Now in early access
AI for scientific discovery
Models trained on 80 million papers and research agents built to explore the unknown, from your first question to a verified finding.
Question
What predicts treatment response in cohort_v2?
- plan.md
Plan, done
Exploration plan drafted and approved
- profile.html
Explore, done
38 variables profiled, 3 anomalies found
- hypotheses.md
Hypothesize, done
12 patterns ranked as hypotheses
- tests.py
Test, done
Top 3 tested, 2 supported by the data
- findings.docx
Write, done
Findings and figures drafted
- verify.md
Verify, needs review
Every number rerun; 1 finding for your review
Built for scientific discovery
Most AI tools finish the task they are given. MindCODE models and agents are built to find what is not yet known.
MindCODE models
Trained on the scientific record
80M
scientific papers in training
~200
years of scientific progress
- Trained specifically for scientific exploration
- Built for research code, statistics and analysis
- Hosted by MindCODE, or private on your own servers
- mc-code-1
- mc-code-1-lite
- mc-code-1-fast
- mc-embed-1
MindCODE agents
Designed to explore the unknown
They start from an open question and look for patterns and laws that no one has described yet. You approve the plan, and the agents do the exploring.
1Explore
Profile every variable and read the literature around your question.
2Hypothesize
Rank the patterns and anomalies they find as hypotheses worth testing.
3Test
Write and run the analyses, and keep what the evidence supports.
4Verify
A separate agent reruns every number and checks every citation.
One platform for AI-driven research
Four products that work alone or together, from a single analysis to a complete study.
Research Agents
Built for discovery
Agents that explore your data and the literature, design studies, analyze results and write and review manuscripts.
- Exploratory analysis that ranks patterns as hypotheses
- Study design, analysis, figures and writing
- Review and journal selection before you submit
Autonomous Runs
Start with a question, come back to findings
An agent plans the exploration, writes and runs the code, tests what it finds and checks the results on its own.
- Plans the work and waits for your approval
- Writes and runs its own code to test hypotheses
- Returns findings, figures and a report
Model API
MindCODE models through one key
The mc-code-1 model family and mc-embed-1, trained for scientific exploration and served through an OpenAI-compatible API.
- Trained on 80 million papers
- Works with the OpenAI SDK
- Hosted, or private in your environment
Verify
Every claim, every number
Checks each citation against its source and reruns each result from the code and data.
- Citations matched to their sources
- Results regenerated from code and data
- Every finding located in your files
Built for medicine and the life sciences
From adolescent cohorts and brain imaging to clinical models and multi-omics.
Youth mental health
Studies of adolescent mental health, from school cohorts to the AI products that teens use.
Learn moreBrain science
Neuroimaging and connectome analysis at the scale of modern brain datasets.
Learn moreClinical AI models
Diagnostic, prognostic and treatment-response models, developed and validated on your clinical data.
Learn morePsychiatry and digital phenotyping
Speech, facial expression, movement and sleep turned into measurable signals of mental health.
Learn moreMedical language models
Language models that read clinical text, running inside the hospital network.
Learn moreMulti-omics and precision medicine
From expression data and gene lists to biomarkers and pathways.
Learn moreStart with a question. Come back to findings.
Every autonomous run follows the same four steps.
- 1
Plan
The agent proposes how to explore your question, and you approve the plan before work starts.
- 2
Explore
It reads the literature, profiles the data, and writes and runs code to test what it finds.
- 3
Verify
A separate agent checks every claim, citation and number.
- 4
Deliver
You receive the findings, the code, the figures and the verification report.
Model API
MindCODE models in a few lines of code
Models trained for scientific exploration, called with the OpenAI SDK you already use. Change the base URL and the key, and keep the rest of your code.
- Chat, embedding and model list endpoints
- Hosted, or private in your environment
- Priced per million tokens, volume pricing negotiable
from openai import OpenAI
client = OpenAI(
base_url="https://api.mindcode.cc/v1",
api_key="sk-mc-...",
)
reply = client.chat.completions.create(
model="mc-code-1",
messages=[
{"role": "user", "content": "Summarize cohort_v2.csv"},
],
)
print(reply.choices[0].message.content)curl https://api.mindcode.cc/v1/chat/completions \
-H "Authorization: Bearer $MINDCODE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "mc-code-1",
"messages": [
{"role": "user", "content": "Summarize cohort_v2.csv"}
]
}'Services
When you want our team to build it with you.
Custom models
Models tuned to your science
We fine-tune open-weight models on your data and test them on your own tasks.
Learn morePrivate deployment
Your data stays with you
Agents and models run inside your cloud account or on your own servers.
Learn moreResearch services
Our scientists and agents, on your project
We deliver analyses, figures, manuscripts and reviews, start to finish.
Learn moreBring us your next study
Start with a pilot on a real project: a manuscript, a dataset or a model.