Academic
OpenAlex literature search
A query and optional filters on author, institution, concept, open access and document type, resolved into a space of 50 to 5,000 papers.
/new-space/academic_synthesis
Cognitive infrastructure on the stack you already run
Mantis reads the public corpora, files, repositories, databases and APIs an organization already depends on, embeds them with the model provider it already trusts, and lets its own developers extend the workspace. This page lists what connects today.
Public sources
Fourteen public APIs behind one interface, each with its own query, limit, date and sort controls. Search, then turn the result into a space.
- Library of Congress
Gather from where the data lives
Each route is a creation flow under /new-space. The path is shown so a reader can find it in the workspace.
Academic scraping
Open-access PDFs on a topic
A keyword list and a date window pull genuinely open-access papers, summarised on the Five Facets model, up to 1,000 at a time.
/new-space/academic_scraping
Web search
Crawl the web into a space
A search term with six to ten crawl iterations, or a list of specific URLs one per line.
/new-space/websearch
GitHub
Map a repository
A repository URL becomes a map of its files and structure. Pick the embedding and chat services and models right on the form.
/new-space/github_synthesis
Your database
SQL into a map
The Multimedia page also takes a database connection and a SQL query, and a natural-language search bar that returns a table you can turn into a space.
/new-space/MultiMedia
Periodically updating
An API that keeps tracking its source
GET or POST with headers and parameters, a records path for nested responses, a type per column, and an ingest interval from one second to ten years.
/new-space/api_ingestion
Bring your own files
Seven upload routes. There is no global row limit: spaces of millions of points are normal, and the map renders them through a level-of-detail budget rather than refusing them.
Spreadsheets
CSV, XLSX or JSON, with a type assigned to every column: title, semantic, numeric, categoric, date, links, image, geospatial, coordinates or vectors.
.csv · .xlsx · .json
AI Assistant
Drop a spreadsheet and let Mantis decide which columns carry titles, meaning and metadata.
.csv · .xlsx
Text and documents
Files or pasted prose, cut into points per chapter, page, section, paragraph, sentence, or a segmentation prompt you write.
.txt · .pdf · .doc · .docx
Audio
One or more recordings, up to 100 MB each, mapped from their content.
MP3 · WAV · M4A
Video
A lecture or a talk becomes a map of its transcript, with a player and transcript panel wired to the points.
MP4 · MOV · WebM · YouTube URL
Molecules
SMILES strings laid out by structural similarity on Morgan fingerprints, with optional property computation and GPU acceleration.
.csv of SMILES
Single-cell data
Mantis inspects the file, proposes a set of maps over one shared cell dataset, and materialises the ones you enable.
.h5ad · HDF5
Reference
Every source type, with its limits
The documentation page this list is drawn from, including the row floors and ceilings that actually exist.
Read the reference →Models and services
The embedding, reduction, clustering and labelling configuration is generated from the backend’s own schema, so the provider is a choice on the form rather than a rebuild.
Provider
OpenAI
Hosted embedding and chat models for spaces that want the quickest path to a map.
Provider
vLLM
Open models served on your own hardware, for spaces whose data cannot leave the building.
Provider
Hugging Face
Open embedding models, including multilingual ones, pulled from the Hub.
Build on the workspace
Extensions, agents, and a CLI, each documented and each in use inside Mantis today.
Extensions
Panels, commands and background behaviour
A host worker, sandboxed panels that talk to the workspace through window.mantis, and an optional Python backend, packaged as one manifest.
Read the docs →Python backend
Backend actions with real limits
Declare actions, invoke them from browser code, and run them through the Mantis API in a container with timeouts and no network unless the manifest asks.
Read the docs →Agents
Custom agents with MCP tools
Agent Studio builds an agent with its own knowledge and Model Context Protocol tools, and manages what the Composer remembers.
Read the docs →CLI
Any coding agent, from the terminal
Give Claude Code, Codex or your own agent access to a workspace without an editor plugin.
Read the docs →Notebooks
Prototype a panel in place
Write Python vertical cells, add one UI cell, test the panel in the space, then export a .mantisx package.
Read the docs →Developer portal
API keys and the workflow engine
Keys for the CLI and the API, and the documentation for the workflow engine behind automated spaces.
Read the docs →