About Gluonify

The project

Gluonify is a personal idea, born of my own experience: I had seen OpenShift rollouts turn into long projects that consumed a lot of people's time. I thought it could be simpler and faster, and that people without particular skills could deploy their ideas with an AI.

How we work · Claude Code plans and reviews, Edge carries out the small tasks, and everything is tested by real end-to-end runs.

The timeline

  1. Start of the project.
  2. Move to the gluonify GitHub organisation, one repository per component.
  3. gluonify-source, the starter application for customers; first version of the Synchrotron console.
  4. Sleep mode (scale-to-zero) tested end to end; sea, the repository of signed binaries.
  5. Gauge 0.1, the MCP server that drives an application; gDown compared with five other graph databases.
  6. Release 0.5.0: everything under the brand; gDown joins the platform; quark, the Linux lab, runs a first orchestrator; Edge, the local model, gets its first tasks.
  7. Benchmark against seven databases, measured at night only; Fermion, Crystallon and the shared UI kit start.
  8. Platform roles gluonify:*; single-node development cluster; nightly incremental benchmark of gDown.
  9. SDK 0.6.0 to 0.7.1 published on sea; Prism, the customer console; application-to-service tokens.
  10. Prism's MCP for customers' AIs; x86-64 v2 and v3 builds; SonarQube rules GN01 to GN11.
  11. SDK 0.7.2; this page.

The development team

  • David CABILLIC — Founder and lead developer

    Founded Gluonify, sets the plan and the architecture, directs Claude Code and decides what is released.

    LinkedIn

    • Claude Code — AI coding agent (Anthropic)

      Plans, writes, reviews, tests, commits and pushes the code.

      • Claude agents in parallel — Subagents in worktrees

        Write code in separate worktrees while the end-to-end tests run; Claude Code merges and tests their work.

      • Ollama · Qwen3-Coder 30B — Local model, called Edge

        Small, fully specified tasks: translations, documentation, tests modelled on an existing one. Claude Code reviews every result.

      • SonarQube — Static analysis

        Checks the code with the gluonify-native rules GN01 to GN11, which detect GraalVM native image pitfalls.

      • x86_64 — Linux lab (quark)

        A 2013 Mac Pro under Debian 13: x86_64 end-to-end tests, the nightly benchmark and SonarQube.

      • arm64 — Development machine (Apple Silicon)

        A MacBook Pro M4 Max under macOS: development cluster, arm64 builds and the nightly benchmark.

Claude and Claude Code are trademarks of Anthropic PBC; Ollama is a trademark of Ollama, Inc.; SonarQube is a trademark of SonarSource SA. Logos are shown to identify the tools used, without endorsement.