01
Literature review
Map a whole field at once and see where the work clusters, overlaps, and thins out.
Cognitive cartography · Built at MIT CSAIL
Mantis turns the collections, systems, and workflows an organization already runs into one space people and AI can reason in together. Born at MIT CSAIL, out of two decades of work making the human genome readable: the same problem at a different scale.
Documents, datasets, media, tickets, and the systems that hold them are gathered into one embedded space, where related material sits together whether or not anyone filed it that way. People and agents work from that single view, reason over the same evidence, and act on the same systems.
Mantis reads open corpora, institutional archives, and your own private collections through the systems that already hold them, then maps the result into one environment your teams and agents can work in.








01
Map a whole field at once and see where the work clusters, overlaps, and thins out.
02
Explore genomics, single-cell, and molecular collections without writing a pipeline first.
03
Index the software an organization already runs and see which parts of it actually relate to each other.
04
Put a live Mantis map inside your own product, notebook, or internal tool, where the work already happens.
05
Bring images, audio, and video into one searchable space alongside your text.
06
Let agents read the map, propose next steps, and cite the points behind every answer.
07
Work directly on arXiv, PubMed, and open government data without staging it yourself.
01
Dimension-reduced embeddings laid out so proximity means similarity. The shape of a collection is visible before you query it.
02
Unsupervised clusters, labelled automatically and nested. Zoom from the whole corpus down to a single record.
03
Search by meaning or by exact string, then narrow with categoric, numeric, and date filters on the same map.
04
Save any subset of points as a named bag, and save the whole view as a thread you can return to or share.
05
Every summary and every agent conclusion links back to the points that support it, so a decision can be checked against the record it came from.
06
Agents read the same map your team does, propose the next action in real time, and extend the workspace through a plugin system.
07
Shared editing, customizable layouts, and a coding widget for computing directly on the data behind the map, inside the workflow rather than beside it.

Talks, walkthroughs, and releases, as they happen.