Skip to main content
Open Source · SSPL v1

AI-Native DocumentDatabase, Reimagined

A standalone database with native vector search, time travel, Raft replication, and automatic hot/cold tiering. MongoDB compatible. 60-80% cost reduction. Zero application changes.

Get StartedView on GitHub
terminal
# Connect with any MongoDB driver — zero changes $ mongosh "mongodb://thermocline:27017/mydb" # Vector search, time travel, hot/cold queries — all native > db.articles.aggregate([{ $vectorSearch: { path: "embedding", ... } }]) // Native storage engine + cold Parquet, queried seamlessly

Why Thermocline?

The document database built for AI workloads, cost efficiency, and operational simplicity

MongoDB Compatible

Connect to Thermocline with any MongoDB driver. Same wire protocol, same MQL, same aggregation pipelines. Just change your connection string.

Native Vector Search

Built-in HNSW and DiskANN indexes for semantic similarity. $vectorSearch aggregation stage. Store documents and embeddings together.

60-80% Cost Reduction

Automatically tier historical data to cost-optimized object storage while maintaining full query access. Pay hot prices only for hot data.

Cloud Agnostic

Runs on any Kubernetes cluster with any S3-compatible object storage. AWS, GCP, Azure, on-premises — your infrastructure, your choice.

Features

Everything you need for an enterprise-grade AI-native document database

Query Federation

Transparently split, route, and merge queries across the hot Storage Engine and cold object storage in a single operation

Policy-Driven Lifecycle

Define archival rules by age, size, or custom predicates. Data moves automatically from hot to cold with zero downtime

Complete MQL Translation

Every MongoDB query operator, aggregation stage, and cursor operation works seamlessly on cold data via Parquet translation

Multi-Cloud Storage

Native adapters for AWS S3, Google Cloud Storage, Azure Blob Storage, and any S3-compatible endpoint like MinIO

Wire Protocol Compatible

Full MongoDB 4.4+ wire protocol implementation. Connect with any MongoDB driver — no SDK, no library, no changes

Security & RBAC

Native SCRAM-SHA-256 auth, TLS 1.3, granular RBAC for vector search and time travel, encryption at rest, and audit logging

Full Observability

Prometheus metrics, structured JSON logging, OpenTelemetry distributed tracing, and pre-built Grafana dashboards

High Availability

Raft consensus replication with sub-5s failover. Multi-replica, multi-AZ deployments with no single point of failure

Kubernetes Native

Production-ready Helm charts, Horizontal Pod Autoscaling, resource quotas, and GitOps-compatible deployment patterns

Backup & Restore

WAL-based point-in-time recovery for hot and cold data. Automated scheduled backups with cross-region DR support

Intelligent Caching

Seven cache tiers including block cache, vector index cache, MVCC version cache, and WAL read cache with adaptive eviction

Parquet Storage Format

Columnar Parquet format with configurable compression (zstd, snappy, gzip) for optimal query performance and storage efficiency

Architecture

A standalone AI-native database with automatic hot/cold tiering

ApplicationsMongoDB Drivers — Zero Changes RequiredWire ProtocolGatewayConnection Management / TLS / AuthenticationQuery CoordinatorAnalysis / Routing / Planning / Result MergingHOT PATHStorage EngineLSM-tree / WAL / RaftVector Search / MVCCTime Travel / TransactionsCOLD PATHMQL TranslatorQuery EngineS3 / GCS / Azure / MinIO

Community

Thermocline is SSPL licensed and community-driven

Contribute

Help build Thermocline. All contributions welcome — code, docs, and ideas.

Discussions

Ask questions, share ideas, and connect with the community.

Roadmap

See what we are building next and help shape the future of Thermocline.