Learn
The Mantis documentation, arranged by what you want to do: understand the map, build a space, put agents on it, or extend the workspace. Each card opens the manual at mantis.csail.mit.edu/docs.
What is Mantis?
Mantis embeds a corpus, lays it out as a navigable map of meaning, and puts agents on that map with you.
Read the page →Embeddings and projection
How Mantis turns your rows into vectors, compresses them to two coordinates, and groups them into a labeled cluster tree.
Read the page →Read the map
What points, clusters, labels, boundaries and colors mean on a Mantis map, and why 2D distance is not the whole truth.
Read the page →Spaces and maps
A space is the workspace, a map is one projection of one dataset.
Read the page →Selections and bags
A selection is temporary and there is only one; a bag is a saved, named, colored set of points you can keep as many of as you like.
Read the page →Clustering methods
How Mantis builds, prunes, stores, and uses cluster hierarchies.
Read the page →Create your first space
Pick a source type, assign your columns, and let Mantis build the map.
Read the page →Source types
Every route into a Mantis space, what each one needs, and the limits that actually exist.
Read the page →Composer
The AI chat panel that reads your space, calls Mantis tools, and drives the map.
Read the page →Agent Studio and memories
Build a custom agent with its own knowledge and MCP tools, and manage what the Composer remembers about you.
Read the page →Inspection
The panel group that reads the data behind the points you selected on the map.
Read the page →Journeys
Build an ordered walkthrough of a space from saved vignettes, and share it by link.
Read the page →Mantis Coding
A Python notebook wired into the workspace, with the space’s points preloaded and cells that can drive the UI.
Read the page →Sharing and visibility
Who can open a space, how per-user access works, and how vignette links are controlled.
Read the page →Extensions overview
What a Mantis extension is: a host worker, sandboxed panels, and an optional Python backend.
Read the page →Take the tour
One session end to end, from a first look at the map to a saved bag and an annotated region.
Read the page →