Skip to content
NewExplore seven live examples

Made at MIT CSAIL

Representation superintelligence for every team.

Mantis understands and reasons over anything your organization knows, sees connections across thousands of dimensions, and gives every team and every agent situational awareness, with every answer traced to its evidence.

Every citation lights the exact records behind it

Understand and reason over anything, from a first question to a decision.

Mantis reads the Space, searches the map, runs the analysis in its own environment, outlines the cluster, saves a bag and answers with citations that light the exact records.

Five windows of Mantis at work, with illustrative data. In the workspace, asked “Which neurons are lost earliest in Alzheimer’s disease, and where do they sit on this map?”, Mantis reads the Space, searches the map, runs an analysis, outlines the Sst cluster, saves a bag and answers with citations that light the exact records. The other windows switch between the public spaces’ maps, turn approved kinase inhibitors in 3D beside their computed properties, compare two groups of Alzheimer’s disease research over time, and show a team sharing a space.

Who it is for

Built for the teams pushing science forward.

Representation superintelligence changes the work of every research team: the whole of what it knows, in view at once, with an agent that reasons over it and shows its evidence.

  • Discovery biology

    Every cell, sample and study in one representation. Mantis finds the population that carries a signal, compares conditions inside it, and cites the cells.

  • Chemistry

    Compounds placed by structure and activity. Mantis triages leads with their computed properties and their neighbours on the map beside them.

  • Literature and evidence review

    A field read at once. Mantis groups thousands of papers by what they say and links each finding to the papers behind it.

  • Data science

    Any table, notebook or repository embedded and navigable. Agents work through the same tools as you, and every step is inspectable.

  • Research leadership

    Situational awareness across every project: what changed, what is related, and what matters now, with the evidence one click away.

Trusted by teams at

  • Massachusetts Institute of TechnologyMIT Computer Science and Artificial Intelligence Laboratory
  • Broad Institute of MIT and Harvard
  • Tsinghua University
  • Stanford University
  • University of California, Berkeley
  • University of Oxford
  • Harvard University
  • The University of Texas at Austin
  • University of Cambridge

How it works

Representation superintelligence

Mantis learns a representation of everything you give it, across thousands of dimensions, and reasons over that representation. That is what gives every team its whole context: every record has a place, every relation is measurable, and every answer can point at its evidence.

  1. 01Records

    Bring anything

    Papers, datasets, documents, code, cells, molecules and conversations, from your own collections or from public corpora such as arXiv and PubMed.

  2. 02Representations

    Learn its representation

    Every record is embedded across thousands of dimensions by models suited to it, so that distance means difference in meaning.

  3. 03Structure

    See every connection

    The representation is laid out, clustered and named, from the whole corpus down to a single record, so people and agents navigate the same structure.

  4. 04Reasoning

    Reason with agents

    Mantis and your own agents work on that structure: they search it, select from it, compare within it, and cite the exact records behind every answer.

1000x faster from question to evidence

Transparency

Every answer traces to its evidence.

Every point is a real record. Every cluster, answer and agent action traces back through the representation to the exact evidence. Nothing is a black box: you see why things are related and where an answer came from.

Cited records
Each citation lights the records behind it on the map, one click from the source.
Visible reasoning
Every tool call, selection and comparison stays in the thread, in order.
Shared structure
People and agents read the same map, so a finding can be checked by anyone on the team.

Live examples

Mantis at work on real datasets. Every point is a real record, every cluster is labelled, and every connection is one click from its evidence. Sign in and move through any of them.

  • COVID-19 PBMC, open this space in Mantis

    Life sciences · 59,572 points

    59,572 blood immune cells from 20 samples: healthy donors, asymptomatic to severe COVID-19 and severe influenza, clustered by cell type, with each patient's clinical measurements and interferon and TNF/IL-1 programme scores.

    Source: CZ CELLxGENE Discover (CC BY 4.0); Lee et al., Science Immunology 5, eabd1554 (2020).

  • SEA-AD brain, open this space in Mantis

    Life sciences · 80,000 points

    80,000 nuclei from the middle temporal gyrus of 89 donors spanning the Alzheimer's disease continuum (SEA-AD), in 24 cell subclasses, with each donor's neuropathology, APOE4 status, cognition and pseudo-progression score.

    Source: CZ CELLxGENE Discover, SEA-AD, Allen Institute (CC BY 4.0); Gabitto, Travaglini et al., Nature Neuroscience 27, 2366-2383 (2024).

  • The human proteome, read by a protein language model, open this space in Mantis

    Life sciences · 20,431 points

    All 20,431 reviewed human proteins (UniProtKB/Swiss-Prot) placed by ESM-2 embeddings, so proteins the model reads as alike sit together, with families, compartments, structures and disease links from UniProt.

    Source: UniProtKB/Swiss-Prot (CC BY 4.0) with ESM-2 embeddings (MIT); The UniProt Consortium, NAR 2025; Lin et al., Science 379, 1123 (2023).

  • Kinase inhibitors, open this space in Mantis

    Chemistry & materials · 26,970 points

    26,970 potent human kinase inhibitors from ChEMBL 37 (top-kinase pChEMBL 7.5 or higher, plus every clinical candidate), laid out by chemical structure and grouped by the kinase family they hit hardest; 145 are approved drugs.

    Source: ChEMBL 37, EMBL-EBI (CC BY-SA 3.0); Zdrazil et al., Nucleic Acids Research 52, D1180-D1192 (2024).

  • Alzheimer's disease research since 2000, open this space in Mantis

    Life sciences · 41,673 points

    41,673 of each year's most cited Alzheimer's disease papers from 2000 to 2026, laid out by what their abstracts say and grouped into 19 research topics, with year, journal, citations and MeSH terms.

    Source: Europe PMC (EMBL-EBI) with MEDLINE records (NLM); abstract text kept only for Creative Commons papers.

  • A map of Mathlib, open this space in Mantis

    Mathematics · 66,154 points

    66,154 theorems of Lean's Mathlib (every documented theorem and every one used by at least five others) placed by what their statements say, in 24 areas, with statements, docstrings and usage counts.

    Source: Mathlib 4, leanprover-community (Apache-2.0), Mathlib v4.34.1; The mathlib Community, CPP 2020.

  • The S&P 500 by business model, open this space in Mantis

    Finance & markets · 492 points

    492 S&P 500 companies laid out by how they make, spend and return money: margins, R&D, capital intensity, leverage, growth and payouts from their SEC filings, grouped into 10 business models that cut across sectors.

    Source: SEC EDGAR XBRL company facts (public domain), latest fiscal year filed; constituent list from Wikipedia (CC BY-SA 4.0). No market prices.

Scale

Context at the scale of an organization.

records mapped
70M+
maps
5,600+
Spaces
4,600+
researchers, builders, engineers and biologists using Mantis
2,000+
points in the largest single map
2.7M

Production, October 2026

Security and deployment

Your data stays yours.

Run Mantis where your data lives, under the access rules your organization sets.

Private Spaces
Keep unpublished and internal records in Spaces your organization controls, with built-in access controls.
Your environment
Deploy Mantis inside your own environment, so records stay in your storage.
Your models
Serve open models on your own hardware for data that cannot leave the building.
Your compute
Run analysis on your own cluster or on hosted workers.
Talk to us about deployment

Developers

Build on Mantis.

Your agents reach Mantis tools through MCP and the Mantis CLI, and the extension SDK adds panels, commands and Python actions to the workspace.

npm install -g mantisai-climantis setup # API key, Space and threadSetup completemantis create map ./papers.csv # embed and lay out a tableMap creation startedmantis use get_space_context # what agents read firstmantis use inspect --uri "mantis://map/<id>"

Research

From the Kellis Lab at MIT CSAIL

Professor Manolis Kellis leads the MIT Computational Biology Group at CSAIL and is a member of the Broad Institute of MIT and Harvard. His group helped map the regulatory circuitry of the human genome through the ENCODE, Roadmap Epigenomics and GTEx consortia, and charted Alzheimer’s disease cell by cell.

Mantis carries that program from the genome to everything an organization knows: learn one representation across thousands of dimensions, reveal the structure no single query can see, and trace every finding to its evidence.

  • ENCODE
  • modENCODE
  • Roadmap Epigenomics
  • GTEx
TalkManolis Kellis at CSAIL + Imagination in Action: AI, creativity, innovation, information
Presidential Early Career Award (PECASE)
2008
Alfred P. Sloan Foundation Award
2008
NSF CAREER Award
2007
MIT Technology Review TR35
2006

Latest

  1. Ask about Mantis answers live, from any page of the site.
  2. A new front page, opening on Mantis at work in its own workspace.
  3. Instant navigation: the sidebar stays put as you move between pages.
  4. Seven public dataset Spaces open to everyone.
  5. Continuum-1, Mantis’s own model, serves the Composer, with usage tiers.

Origins

From the genome to the organization

Mantis is made at MIT CSAIL, in Professor Manolis Kellis’s group. It extends a long-standing research program, using computation to reveal hidden structure in complex systems, from genomics and biological circuitry to everything an organization knows.

Made at

About Mantis ›

Work with us

Bring representation superintelligence to your team.

Start in the workspace today, or talk to us about your data, your deployment and your questions.