgres + TimescaleDB instance.

Override the bind host/port from the CLI if needed:

acquirium server --config acquirium.toml --host 127.0.0.1 --port 8000
acquirium server --config acquirium.toml --reload          # uvicorn auto-reload

Driver-only mode

To run only [[drivers]] against a remote Acquirium server (no FastAPI on this host), set:

[server]
enabled = false

and configure [driver].server_url / server_port to point at the remote instance. Then:

acquirium server --config acquirium.toml

When enabled = false, the server subcommand starts only the drivers.

Docker stack (optional)

A compose.yaml is provided for an all-in-one local stack (Acquirium + TimescaleDB + Grafana):

make up                              # start
make up ACQUIRIUM_RECREATE=true      # wipe data + start
make down                            # stop

By default each Docker run resets the system. To preserve data across runs, set ACQUIRIUM_RECREATE=false in compose.yaml.

WaterTAP integration

The watertap extra installs the Python packages needed for the built-in WaterTAP driver:

pip install "acquirium[watertap]"
acquirium server --config acquirium.toml   # with a [[drivers]] entry for WaterTAP

Some WaterTAP setups also require native extensions that are not installed by the extra:

pyomo download-extensions
python -m pip install setuptools && pyomo build-extensions
idaes get-extensions

For a full demo (WaterTAP + streaming simulator + API examples):

make watertap-up
uv run scripts/api_example.py
# or open notebooks/watertap-single-pump.ipynb
make watertap-down

Logging

Acquirium supports user logs attached to entities in the system. See scripts/logging_example.py:

acquirium server --config acquirium.toml &
python scripts/logging_example.py

Text Matcher

Acquirium uses a text matcher to map natural-language input to ontology URIs (classes, predicates, units, quantity kinds). The match algorithm uses semantic embedding similarity powered by FastEmbed (default model: BAAI/bge-small-en-v1.5). Each ontology concept is represented by one or more surface strings, embedded and stored in an in-memory vector index. At query time the input phrase is embedded and compared against the index using cosine similarity.

There are two separate matchers, each with its own index:

1. Graph matcher — indexes classes and predicates from user-inserted RDF graphs. Surface strings are derived from rdfs:label values and CamelCase/underscore-split local names.
2. QUDT matcher — indexes units and quantity kinds from the QUDT ontology (fetched over HTTP with local fallback). Surface strings include rdfs:label, skos:prefLabel, skos:altLabel, symbols, UCUM codes, and split local names.

Both indexes are cached to disk and updated incrementally when graphs change. Results can be filtered by kind (class, predicate, unit, quantity_kind) and are ranked by cosine similarity, deduplicated to the highest-scoring surface per URI. See scripts/text_matcher_example.py for usage.

Tests

pytest tests/unit            # unit tests only
make test                    # full suite (Docker required)

Status

Acquirium is under active development. Planned work is tracked in improvements.md. Bug reports and feature requests are welcome — please open an issue.