Science meets agentic AI

Scientific Software,
AI-Orchestrated

Fragmentum builds album — the open-source platform that makes scientific tools reproducible, shareable, and now AI-orchestrable. Born in research. Built for scale.

See album-mcp → Get Started album.solutions ↗
10+
Public Catalogs
FAIR
Principles Built-in
3
Platforms
MCP
AI-Native Protocol
MIT
Open Source
The Problem

Scientific software is broken — and it costs billions

Researchers waste time fighting tooling instead of doing science. Industry loses money on pipelines that can't reproduce.

40%
Researcher time wasted
Scientists spend nearly half their time wrestling with software installation, dependency conflicts, and environment setup instead of doing research.
$28B
Annual reproducibility cost
The US alone loses an estimated $28 billion per year on irreproducible preclinical research — much of it due to computational failures.
6–8 wks
Tool integration time
Onboarding a new computational tool into an existing lab or enterprise pipeline takes weeks of engineering effort — for every single tool.
The Platform

One framework for every scientific workflow

Album solves the reproducibility crisis in scientific computing. Wrap any existing code into a versioned, shareable solution file. Run it anywhere — full data ownership guaranteed.

Decentralized by design Host your catalog on any Git platform. No central server lock-in, no vendor dependency. Your IP stays yours.
Conda-powered isolation Each solution ships its own environment. Zero dependency conflicts — 50 tools co-existing on one machine.
Research-validated Research (arXiv 2110.00601). Used in production across multiple research institutes and scientific communities.
Multi-interface access CLI, Python API, REST API, GUI, and now MCP — use whichever fits your workflow or product integration.
image-analysis · 12 solutions ● connected
🔬
cellpose-segmentation
v2.1.0 · Python · GPU
Ready
🧬
n2v-denoising
v1.0.2 · Python · HPC
Ready
📊
fiji-macro-runner
v3.0.1 · Java · Interactive
Running
🤖
sam2-segmentation
v1.0.0 · Python · CUDA 12
Installing
ilastik-classifier
v1.3.3 · Python · C++
Ready
Features

Reproducible science. Scalable infrastructure.

From a single laptop to HPC clusters and cloud deployments — album handles the complexity so researchers and teams don't have to.

🌐

Decentralized Catalogs

Host your solution catalog on any Git platform — GitLab, GitHub, or self-hosted. No central registry, no lock-in. Full data sovereignty.

♻️

Reproducible Environments

Each solution defines its own conda environment with locked dependencies. Install once, reproduce exactly — across any platform.

🔌

Multi-Language Support

Wrap solutions in Python, R, Java, Kotlin, C++, and more. One unified registry for heterogeneous tool stacks.

📋

FAIR Compliant

Findable, Accessible, Interoperable, Reproducible. Built-in support for FAIR research data principles and regulatory traceability.

🐳

Cloud, HPC & Edge

Deploy via Docker, Singularity, or natively on HPC clusters. Same solution definition, zero reconfiguration across environments.

REST API + SDK

Sync and async REST endpoints, Python SDK, and CLI. Integrate album into existing platforms, LIMS, or orchestration layers in hours.

✨ Agentic AI · album-mcp

Let AI orchestrate
your science

album-mcp brings the Model Context Protocol to album — enabling AI agents to discover, install, compose, and execute scientific solutions through natural language. 16 tools. One protocol. Full lifecycle.

🔍

Discover & Inspect

AI agents search across all connected catalogs, inspect solution arguments and metadata, and find the right tool for the task — no CLI memorization needed.

🔗

Autonomous Pipeline Composition

The AI scaffolds multi-step pipelines, wires persistent data paths between steps, generates complete solution code, validates against the RDF schema, and deploys — all autonomously.

🔒

2FA-Secured Execution

TOTP-based authentication gates every destructive operation — install, run, deploy, remove. Enterprise-grade security for AI-driven scientific workflows.

🏗️

AI Writes Solutions, Not Just Runs Them

album-mcp doesn't just execute — it authors. The AI generates scaffold code, writes lifecycle functions (run, install, test), assembles valid solution files, and publishes to catalogs. Solutions creating solutions.

album-mcp · AI Session
User » Denoise and segment all images in /data/batch_042/

── album_search_solutions("denoise") ──
Found: n2v-denoising (v1.0.2)
── album_get_solution_info("imaging:n2v:1.0.2") ──
Args: --input (file) --output (file)
── album_install_solution ── 2FA ✓
Conda env provisioned (cached)
── album_run_solution ── 2FA ✓
128 images denoised → /data_path/n2v/output/

── album_search_solutions("segment") ──
Found: cellpose-segmentation (v2.1.0)
── album_run_solution ── 2FA ✓
12,847 cells detected in 128 images

── album_get_solution_scaffold ──
Pipeline assembled: denoise → segment
── album_validate_solution ──
Metadata valid. Ready for deploy.
10x
Faster pipeline setup
80%
Less onboarding time
0
Lines of glue code
The Protocol

16 AI-callable tools. Full lifecycle coverage.

Every album operation — from catalog management to solution authoring — exposed as a standardized MCP tool that any AI agent can invoke.

Discoverylist_catalogsBrowse all connected registries
Discoverylist_solutionsEnumerate available tools
Discoverysearch_solutionsKeyword search across catalogs
Discoveryget_solution_infoInspect args, metadata, schema
Lifecycleinstall_solutionProvision isolated env 2FA
Lifecyclerun_solutionExecute with typed args 2FA
Lifecycleuninstall_solutionClean up env & cache 2FA
Authoringget_solution_scaffoldGenerate template + wiring code
Authoringassemble_solutionBuild .py from lifecycle bodies
Authoringvalidate_solutionCheck against RDF schema
Publishingdeploy_solutionPush to catalog 2FA
Publishingremove_solutionUndeploy from catalog 2FA
Managementadd_catalogConnect a Git-hosted registry
Managementremove_catalogDisconnect a catalog
Managementupdate_catalogSync index from remote
Adminregister_userTOTP 2FA user management
How It Works

From code to reproducible solution in minutes

A three-step workflow that scales from a single laptop to cloud infrastructure — with AI orchestration built in.

1

Wrap & Register

Package any existing tool into a versioned album solution file with its own environment spec. Push to your Git-hosted catalog.

2

Install & Reproduce

One command provisions the exact conda environment. Colleagues reproduce your workflow by pointing to the same catalog and version.

3

AI Orchestrates & Authors

Connect album-mcp. Your AI assistant discovers solutions, chains them into pipelines, generates new solution code, validates metadata, and deploys — all through natural language.

Traction

Research-born. Production-proven.

Album is already used across leading research institutions and scientific communities worldwide.

🏛️
Active
Production use across research institutes
📄
Cited
Published (arXiv 2110.00601)
📦
10+
Public catalogs (imaging, cryo-ET, DL)
🧬
5+
Scientific communities using album
Market Opportunity

The infrastructure layer for AI-driven science

As AI transforms research workflows, the need for standardized, reproducible, AI-orchestrable tool infrastructure is exploding — and no one owns this layer yet.

🎯

Beachhead: Life Sciences & Bioimaging

Established user base in microscopy, cryo-ET, and cell biology. Expanding into pharma, genomics, and clinical research pipelines.

🔄

Expand: Any Computational Discipline

Album is domain-agnostic by design. The same framework works for climate science, materials research, financial modeling, and data engineering.

💰

Monetization: Managed Catalogs + Enterprise MCP

Open-source core with premium managed catalog hosting, enterprise SSO, audit logging, SLA-backed support, and hosted album-mcp orchestration.

Why now — converging tailwinds
AI Adoption
92%
MCP Ecosystem
78%
Reproducibility
85%
FAIR Mandates
70%
Open Source
88%
Get Started

Ready to defragment your workflow?

Install album in seconds. Open source, MIT licensed. Works on Linux, macOS, and Windows.

$ pip install album
Read the Docs → View on GitLab ↗
Research

Research-validated, open-source

Album is backed by published research. If you use it in your work, please cite:

@article{album2021,
  author = {Albrecht, Jan Philipp and Schmidt, Deborah and Harrington, Kyle},
  title = {Album: a framework for scientific data processing with
            software solutions of heterogeneous tools
},
  year = {2021},
  journal = {arXiv preprint arXiv:2110.00601}
}
View on arXiv ↗