Skip to main content

Thermocline: Built by StronglyAI, Open for Everyone

· 3 min read
Thermocline Team
Thermocline Team
Core Team

We're excited to share the story behind Thermocline and our commitment to the open-source community.

About​

Thermocline is developed and maintained by StronglyAI, Inc.. We built it because we needed a document database that could handle AI workloads, manage storage costs at scale, and provide capabilities like vector search and time travel natively — without vendor lock-in.

We're committed to keeping Thermocline fully open source under the Server Side Public License (SSPL) v1.

Why We Built This​

At StronglyAI, we build AI-native applications that demand more from their database than what's available today. We needed:

  • A database with native vector search for RAG and semantic retrieval
  • Automatic hot/cold tiering to control storage costs at scale
  • Time travel queries for debugging, compliance, and reproducible AI pipelines
  • Strong consistency with fast automatic failover
  • Full MongoDB compatibility so existing applications just work

Nothing available met all these requirements, so we built Thermocline from the ground up — a standalone database with its own storage engine, not a proxy or compatibility layer. Now we're sharing it with the community.

What This Means for You​

Whether you're building AI applications, managing time-series data, or looking for a cost-effective document database, Thermocline gives you:

  • Zero application changes — connect with any MongoDB driver, same queries, same aggregation pipelines
  • Native vector search — HNSW and DiskANN indexes with $vectorSearch, no separate vector database needed
  • MVCC time travel — query data at any historical timestamp for debugging, compliance, or reproducible retrieval
  • Massive cost savings — automatic hot/cold tiering with 60-80% storage cost reduction
  • Production-grade reliability — Raft consensus with <5s failover, ACID transactions with >50K TPS

Our Commitment​

As the maintainers of Thermocline, StronglyAI is committed to:

  1. Keeping the core open — SSPL v1, the complete database engine is open source
  2. Active development — we use Thermocline in production daily, so it stays maintained
  3. Community-first — accepting contributions, responding to issues, and building what users need
  4. Transparency — our roadmap is public, our decisions are documented

AI-Native from Day One​

Thermocline was designed with AI workloads as a first-class concern, not an afterthought:

  • Vector search — Built-in HNSW (in-memory, <50ms for 1M vectors) and DiskANN (disk-resident, up to 1 billion vectors) indexes
  • RAG pipeline integration — Change streams can automatically trigger embedding generation on insert/update
  • Hybrid search — Combine metadata filters with vector similarity in a single $vectorSearch query
  • Time travel for reproducibility — Query the exact state of your data at any historical timestamp, enabling reproducible retrieval results
  • Columnar analytics — Parquet-based cold storage enables efficient analytical queries for feature engineering and training data

Get Involved​

We welcome contributions from the community. Whether it's bug reports, feature requests, documentation improvements, or code contributions — every bit helps.

About StronglyAI​

StronglyAI, Inc. builds an end-to-end AI platform supporting apps, agentic workflows, intelligent operations and more. Thermocline is a core component of our platform, and we're proud to share it with the world.

Need managed Thermocline or enterprise support? Visit thermoclinecloud.com.

Questions? Reach out at support@strongly.ai or open a discussion on GitHub.