Persistent documents
Ideas, tasks, events, places, notes, and other app data remain available after refresh and across devices.
BUILT INTO EVERY APP
MONGODB ATLAS
+ AN LLM.
MongoDB Atlas stores the app data. The builder selects and connects the right LLM. Both are ready when the app is deployed.
MONGODB ATLAS
Store, find, and update app data.
Search and discovery
MongoDB Atlas Search finds exact words and phrases. MongoDB Atlas Vector Search finds similar meaning. Hybrid search combines both result sets.
Live and changing data
Change streams update shared boards, feeds, maps, polls, dashboards, and trends as saved data changes.
MONGODB ATLAS FEATURES IN USE
See what powers the platform and each app.
Documents + indexes
Document storage, schema validation, relationships, aggregation, TTL, time-series, and geospatial indexes support the apps that need them.
Hybrid search
MongoDB Atlas Search and MongoDB Atlas Vector Search retrieve keyword and meaning-based candidates. Native rank fusion or reciprocal rank fusion combines them.
Voyage rerank-2.5
Every platform-managed generated-app search and the Build Assistant rerank MongoDB Atlas hybrid candidates with Voyage AI rerank-2.5 by default. If reranking is temporarily unavailable, the search safely keeps the MongoDB Atlas hybrid order.
Build Details
Each published app links to its own Build Details page with the MongoDB Atlas operations and advanced features that were declared and verified, plus its actual search and reranker status.
These are capabilities used across the Build Lab platform; an individual generated app only uses the subset shown on its Build Details page. When an app uses the platform-managed search capability, hybrid MongoDB Atlas Search + MongoDB Atlas Vector Search and Voyage AI rerank-2.5 are the defaults.
THE LLM
Use saved data to produce a useful result.
Summarize and explain
Ground a response in a focused set of documents already saved by the app.
Classify and compare
Group information, find patterns, compare choices, or suggest a next step.
Draft and transform
Generate a useful result, show it to the user, and save it to MongoDB Atlas when the app needs it.
PDF RAG
Upload, retrieve, then generate.
1
Upload: choose a non-sensitive PDF up to 20 pages and 3 MB. Each app accepts 25 PDFs.
2
Embed: the platform renders each page and embeds the page image with Voyage multimodal-3.5.
3
Retrieve: MongoDB Atlas Vector Search finds the pages most relevant to the question.
4
Generate: the built-in vision LLM reads those pages and answers with page citations.
Scans, charts, figures, tables, and selectable text work without OCR. At the kiosk, choose whether uploads are private to each browser or deliberately shared with everyone using the app. Never upload credentials, private company information, payment cards, government IDs, medical records, biometrics, or other sensitive files.
MONGODB ATLAS RECORD LIMITS
15 starter records by default.
15 records
When starter data improves the experience, the builder writes 15 sourced public records to MongoDB Atlas. A user-created collection may start empty.
100 records
The initial deployed app cannot start with more than 100 records.
500-record ceiling
After launch, the app can grow through user input or user-triggered LLM actions, up to 500 records. The separate PDF allowance is 25 files per app.
AFTER DEPLOYMENT
Public web research happens during the build.
1
The builder may research public information before the app is deployed.
2
The live app cannot search the open internet to collect new documents.
3
Users can add ordinary app data through the interface.
4
Users can trigger the LLM to create or transform a result and save it to MongoDB Atlas.
BUILD ASSISTANT