The portfolio
Open primitives, yours to fork
Each project works standalone and shares the same foundation. Fork one, run it on your own infrastructure, extend it. The list grows as more open up.
roxabi-factory
In constructionA self-hostable agent engine. Hub-and-spoke over a NATS bus, per-scope agent pools, and a worker fleet that plugs in over the wire.
How roxabi-factory works →
roxabi-plugins
Open sourceReusable Claude Code plugins. dev-core turns one command into the full dev lifecycle — frame, spec, plan, implement, review, ship.
How /dev works →
voiceCLI
Open sourceOne CLI for local speech synthesis and transcription. A single Markdown script drives every engine — Qwen3-TTS, Chatterbox, Whisper — with a NATS satellite for cross-host voice.
How voiceCLI works →
llmCLI
Open sourceServe local LLMs on your own GPU. llama.cpp + GGUF behind an OpenAI-compatible endpoint, hot-swap from a TOML catalog, one LiteLLM proxy for the whole LAN.
How llmCLI works →
imageCLI
Open sourceOne CLI for local image generation. FLUX.2, FLUX.1 (+ PuLID face lock) and SD3.5 behind one command — quantization and VRAM handled, a NATS satellite for cross-host rendering.
How imageCLI works →
roxabi-forge
Open sourceA Claude Code plugin that turns "explain this" into self-contained HTML — diagrams, galleries, decks, guides. Eight skills, one-command deploy.
How roxabi-forge works →
roxabi-boilerplate
Open sourceA production-ready SaaS starter. Bun + TurboRepo monorepo with auth, multi-tenancy, RBAC and email wired — plus a built-in Claude agent team.
How roxabi-boilerplate works →
On the way
Opening up- cortexmemory
- vaultknowledge