— AI-Native Data Platform · 2026
A document database, vector search, and stream processing — unified behind one query language. Power the next generation of AI applications without stitching together half a dozen services.
// Find product reviews semantically // similar to the user's query. const results = await db .collection("reviews") .aggregate([ { "$vectorSearch": { index: "plot_embedding_idx", path: "embedding", queryVector: embed(query), numCandidates: 400, limit: 5 } }, { "$project": { title: 1, score: { "$meta": "vectorSearchScore" } } } ]) .toArray();
The same database across AWS, GCP, and Azure — with cross-cloud replication built in.
Store embeddings next to your documents. Query them with the same aggregation pipeline.
Kafka-compatible streams that land in your collections — without the cross-system glue.
Cloud-managed, self-hosted, or air-gapped on-prem — every deployment runs the same query language and the same SDKs.
Fully managed across AWS, GCP, and Azure. Autoscaling, point-in-time recovery, and queryable encryption — without the ops team.
The source-available core. Run MongoLeaf on your own hardware, in your own VPC, with the developer tools we use ourselves.
Air-gapped, FIPS-compliant, and certified for the most demanding workloads. Run anywhere, including disconnected environments.
Documents that map to objects. Indexes that learn your queries. SDKs in every language you care about. MongoLeaf is the database engineers actually want to write against.
Objects in your code map 1:1 to documents in the database. No ORM, no join hell, no impedance mismatch.
Semantic search lives inside the same aggregation pipeline as your filters. No separate vector store to operate.
Three-node replica sets, zone-aware sharding, and read preferences that let you push reads to the closest region.
Encrypt sensitive fields client-side and still run equality and range queries on the encrypted ciphertext.
Embeddings live next to the documents they describe. One query language. One operational footprint. One source of truth.
Run vector search, then chain a $project, $lookup, and $match — all in the same pipeline. The hard part of RAG isn't the embeddings; it's keeping them in sync. MongoLeaf moves that problem into the database.
// Hybrid: vector + metadata + rerank [ { "$vectorSearch": { index: "product_emb_v2", path: "embedding", queryVector: embed(query), filter: { in_stock: true, price: { $lt: 200 } }, numCandidates: 400, limit: 20 } }, { "$lookup": { from: "reviews", localField: "_id", foreignField: "product_id", as: "reviews" } }, { "$project": { name: 1, score: { "$meta": "vectorSearchScore" }, avg_rating: { "$avg": "$reviews.rating" } } } ]
Free, self-paced courses taught by the engineers who built the product. Certify when you're ready.
You know JOINs. Now learn $lookup. A 4-hour bridge for engineers coming from Postgres or MySQL.
Build a semantic search index from scratch. Includes a free Atlas cluster for the duration of the course.
Design read-preference strategies that route queries to the nearest replica without sacrificing consistency.
Ship a working retrieval-augmented chatbot. Embedding pipeline, hybrid search, rerank, eval harness — all of it.
$match, $lookup, $group, $bucket, $facet — the seven stages you'll use in 95% of your queries.
Encrypt sensitive fields client-side and still run equality + range queries against the ciphertext.
A free forever tier for evaluation. Pay-as-you-go credit beyond that. No seat minimums, no procurement gates until you genuinely need them.
$0
For evaluation, prototypes, and personal projects. Free forever — no credit card required.
$0.012/hr · pay-as-you-go
Right-sized for production workloads. Pay for the compute and storage you actually use.
$0.08/hr · starting
Isolated compute, customizable hardware, and the controls enterprise procurement teams expect.
Source: DESIGN.md. Dual-mode identity. Pill CTAs (r-full). 12px rounded cards. 3-tier pricing with mint-bordered featured tier.
Brand — the unmistakable signal
Teal — the dark canvas
Category accents — course tags ONLY, never system colors
Ink ladder — text on light surfaces
Euclid Circular A is the contemporary geometric brand face. Inter is the open-source substitute used here. Source Code Pro for every code mockup (JetBrains Mono sub). Weight 500 across display; weight 600 reserved for buttons and small emphasis.
Every CTA is a pill (r-full). Every card is r-lg (12px). Course category tags use r-sm (6px) — the only sharp-cornered surface in the system.
Buttons on white canvas
Buttons on deep-teal hero band
Inputs — focus is a 2px green-dark ring
Cards — base / featured / dark
White canvas + 1px hairline. The default container. Flat by default — no shadows on documentation surfaces.
Mint surface + 2px brand-green border. The signature featured tier — never green text on green.
Deep teal fill for product showcase cards on the hero band. White text, no border.
Badges & tags — category palette
Tabs — pill style (pricing)