Request access

Where scientists and AI agents do research together.

Write your goal in plain words. Cytogent turns it into a research brief, then AI agents and your team do the work.

Request accessSee an example
NVIDIA Inception Program member badge

From a rough idea to a signed plan.

Pick an example. Watch Cytogent ask the right questions, write the brief, split the work between agents and people, and build the file.

  1. Dr. Berg · 09:00

    Our lung cancer cell line stopped responding to a KRAS inhibitor after six weeks. Why, and what should we test next?

    • rna-seq_wk0_wk6.csv
    • variants.vcf

  2. Planner · 09:01

    Three questions before I plan.

    1. What counts as a good answer?Answer: Top 3 causes, each with one test
    2. What can the lab do in the next four weeks?Answer: Two CRISPR knockouts, no animal work
    3. What do we compare with?Answer: The parental line

  3. Brief signed by Dr. Berg09:04

  4. Planner · 09:05

    I split the brief into 6 tasks: 4 for agents, 2 for people.

  5. Analyst · 11:40

    The week-6 sample has a mutation at a known resistance site. Flagged for review.

  6. Writer · 14:12

    Report v0.2 is ready: 3 causes, 3 tests, 18 citations.

Research brief RB-014

DraftSigned
Question
What causes acquired resistance to the KRAS inhibitor in line LC-7R?
Hypothesis
A second mutation in KRAS, or a bypass pathway.
Data
RNA-seq at week 0 and week 6, variant list, dose–response curves.
Control
Parental line LC-7.
Success
Top 3 causes, each with evidence and one test.
Limits
Two CRISPR knockouts. No animal work.
Owner
Dr. A. Berg · signed 09:04

Task board RB-014

To do 1

  • CG-105

    Order guides for the top 2 targets

    Jonas, lab manager
    Person

In progress 3

  • CG-101

    Read 40 papers on KRAS inhibitor resistance

    Readerlong-context LLM
  • CG-102

    Compare variants with known resistance sites

    Analystcode model
  • CG-104

    Approve the knockout plan

    Dr. Berg
    Needs Dr. Berg

Done 2

  • CG-100

    Brief signed

    Dr. Berg
  • CG-103

    Dock the inhibitor into the mutant pocket

    Analyststructure model

Resistance report v0.2

3 causes · 3 tests · 18 citations

  1. Summary
  2. Evidence table
  3. Docking results
  4. Proposed tests
  5. Bench protocol: CRISPR knockout

Every claim links to its source. [1] [2] [3]

ExportELNWordPDF

Illustrative example. Names, data and numbers are made up.

See how the platform works

You don't need the perfect prompt.

A research question is not a prompt. It hides choices about controls, endpoints, sample size, ethics and regulation. Cytogent asks about them first, so agents plan from a brief your PI would sign.

Write it your way

Type the goal the way you would tell a colleague. Attach data if you have it.

Answer a few questions

Cytogent asks only what changes the plan: data, what counts as an answer, limits and rules.

Sign the brief

One page with the question, hypothesis, data, controls, endpoints, rules and owner. Agents plan from it, and every result links back to it.

Agents plan the work and give work back to people.

After you sign the brief, agents split it into tasks. Some go to agents. Some go to people: a lab run, a review, an approval. Results come back into the same record.

  • Agents take the reading, analysis and drafting.
  • People get lab work, reviews and sign-offs, with the protocol attached.
  • New tasks appear as results come in.
  • Anything that leaves the project waits for a yes.

An illustrative project board · agents and a person at work · one draft waiting for you

Built for the people who move science forward.

Pharma & biotech R&D

Too many tools, too little evidence per decision.

Target triage, candidate ranking and design loops with cited sources.

See this page

CROs & clinical teams

Protocols, CRFs and reports take months to draft.

Trial documents drafted from your study data, ready for review.

See this page

Hospitals & academic labs

Cohort analysis under strict ethics and data rules.

Analysis and writing with a full audit trail inside your permissions.

See this page

Regulatory & IP teams

Submissions and filings built by hand from scattered results.

Structured drafts for IND/CTA, IVDR files and patent applications.

See this page

From question to signed result in four steps.

  1. 01

    Write

    Your goal in plain language. Attach data or connect your sources.

  2. 02

    Brief

    Cytogent asks a few questions. You confirm and sign the research brief.

  3. 03

    Work

    Agents and people do the tasks. Each step goes to the best model and is logged.

  4. 04

    Decide

    Review cited results, sign off, and export to your notebook, LIMS or report.

You → Planner → Reader, Analyst, Writer and a person → a signed result

Seven workflows. One evidence trail.

One query over papers, patents and protocols · three become citations

Literature & evidence

Papers, patents, protocols and registries in one query. Every claim points to its source.

PubMedpatentsprotocols
Open literature & evidence

A ligand docks into the pocket · candidates ranked

In-silico studies

Screen, simulate and compare before the bench: docking, ADMET, pathway and PK models.

dockingADMETsimulation
Open in-silico studies

Sequence → predicted fold → candidate with validation

Protein studies & design

Structure prediction, binding analysis and sequence design, with validation reports.

structurebindingdesign
Open protein studies & design

Guide finds the site · cut · repaired · off-targets reviewed

CRISPR & genome editing

Guide design, off-target review and screen analysis, documented for the lab.

guide RNAoff-targetscreens
Open crispr & genome editing

Plan → Start → Run → Close · the cohort fills in

Clinical trials, full cycle

Protocol drafts, site feasibility, cohort criteria, CRFs, statistical analysis plans, monitoring summaries and CSR drafts.

protocolCRFSAPCSR
Open clinical trials, full cycle

IND modules checked one by one · a reviewer question answered

Regulatory documentation

IND/CTA modules, IVDR technical files and SOPs, structured for reviewer questions.

INDCTAIVDRSOP
Open regulatory documentation

Claim tree · prior art on a timeline · novelty argued

Patent & IP documentation

Prior-art search, invention disclosures and claim-drafting support from your own results.

prior artclaimsdisclosure
Open patent & ip documentation

Follow a single question through the workspace.

You ask · two datasets attach to the question

The Planner asks · the brief fills in · the PI signs

The Reader agent reads 40 papers · three become citations

The Analyst docks the ligand · candidates ranked · G12C flagged

Guides designed · off-targets checked · a lab task goes to the bench

The Writer assembles the protocol · cohort defined · SAP drafted

IND modules and claim draft from the same sources · a scientist signs off

Illustrative example

  1. 01

    A question arrives

    The team asks why a cell line resists an inhibitor. They attach assay results and a variant list.

  2. 02

    The brief is signed

    Cytogent asks which data to use, what counts as an answer and what the limits are. The PI signs a one-page brief.

  3. 03

    Evidence is gathered

    The Reader agent reviews 40 papers and two cohort datasets and cites the three that explain the mutation.

  4. 04

    Models test the idea

    In-silico docking and a variant-effect model rank the mutation and suggest a binding-site change.

  5. 05

    An edit is designed

    Guide RNAs are designed and checked for off-targets. An agent creates a lab task for the bench team, with the protocol attached.

  6. 06

    A study is planned

    A cohort definition, protocol draft and statistical plan are prepared for the clinical team to review.

  7. 07

    The record is written

    Regulatory and patent drafts are built from the same cited results. A scientist signs off at each checkpoint.

Keep your AI tools. Choose Cytogent for research.

ChatGPT, Claude, Copilot and agent workspaces are great for general work. Cytogent does the same agent work, and is built for how research moves from question to file.

Assistants e.g. ChatGPT, Claude, CopilotWorkspaces e.g. ChatGPT dots, Grok Bots, Dust

FeatureAI assistantse.g. ChatGPT, Claude, CopilotAgent workspacese.g. ChatGPT dots, Grok Bots, DustCytogent
Agents that work in the backgroundSomeYesYes
One workspace for people and agentsSomeYesYes
Connects to your apps and the webYesYesYes
Starts from a signed research briefGeneral plan stepGeneral plan stepBuilt for research
Agents create lab tasks for peopleNot built inApproval requestsWith protocol and owner
Life-science models built inVia pluginsVia pluginsBuilt in
One cited trail to the final filePer answerPer taskAcross 7 workflows
Trial, regulatory and patent documentsGeneral writingGeneral writingTemplates and checks
Scientist sign-off in the audit logVariesApproval rulesEvery checkpoint
Models from many providersVariesVariesYes, logged

Examples are for orientation, based on public product information checked on 2 October 2026. Product names are trademarks of their owners.

See the full comparison

Two pilot studies, running now.

Both turn a science finding into industry-ready work.

Guide finds the site · cut · repaired · off-targets reviewed

Cervixel

CRISPR research for cervical cancer diagnosis.

CRISPR & genome editingLiterature & evidence

Plan → Start → Run → Close · the cohort fills in

Eipha Biosciences

Clinical trial for a vitiligo skin patch.

Clinical trials, full cycleRegulatory documentation

Read about the pilots

NVIDIA Inception Program member badge

Built on NVIDIA technology.

WelloWork AB, the company behind Cytogent, is a member of the NVIDIA Inception Program. The science models our agents use run on NVIDIA GPUs and NVIDIA BioNeMo NIM microservices.

  • OpenFold
  • DiffDock
  • ESM
  • RFdiffusion + ProteinMPNN
  • Parabricks
See data and models
NVIDIA stackThree layers. At the bottom, a row of NVIDIA GPUs. In the middle, NVIDIA BioNeMo NIM microservices: OpenFold, DiffDock, ESM, RFdiffusion + ProteinMPNN and Parabricks. On top, three Cytogent agents: Reader, Analyst and Writer. Thin lines join the layers, and a pulse runs up from a GPU through a model to the Analyst, one route at a time. A run log reads: 12:05, predict structure on OpenFold; 12:06, dock ligand on DiffDock.CYTOGENT AGENTSReaderAnalystWriterBIONEMO NIM MICROSERVICESOpenFoldDiffDockESMRFdiffusion+ ProteinMPNNParabricksNVIDIA GPUS12:05predict structure →OpenFold12:06dock ligand →DiffDock

NVIDIA GPUs → BioNeMo NIM models → Cytogent agents

Built on data you can trace.

Datasets

Public and licensed collections, cleaned, versioned and documented. Each one lists its source, version and licence inside the workspace. Literature includes full-text papers from ScienceDirect, through Elsevier's API.

Trained models

Domain models for prediction and screening (variant effect, binding affinity, assay QC). Each shows its validation on its own card. Trained and served on NVIDIA GPUs.

In progress

Protocols

Protocols you can search, adapt and cite, with every step attributed to where it came from.

Health data standards

Patient data is modelled with openEHR, exchanged with FHIR, mapped to OMOP for multi-site studies, and coded with SNOMED CT.

See data and models

Datasets with source, version and licence · models with validation · the curation path

Your data stays yours.

  • Project isolation

    Each project has its own storage scope. Agents see only its data.

    Done
  • No training on your data

    Your data never trains shared models.

    Done
  • Encryption in transit

    TLS 1.2 or higher on every connection.

    Done
  • Encryption at rest

    Managed keys, rotated on a schedule.

    In progress
  • Role-based access

    Owner, editor and viewer roles per project and dataset.

    Done
  • EU data residency

    Storage and processing inside the EU.

    In progress
  • SOC 2 Type II

    Independent audit of controls.

    Planned
  • Analytics only with consent

    Google Analytics runs only if you accept it in the cookie banner.

    Done

Cytogent is a research tool. It is not a medical device and makes no clinical decisions.

Read the full security page

Three isolated projects · a crossing refused · roles · the audit log

Access is by request.

Individual

Scientists and students with an institutional email.

Request access

Institute

Companies, academic institutes and CROs.

Request access

Hospital

Clinical teams and hospital units working with patient data.

Request access

Tell us what you want to find out. We reply within five working days with a first draft of your research brief.

Questions scientists ask first.

What is Cytogent?

Cytogent is an agentic workspace for life science research. Scientists write what they want to find out, Cytogent turns it into a signed research brief, and AI agents and people do the work together, from literature and in-silico studies to trials, regulatory files and patents. Every result is cited. Cytogent is operated by WelloWork AB in Sweden.

How is Cytogent different from ChatGPT, Claude or Copilot?

Keep them for everyday work. Cytogent uses the same kind of frontier models, but it is built for research. It starts from a signed research brief, gives lab tasks to people, runs life-science models, and keeps one cited trail from the first question to the final protocol, regulatory or patent file.

How is Cytogent different from agent workspaces like ChatGPT dots, Grok Bots or Dust?

Those tools run always-on agents for any kind of work, and Cytogent does too. The difference is focus. Cytogent plans from a research brief, knows trial, regulatory and patent formats, records scientist sign-off at each checkpoint, and keeps every claim linked to a source you can open.

Which science models does Cytogent use?

Cytogent runs life-science models through NVIDIA BioNeMo NIM microservices: OpenFold for structure prediction, DiffDock for docking, ESM for protein embeddings, and RFdiffusion with ProteinMPNN for protein design. NVIDIA Parabricks handles genomics. WelloWork AB, the company behind Cytogent, is a member of the NVIDIA Inception Program. Membership does not imply NVIDIA endorsement.

What is a research brief?

A one-page plan you sign before agents start. It holds the question, hypothesis, data, controls, endpoints, limits, rules and owner. You write the goal in your own words, and Cytogent asks only the questions that change the plan. Agents plan from the brief, and every result links back to it.

Can agents give tasks to people?

Yes. When a step needs hands or judgment, an agent creates a task for a person: a lab run with the protocol attached, a review, or an approval. Each task has an owner and a due date. When the person adds the result, it goes back into the same record for everyone to see.

Who is Cytogent for?

Four groups. Pharma and biotech R&D teams who need evidence behind every decision. CROs and clinical teams who draft protocols, CRFs and reports. Hospitals and academic labs doing cohort work under strict ethics and data rules. Regulatory and IP teams who build submissions and patent filings from scattered results.

Is my data used to train models?

No. Data you bring to Cytogent stays inside your project and never trains shared models. Projects are isolated from each other, access is granted per role, and every agent action against your data is recorded in an audit trail you can read. WelloWork AB acts as processor under a data processing agreement.

How do I get access?

By request. There is no self sign-up and no free trial. Choose Individual, Institute or Hospital, and tell us what you want to find out. We read every request, reply within five working days, and set up a workspace with your permissions, starting from a first draft of your research brief.

Is Cytogent a medical device?

No. Cytogent is a research tool. It makes no clinical decisions, gives no diagnosis and is not certified as a medical device under the MDR or IVDR. Agents draft, search, predict and support. A qualified scientist or clinician reviews and decides, and that review is part of the record.

Bring your next question.

Tell us what you want to find out. We set up the workspace and send you a first research brief.

Request access