Chat, not code
Describe the task as you would to a colleague. The agent maps it to ready-made tools and shows the plan before it runs.
For principal investigators, clinician-scientists, and medical doctors
Run reproducible analysis
by describing it.
Ask in plain language, and a Claude agent finds the right curated tool, installs it, and runs it on your own data. No pipelines to wire, no code to write. Every result traces back to the exact signed tool that produced it. For research and decision support, not a medical device.
No card required for the free tier. Built on the open Album framework and installable upstream tools (bioconda, PyPI, Bioconductor).
What it is
A scientist’s agent, not another pipeline manager
You know the analysis you want. fragmentum removes the friction around it, finding, installing, and running a trustworthy tool, and keeps the science attributable.
Reproducible by construction
Each tool has a pinned conda environment at catalog:group:name:version. The same address rebuilds the same result.
Traceable results
Every result declares its tool's coordinates, arguments, citations, and license, traceable to the exact method and code.
Your data, your machine
Tools run locally in isolated environments on the host you control. No hidden cloud step.
How it works
Discover → check → install → run → verify
Five steps, all driven from the chat. You stay in the loop: the agent shows its plan and asks before it changes anything on your machine.
-
01
Discover
Ask in plain language. The agent searches the signed catalog and proposes the tools that fit your task, with their inputs, outputs, and citations.
-
02
Check the host
Before running, the agent reports what your machine can do (CPU, RAM, disk, GPU, SLURM, micromamba) and whether it meets each tool’s requirements, with any blockers or caveats named up front.
-
03
Install
The tool’s pinned conda environment is created with micromamba, pulling real upstream packages from bioconda, PyPI, or Bioconductor. Nothing is reimplemented or hand-rolled.
-
04
Run
The agent runs the tool on your inputs and can chain tools into a pipeline, passing one step’s output file to the next. Write actions ask for your approval first.
-
05
Verifiable result
You get outputs plus the provenance to defend them: the coordinates, arguments, and citations behind every step, so the result stands up to review.
A pipeline is just tools, chained
The agent wires the output of one step to the input of the next. Each step keeps its own data directory, so the intermediate files survive and the wiring stays explicit.
ngs-qc:fastp-preprocess
raw reads (FASTQ) → cleaned reads
read-alignment:align-short-reads
cleaned reads → aligned BAM
variant-calling:call-variants-bcftools
aligned BAM → variants (VCF)
What’s inside
About 58 curated solutions, across the work you actually do
Each solution wraps a real, installable upstream tool and is hand-checked before it lands. The catalog grows with every release. These are research tools, not clinical software.
Rare-disease genomics
Turn symptoms into standard phenotype terms, rank candidate diseases and genes, and prioritise variants in an exome.
Secondary analysis (FASTQ → VCF)
Read QC, alignment, and variant calling as chainable building blocks, from raw reads to a filtered variant set.
Functional genomics
Tools for regulatory and functional questions, including transcription-factor binding-site scanning and related analyses.
Single-cell
Quality control, clustering, and downstream steps for single-cell data, wired as reproducible solutions.
Clinical biostatistics
Survival analysis, group comparisons, and the statistical work that supports a study, with pinned dependencies.
Protein & sequence
Structure and sequence tools for protein and nucleotide work, each with declared inputs and citations.
Pharmacogenomics
Tools that connect genetic variation to drug response, for research into how variants affect treatment.
◇ Growing every release
New solutions land regularly. When you connect a catalog, the agent reads the actual solutions and arguments available to you.
Signed, immutable, and guarded
Signed & immutable
Every fragmentum solution is cryptographically signed with ed25519 over a content digest, so a solution cannot be altered without breaking its signature. The agent can verify integrity before it runs.
Host capability checks
The agent checks whether your machine can run a solution before running it, and tells you the blockers, warnings, and notes rather than failing halfway through.
Guardrails
A LiteLLM-style guardrail layer screens every tool result for prompt injection, secret leakage, and exfiltration patterns before it reaches the model loop.
Plans
Start free, scale when your work does
Plans are metered by agent runs and tokens per day. Prices are shown at checkout; create an account to see current pricing for your region.
Free
€0 to start
For trying the agent on small tasks.
- 10 agent runs per day
- 50,000 tokens per day
- Discover, inspect, and plan tools
- No file uploads
- No card required
Researcher
Contact us for pricing
For an individual scientist running real analyses.
- 200 agent runs per day
- 1,000,000 tokens per day
- Upload your own input files
- Everything in Free
- Priority for new solutions
Team
Contact us for pricing
For a lab or group with heavier throughput.
- 2,000 agent runs per day
- 10,000,000 tokens per day
- Upload your own input files
- Everything in Researcher
- Institutional SSO available
Quotas are daily limits on agent runs and tokens, reset each day. Prices are handled at checkout through Paddle and vary by region, so we do not list them here. Comped access is available for accelerator partners and testers.
Bring an analysis. Leave with a reproducible result.
Create a free account and ask the agent your first question, or watch the recorded walkthrough first.
Research use only. fragmentum helps you discover and compose analysis software. It supports expert work and is not a medical device. It does not provide a diagnosis or any clinical recommendation. Every result depends on the underlying solutions and their inputs, and must be reviewed and confirmed by qualified people using independent evidence before it informs any decision about a person’s care.