LaminDB: Data management for traceable, multimodal AI

LaminDB is an open-source data management tool that makes it easy to query, trace and govern datasets across diverse storage formats and locations. Like git, LaminDB is a distributed system that runs anywhere and captures all relevant context about your work. This includes the data flow through models and analyses, the entities and notes defining your work, and the features & schemas of datasets. It takes a few seconds to install LaminDB and create a database on your laptop.

How?

• lineage → trace results across agent sessions, notebooks, scripts & workflows
• lakehouse → manage datasets in any format (parquet, zarr, ...) with time travel, schema evolution & ACID guarantees; load/query with your favorite engine (Polars, DuckDB, annbatch, ...)
• LIMS & ELN → unified schema-based records management with support for ontologies & notes
• FAIR datasets → validate & annotate files, DataFrame, AnnData, SpatialData, …
• governance → manage changes via branching & versioning

Architecture?

• zero lock-in → open source, using open standards (metadata in SQLite/Postgres, data in parquet, zarr, etc.)
• scalable → hit storage & database directly through your pydata or R stack, no REST API involved
• simple → pip install lamindb or install.packages('laminr') · no Docker required, no separate backend
• unified access → local, S3, GCP, etc. · Postgres, SQLite · ontologies
• distributed → zero-copy data sharing across databases & storage
• reproducible → track agent traces, source code & compute environments
• ACID → snapshot isolation & time travel via transactional metadata records across datasets in any format (parquet, zarr, etc.)
• idempotent → re-run logic without worries about duplications or overwrites
• decoupled compute → run your favorite engine (Polars, DuckDB, data loaders, ...) with all its benefits
• integrations → bio ontologies, git, nextflow, vitessce, redun, and more
• extensible → create custom plug-ins based on the Django ORM, the basis for LaminDB's registries

Read more: docs.lamin.ai/architecture.

UI, permissions, audit logs? Lamin is a collaboration hub built on LaminDB similar to how GitHub is built on git.

Quickstart

To install the Python package with recommended dependencies, use:

pip install lamindb

pip install lamindb-core

Agent? See .agents/, docs/ and llms.txt.

Query databases

You can browse public databases at lamin.ai/explore. To access laminlabs/cellxgene, run:

import lamindb as ln

db = ln.DB("laminlabs/cellxgene")  # a database object for queries
df = db.Artifact.to_dataframe()    # a dataframe listing datasets & models

To get a specific dataset, run:

artifact = db.Artifact.get("BnMwC3KZz0BuKftR")  # a metadata object for a dataset
artifact.describe()                             # describe the context of the dataset

See the output.

Access the content of the dataset via:

local_path = artifact.cache()  # return a local path from a cache
adata = artifact.load()        # load object into memory
accessor = artifact.open()     # return a streaming accessor

For broader queries of cellxgene, see docs.lamin.ai/cellxgene.

Create artifacts

You can create a hosted database at lamin.ai and invite collaborators. To connect to an existing database, run:

lamin login
lamin connect account/name  # tip: add flag `--here` to scope to current directory

lamin init --modules bionty

On the terminal and in a Python session, lamindb will now auto-connect.

To save a file or folder via the API:

import lamindb as ln
# → connected lamindb: account/instance

open("sample.fasta", "w").write(">seq1\nACGT\n")        # create dataset
ln.Artifact("sample.fasta", key="sample.fasta").save()  # save dataset

To save a file or folder via the CLI, run:

lamin save sample.fasta --key sample.fasta

To load an artifact via the CLI into a local cache, run:

lamin load --key sample.fasta

Read more about the CLI: docs.lamin.ai/cli.

Trace data, code & agents

To track agent sessions, install the library skill: uvx library-skills --skill lamindb.

To track source code, inputs, outputs, logs, and environment of a script, use ln.track() and ln.finish():

import lamindb as ln
# → connected lamindb: account/instance

ln.track()                                              # track code execution
open("sample.fasta", "w").write(">seq1\nACGT\n")        # create dataset
ln.Artifact("sample.fasta", key="sample.fasta").save()  # save dataset
ln.finish()                                             # mark run as finished

If running this snippet as a script (python create_fasta.py), you'll get the following data lineage:

artifact = ln.Artifact.get(key="sample.fasta")  # get artifact by key
artifact.describe()      # context of the artifact
artifact.view_lineage()  # fine-grained lineage

Watch a mini video: youtu.be/yK3ODFZLL1A

Access run & transform.

run = artifact.run              # get the run object
transform = artifact.transform  # get the transform object
run.describe()                  # context of the run

transform.describe()  # context of the transform

ln.track(project="My project", plan="./plans/curate-dataset-x.md")

# create a project with the CLI
lamin create project "My project"

# save an agent plan with the CLI
lamin save /path/to/.cursor/plans/curate-dataset-x.plan.md
lamin save /path/to/.claude/plans/curate-dataset-x.md

ln.Project(name="My project").save()  # create a project in Python

You can track workflows by decorating functions:

import lamindb as ln

@ln.flow()
def create_fasta(fasta_file: str = "sample.fasta"):
    open(fasta_file, "w").write(">seq1\nACGT\n")    # create dataset
    ln.Artifact(fasta_file, key=fasta_file).save()  # save dataset

if __name__ == "__main__":
    create_fasta()

Beyond what you get for scripts & notebooks, this automatically tracks function & CLI params and integrates well with established Python workflow managers: docs.lamin.ai/track. To integrate advanced bioinformatics pipeline managers like Nextflow, see docs.lamin.ai/pipelines.

A richer example.

Here is an automatically generated re-construction of the project of Schmidt et al. (Science, 2022):

A phenotypic CRISPRa screening result is integrated with scRNA-seq data. Here is the result of the screen input:

You can explore it here on Lamin or here on GitHub.

Annotate & query artifacts

You can label an artifact by running:

my_label = ln.ULabel(name="My label").save()   # a universal label
my_project = ln.Project(name="My project").save()  # a project label
artifact.ulabels.add(my_label)
artifact.projects.add(my_project)

Query for it:

ln.Artifact.filter(ulabels=my_label, projects=my_project).to_dataframe()

You can also query by automatically generated metadata:

ln.Artifact.filter(run=run).to_dataframe()              # by creating run
ln.Artifact.filter(transform=transform).to_dataframe()  # by creating transform
ln.Artifact.filter(size__gt=1e6).to_dataframe()         # size greater than 1MB

If you want to include more information into the resulting dataframe, pass include.

ln.Artifact.to_dataframe(include=["created_by__name", "storage__root"])  # include fields from related registries

Let's define some features:

from datetime import date

gc_content = ln.Feature(name="gc_content", dtype=float).save()
experiment_note = ln.Feature(name="experiment_note", dtype=str).save()
experiment_date = ln.Feature(name="experiment_date", dtype=date, coerce=True).save()  # accept date strings

The most basic thing you can do with features is annotating artifacts, records, or runs:

artifact.features.set_values({
    gc_content: 0.55,
    experiment_note: "Looks great",
    experiment_date: "2025-10-24",
})

# query
ln.Artifact.filter(experiment_date == "2025-10-24").to_dataframe(include="features")  # query all artifacts annotated with `experiment_date`

Read more: docs.lamin.ai/organize.

Create records & notes

You can create records for entities underlying your experiments (samples, perturbations, instruments, etc.):

ln.Record(name="Sample 1", features={gc_content: 0.5}).save()

You can create record types:

# create an Experiments record type
experiments = ln.Record(name="Experiments", is_type=True).save()

# create a data record of that type
experiment1 = ln.Record(name="Experiment 1", type=experiments).save()

# create a feature that links experiments (a relationship)
experiment = ln.Feature(name="experiment", dtype=experiments).save()

# create a sample record
ln.Record(name="Sample 2", features={gc_content: 0.5, experiment: experiment1}).save()

# export all experiments
experiments.to_dataframe()

In your development directory, you can save markdown notes as records:

lamin save <topic>/<my-note.md>

To load a record back as a markdown file, run:

lamin load https://lamin.ai/laminlabs/lamindata/record/mNDJgWFrkWQVW3ox

Watch a mini video: youtu.be/NRzVQXJaRH8

Read more: docs.lamin.ai/organize.

Validate datasets

Here is how you ingest a DataFrame:

import pandas as pd

df = pd.DataFrame({
    "sequence_str": ["ACGT", "TGCA"],
    "gc_content": [0.55, 0.54],
    "experiment_note": ["Looks great", "Ok"],
    "experiment_date": [date(2025, 10, 24), date(2025, 10, 25)],
})
ln.Artifact.from_dataframe(df, key="my_datasets/sequences.parquet").save()  # no validation

The simplest way to validate & annotate the content of the dataframe is the built-in schema valid_features:

ln.Feature(name="sequence_str", dtype=str).save()  # define a remaining feature
artifact = ln.Artifact.from_dataframe(
    df,
    key="my_datasets/sequences.parquet",
    schema="valid_features"  # validate columns against features
).save()
artifact.describe()

You can filter for datasets by schema and then launch distributed queries or batch load distributed datasets. For tables, see: docs.lamin.ai/tables. For arrays, see: docs.lamin.ai/arrays.

Watch a mini video: youtu.be/Ji6E7hTnReQ

To validate an AnnData, call:

import anndata as ad
import numpy as np
import pandas as pd

adata = ad.AnnData(
    X=np.ones((21, 10)),
    obs=pd.DataFrame({'cell_type_by_model': ['T cell', 'B cell', 'NK cell'] * 7}),
    var=pd.DataFrame(index=[f'ENSG{i:011d}' for i in range(10)])
)
artifact = ln.Artifact.from_anndata(
    adata,
    key="my_datasets/scrna.h5ad",
    schema="ensembl_gene_ids_and_valid_features_in_obs"
).save()
artifact.describe()

To validate a SpatialData or any other array-like dataset, you need to construct a Schema. You can do this by composing simple pandera-style schemas: docs.lamin.ai/curate.

Use branching & versioning

LaminDB co-versions code and datasets for you. If edit and run the create_fasta.py script, you'll automatically create a new version of the transform and the sample.fasta artifact.

# create_fasta.py
import lamindb as ln

ln.track()
open("sample.fasta", "w").write(">seq1\nTGCA\n")  # a new sequence
ln.Artifact("sample.fasta", key="sample.fasta", features={"experiment": "Experiment 1"}).save()  # annotate with the new experiment
ln.finish()

artifact_latest = ln.Artifact.get(key="sample.fasta")  # pass version for a previous version: ln.Artifact.get(key="sample.fasta", version="1.0")
artifact_latest.versions.to_dataframe()                # all versions of that artifact
artifact_latest.transform.versions.to_dataframe()      # all versions of the transform that created the artifact

To isolate changes, create a contribution branch and switch to it as in git:

lamin switch -c my_branch

To merge a contribution branch into main, run:

lamin switch main  # switch to the main branch
lamin merge my_branch  # merge contribution branch into main

Read more: docs.lamin.ai/manage-changes.

Watch a mini video: youtu.be/rzRwcMj6-fc

Share datasets

To share data in a lineage-aware way, transfer objects from a source database to your default database:

db = ln.DB("laminlabs/lamindata")
artifact = db.Artifact.get(key="example_datasets/mini_immuno/dataset1.h5ad")
artifact.save()

This is zero-copy for the artifact's data in storage. Read more: docs.lamin.ai/transfer.

Manage ontologies

Plugin bionty gives you >20 public ontologies as SQLRecord registries. This was used to validate the ENSG ids in the adata just before.

import bionty as bt

bt.CellType.import_source()  # import the default ontology
bt.CellType.to_dataframe()   # your extensible cell type ontology in a simple registry

You can then create objects, e.g. for labeling, analogous to ULabel, Project, or Record:

t_cell = bt.CellType.get(name="T cell")
artifact.cell_types.add(t_cell)

Read more: docs.lamin.ai/manage-ontologies.

Watch a mini video: youtu.be/3vpWjHj3Kw8