Harper 5.0 is the most significant release in the history of the Harper project. The open-source core is now fully available under Apache 2.0. A new storage engine, RocksDB. And the entire platform has been re-architected to give AI agents the tightest possible surface area for building production-ready applications. Read the release notes here.
We sat down with Bari, Kris, and Ethan to break down what 5.0 means for developers, for the open-source community, and for the future of how software gets built. You can watch the full conversation below, or keep reading for the highlights.
A Unified Runtime, Now Fully Open
Harper has always been built on a single guiding principle: simplicity without sacrifice. That means collapsing database, cache, messaging, and application logic into one in-memory process so developers can focus on what they are building rather than managing how half a dozen services talk to each other.
With Harper 5.0 (release notes), that same runtime is now open source under the Apache 2.0 license. The full core is available at github.com/HarperFast/harper. Anyone can download it, build on it, extend it, fork it, or build a company around it. There are no caveats, no feature gating, no asterisks.
Alongside the open-source core, Harper Pro is now source-available under the Elastic License v2 (ELv2). Harper Pro adds the distributed capabilities that power Harper Fabric: replication, clustering, certificate management, and the infrastructure layer that enterprise customers rely on at scale. Source-available means developers can read the code, build it, and run it. The only restrictions are that it cannot be repackaged as a competing hosted service and licensing keys must be respected.
This is not a cosmetic change. Having the full source code accessible transforms how developers and AI agents interact with the platform. As Bari put it during our conversation, when agents can process the entire codebase in real time, the quality of what gets built goes up significantly. The condensed surface area, the minimal serialization, the reduced network hops already make Harper easier for an agent to work with. Full source visibility takes that further.
RocksDB: A New Native Storage Engine
Harper 5.0 introduces RocksDB as a native storage engine option alongside LMDB which was retained for legacy compatability. This is the headline architectural change in the release, and the engineering effort behind it has been substantial.
LMDB has served Harper well. It is known for highly optimized read performance, and the results that enterprise customers have seen on read-heavy workloads speak for themselves. But as Harper has expanded into new use cases, the need for a storage engine optimized for write-heavy, high-throughput scenarios became clear.
RocksDB fills that gap. Originally built on research from Google's LevelDB project and refined through years of contributions from Meta and others, RocksDB represents something close to a consensus on low-level storage engine design. Its continuous compaction system handles large, heterogeneous record sizes efficiently, eliminating the fragmentation challenges that can emerge with LMDB under heavy write loads. It also introduces a significantly more robust transaction handling model with better isolation, something Kris and the engineering team are particularly excited about.
Critically, developers do not have to choose one or the other at the outset. Harper 5.0 ships with full support for both engines. Existing deployments on LMDB will continue to work without modification, and migration scripts are provided for teams that want to move to RocksDB when the time is right. No application code changes are required to switch. The upgrade path from 4.x was designed to be as seamless as possible.
The new RocksDB JavaScript binding is also fully open source, available at /HarperFast/rocksdb-js. This is not a Harper-only tool. It was designed as a generally useful, high-performance key-value store for the broader Node.js ecosystem, continuing the same tradition as the LMDB-JS library that Harper has maintained for years.
Built for Agentic Engineering
The timing of this release is not accidental. AI coding agents are producing increasingly impressive prototypes. The gap is no longer in generating application code. The gap is in going from prototype to production.
Kris described this clearly in our conversation. An agent like Claude Code can whip up an impressive application from a prompt. But that application typically runs locally against seed data. The hard part begins when it needs to connect to a real database, wire up a caching layer, integrate messaging, configure authentication, and deploy across distributed infrastructure. Each additional system expands the agent's context, introduces new credentials, and multiplies the surface area for failure.
Harper collapses that entire chain. Because database, cache, messaging, and application logic all live in one process, an agent working with Harper does not need to orchestrate across multiple systems. The full context is already contained. And because 5.0 makes all of that source code available, agents can reason about the platform itself, not just the application being built on top of it.
The security implications matter here too. Every additional service an agent touches is another set of credentials it needs access to, another potential exposure point. A unified runtime means the principle of least authority is built into the architecture. The agent operates within one contained boundary.
The results are already visible internally. Team members without engineering backgrounds have gone from prompt to production-ready, geographically distributed applications with OAuth built in. That is not a prototype. That is a deployed service, running on Harper Fabric, ready to scale.
What Is in the 5.0 Upgrade
For developers on Harper 4.x, the upgrade to 5.0 should be straightforward for most workloads. It is a major version, so some breaking changes exist, but they have been minimized:
The Resource API for retrieving database records has been simplified. Objects returned from get are now cleaner and enumerable, with some previously attached methods moved to other access points. Application code sandboxing has been tightened for security. Code that spawns new processes will need to declare allowed processes in configuration. Migration scripts handle the LMDB-to-RocksDB transition without application code changes.
Enterprise customers on Harper Fabric are supported through use-case-driven QA that tests real customer workloads against 5.0, forward-deployed engineering support, and comprehensive documentation.
Speaking of documentation: Ethan has led a ground-up rewrite of the docs over the past several months. The new structure splits into two sections. Harper Reference provides deep technical documentation covering every aspect of the platform. Harper Learn offers prose-style, step-by-step guides ranging from getting started to advanced topics like cache table implementation and administration. Both are designed to serve human developers and AI agents equally well.
The Product Landscape: Harper, Harper Pro, Harper Fabric
For anyone new to the ecosystem, here is how the pieces fit together:
Harper is the open-source core, licensed under Apache 2.0. It is a fully capable data application platform: database, endpoints, subscriptions, real-time messaging, all in-process. Run it on a single node, build plugins for it, build applications on it.
Harper Pro extends the core with distributed capabilities: replication, clustering, certificate management, and the infrastructure layer needed to run at enterprise scale. It is source-available under ELv2. Teams that want to self-manage distributed deployments use Harper Pro.
Harper Fabric is the managed cloud platform. Deploy Harper without managing infrastructure. Pick a performance level, pick a region (or go global), and deploy. A free tier is available with no credit card required.
Applications and plugins built on the open-source Harper core work identically on Harper Pro and Harper Fabric. There is no re-architecture required as a project scales.
Get Started
Harper 5.0 is available now.
Run npm create harper@latest to scaffold a new project with the skills files and templates needed to start building agentically. The scaffolder sets up CLAUDE.md, AGENTS.md, GEMINI.md, and .cursorrules depending on the agent, so the AI has full context from the first prompt.
Explore the open-source core on GitHub. Deploy to Harper Fabric for free. Join the Discord community.
And stay tuned. This week we will be publishing deeper dives into the open-source story, the agentic engineering thesis, the product vision, and the technical architecture behind RocksDB. This is just day one.
.jpg)
%20(1).png){kind=icon}
.png){kind=icon}
.png){kind=icon}