I build local-first AI tools, agent workflows, speech integrations and desktop experiments. On my Omarchy desktop, the window manager and the tooling around it are part of the project.
A curated mix of original projects and substantial work on upstream foundations, with attribution in the linked repositories. Status labels refer to public main, not private branches or unmerged drafts.
| Project | Purpose | Maturity |
|---|---|---|
| mcp-openapi-proxy | Turns OpenAPI-described REST endpoints into MCP tools over Streamable HTTP or stdio. | Established, versioned. v0.4.0 tagged source, PyPI package and published GitHub Release. |
| operating-swarm | Coordinates model APIs, local agent CLIs and remote harnesses through a shared API, CLI and WebUI. Evolved from OpenAI's experimental Swarm. | Experimental alpha. A model-and-tool workflow is verified; integration coverage varies, and the published package lags source. |
| funes-vision | Local watched-folder browser for JPEG snapshots and grouped visits. | Public alpha (synthetic-only). Public main demonstrates 10 fictional JPEGs grouped into 7 visits, without live cameras or model inference. Separate private deployment (verified 23 October 2026): active camera-ingestion pipelines and gallery/API with YOLOv4-tiny (OpenCV) object detection; optional deep vision/LLM decisions currently disabled. |
| mcp-flowise | Lets MCP clients discover and run Flowise chatflows, with authentication and flow filtering. | Experimental. Core chatflow integration verified. Verification details. |
| openai-f5-tts | Self-hosted speech for AI assistants and voice apps. Independently developed OpenAI-style API wrapper around upstream F5-TTS, with selectable voices, support for your own trained or fine-tuned checkpoints, sentence-level PCM streaming, and estimated timestamps. | SOHO-oriented community beta, verified on one machine. NVIDIA GPU speech generation verified in Docker Desktop on 20 October 2026: the author's privately trained Scarlett (v0, not distributed) and upstream Emilia (v1), non-streaming WAV/MP3, incremental sentence streaming, estimated timestamps, and authentication. Generated speech was transcribed correctly. Intended for trusted, sequential use; simultaneous streaming and cancellation during inference remain unverified. Upstream pretrained weights are noncommercial; custom-checkpoint rights depend on their provenance. |
| openai-kokoro-tts | Independently developed OpenAI-compatible API wrapper around the upstream Kokoro TTS model; not affiliated with OpenAI or the Kokoro authors. | Community beta, verified on one machine. CPU and NVIDIA CUDA speech generation verified in Docker Desktop on 20 October 2026, including WAV/MP3 output and bearer authentication. CPU image contains no CUDA packages. AMD and other accelerators remain unverified; see the verification report. |
| llama-stack-bridge | Nginx/njs adapter for Llama Stack model lists and non-streaming chat completions. | Legacy prototype. Current compatibility, streaming and deployment unverified. |
| waltop-osx | Userspace macOS shim that restores pen pressure for the WALTOP Slim Tablet 12.1" (USB 172f:0031), whose firmware mislabels its pressure field. | Working on one machine. Verified drawing with pressure in Krita on one Intel Mac (macOS 14.8.7) with one tablet; Apple Silicon, other macOS versions and other apps untested. |
A synthetic preview is not a live camera or inference test, and an unmerged draft is not a release. These labels describe public evidence, not production certification.
These are broad areas of ongoing work, not claims that every prototype is released or production-ready. Individual projects vary in maturity; public repositories above provide examples where available.
| Area | Focus |
|---|---|
| Cross-Platform Desktop Experiences | Omarchy-inspired desktop workflows across Linux, macOS, Asahi Linux and Windows, including touch, pen and accessible interaction. |
| Edge AI & Hardware Acceleration | Local multimodal AI on consumer hardware, including Snapdragon NPUs, NVIDIA Tegra and AMD GPUs; accelerator claims require measured evidence. |
| Embodied AI & Interactive Systems | Robotics, speech, computer vision, Kinect and alternative controllers for physical and expressive interaction. |
| Smart Home & Intelligent Automation | Home Assistant, presence sensing, local media workflows and AI-assisted home operations. |
| Parametric Design & Digital Fabrication | Functional OpenSCAD design, practical 3D printing, mechanical parts and AI-assisted CAD workflows. |
| AI Infrastructure & Developer Tools | Agent orchestration, MCP interoperability, API gateways, local model services and developer tooling. |
Development work may be private or experimental. A theme here is not a statement of public availability or hardware certification.