Tumbleweed Scientific

Life Science in Continuity

Papers, models and know‑how, in one research context

Tumbleweed Scientific is a research agent platform for life science. It reads full papers, not abstracts, runs AI4Science models on demand, and turns your team's methods into reusable SOPs. No more digging for papers, renting GPUs or moving data between tools.

AI4Science models, ready to run
50+
Public datasets, one click away
70+
Full-text academic sources
6

Capabilities

One agent platform, three research infrastructure services

Research stalls in the hand-offs between steps. Tumbleweed puts conversation, models, literature and data in one context.

TumbleweedAgent platform

You ask and decide. The agent does the rest.

In your project workspace the agent searches, runs models and organises results. It checks key decisions with you through forms and approval cards, answers with charts and 3D structures, and ties every step to its evidence and parameters.

01 / 03Human in the loop

Powered by three infrastructure services

  1. Models

    AI4Science model runs

    Frontier models, one sentence away

    50+ models across 8 areas, including structure prediction, docking, protein design and single-cell, queued and run in isolation. Results return to the chat with their parameters.

    • 50+ models
    • 8 areas
    • No GPUs to buy
    • Multi-model pipelines
  2. Knowledges

    Literature library and processing

    Turn your library into knowledge

    Build a literature library from 6 sources including PubMed and OpenAlex, parse the full text, then turn it into RAG indexes, knowledge graphs and an LLM Wiki.

    • Full-text parsing
    • RAG
    • Knowledge graph
    • LLM Wiki
    • MCP
  3. HyperDrive

    Dataset connections

    Public and local data, in one place

    Connect 70+ public datasets in moments and mount your lab's own private datasets, so the agent can search and analyse them directly.

    • 70+ public datasets
    • Local data mounts
    • Ready to use

Use Cases

One conversation, from study design to a deliverable

Full conversation replays from the platform.

Completed

Liver cancer RNA-seq differential expression

The ask

Transcriptome differences between HCC tumour and adjacent tissue, as a full analysis and report ready for lab meeting.

What the agent did

  1. Loaded the team's rnaseq-de-sop skill
  2. Confirmed species, paired design, sample size and goal in a form
  3. Pulled public datasets, ran differential expression and pathway enrichment
  4. Delivered a summary dashboard and an exportable biomarker table

Replays use example data, not real experimental results.

Testimonials

Trusted by researchers

Researchers at universities, institutes and pharma companies move their projects forward with Tumbleweed.

  • Testimonial · to be added
  • Testimonial · to be added
  • Testimonial · to be added

Frequently asked questions

How is this different from a general AI assistant?

A general assistant answers from memory. Tumbleweed reads full papers and runs real scientific models and datasets, traces every conclusion back to its source and parameters, and keeps your project context.

Is my data safe?

Projects, conversations and data are isolated per workspace, and only members you invite can see them. Institutions with stricter compliance needs can find the Enterprise plan on the pricing page.

Which models and data are supported?

Models offers 50+ AI4Science models across 8 areas, including structure prediction, docking, protein and peptide design, single-cell and genomics. HyperDrive connects 70+ public datasets and can mount your local data.

Do I need to code?

No. The agent confirms key details through forms, options and approval cards, and shows results as charts, tables and 3D structures. You make the call on every key step.

How do I get started?

Individual researchers can sign up right away. For teams and institutions, see the pricing page.