Skip to content
View RemiPelloux's full-sized avatar
10 followers · 16 following
  • France

Achievements

Achievement: StarstruckAchievement: Pair Extraordinairex3Achievement: Pull Sharkx3Achievement: YOLOAchievement: Quickdraw

Achievements

Achievement: StarstruckAchievement: Pair Extraordinairex3Achievement: Pull Sharkx3Achievement: YOLOAchievement: Quickdraw

Block or report RemiPelloux

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
RemiPelloux/README.md
Rémi Pelloux, CTO at OpenPro in France. Building AI coding workspaces, local-first native apps, Rust systems tools, and leading engineering at OpenPro.

OpenPro LinkedIn Sponsor GitHub followers

OpenVision  ·  Open source  ·  At OpenPro  ·  Activity

I build software that people actually run: AI tools that work with your own models, native apps that keep data on your machine, and systems tools without hidden daemons. During the day I lead engineering at OpenPro. The rest of the time I ship in public.


👁️ Latest: OpenVision

OpenVision: an editor where an AI model suggests code that is accepted with Tab, next to a blinking eye logo and a bring-your-own model provider list.

An AI coding workspace where you choose the models. OpenVision brings together a project explorer, editor, terminals, Git and worktrees, AI agents and Tab autocomplete in one desktop app. Point it at any OpenAI-compatible API or a server on 127.0.0.1, and it discovers the available models. You can pick a separate model just for autocomplete.

  • No account and no login. Built-in analytics and crash uploads are disabled at build time.
  • Agents with approved tools. Agents read files, propose edits and run tools only after you approve them. They also get MCP, project rules, skills and plans from .openvision/.
  • A portable Windows app. Download the zip, extract it, and double-click to run. You don't need to set up Node or npm.
  • Honest progress. The roadmap tracks acceptance stories, and only the ones that pass are marked verified.

Download v0.2.0-preview.6 →  ·  Read the docs  ·  Electron TypeScript MCP · development preview, tested on Windows x64


🧪 Open source

OpenPlod: a recorder sends audio over Bluetooth, the waveform plays, and transcript lines become a Markdown document. OpenNative: a gaming handheld showing a stable frame-time graph, next to the Wine, DXVK, VKD3D, Box64 and FEX runtime stack.

OpenPlod  stars

A local-first vault for your Plaud Note Pro recordings. OpenPlod imports audio directly over Bluetooth, keeps the original files, and turns them into transcripts, Markdown documents and AI chats that cite their sources.

  • Transcript Studio: speaker editing, version diffs, chapters and subtitles
  • Transcription with Mistral/Voxtral, OpenAI or AssemblyAI, plus local Ollama for text
  • REST API and a read-only MCP server, so your agents can query your recordings

Tauri Rust TypeScript · macOS / Linux / Windows · experimental

OpenNative  stars

Windows games on Android gaming handhelds. OpenNative combines Wine/Proton, DXVK, VKD3D, Mesa, Box64 and FEX in an interface built for controllers.

  • Warm, shared Wine prefixes that launch quickly, with per-game profiles
  • Shader caches tracked per game, safe cache invalidation, and a performance view on a second screen
  • APK updates checked against their signature and checksum. FPS gains are claimed only after measurement

Kotlin C++ Android ARM64 · validated on AYN Thor · GPL-3.0
Containust: a terminal runs ctst pull, plan and run with offline mode, and an api component connects to a db component without a background daemon.

Containust  release

A container runtime written in Rust, with no background daemon. Each ctst command talks to the backend directly and writes file state. You describe stacks in a small declarative language (.ctst: COMPONENT, CONNECT, EXPOSE). Images are pulled by pinned digest, and --offline blocks network access before anything connects, so it works on air-gapped machines. On Linux it uses namespaces and cgroups v2 directly. On macOS and Windows it runs through QEMU. It also ships a Rust SDK for the parser, the graph and the runtime.

Rust Linux namespaces cgroups v2 OCI · GA 1.2 · source-available, commercial license

More public work
Project What it is Stack
OpenSw Android ARM64 Switch emulator with Vulkan rendering, per-game profiles and AYN Thor tuning C++ Vulkan
agent-mangaka-forge AI manga-page skill pack with persistent character memory for consistent characters Python
OpenAgents Self-hosted, multi-provider AI agents with memory and MCP, forked from Hermes Agent Python MCP
LumaSync Local Philips Hue ambilight driven by fast screen capture Rust Tauri React
OpenPro Connector SDK Sync ATS jobs into OpenPro through PHP, HTTP or MCP PHP
OpenWhistle SDKs OpenWhistle API clients for five languages JS Python PHP Rust Go
BurpHunterTK Burp Suite extension that uses LLMs to triage high-impact web vulnerabilities Python
delonghi-coffee-api Local REST API for De'Longhi machines, built on the reverse-engineered ECAM protocol Python FastAPI

🏗️ At OpenPro

OpenPro architecture sketch: web, iOS and Android clients and ATS integrations talk to a PHP and Symfony core platform, which uses an AI layer in Python with RAG and MCP, and PostgreSQL and Redis, on AWS with Docker and CI/CD.

OpenPro is a French professional network for people, jobs and companies. It offers video CVs, AI matching, real-time messaging and Mia, our AI assistant, on web, iOS and Android. The code is private, but this is the engineering I'm responsible for as CTO:

  • Product platform. A PHP/Symfony core with TypeScript/React on the web and apps for iOS and Android, all shipped from one roadmap.
  • Applied AI. Matching, retrieval (RAG) and the Mia assistant, built with Python and connected through MCP.
  • Open integrations. ATS connectors through a public SDK over HTTP and MCP, so partners don't need custom work.
  • Infrastructure. PostgreSQL and Redis on AWS, containerized with Docker and released through CI/CD.
How I like to engineer
  • Useful workflows first: from the first install to everyday use.
  • Your data stays yours: local-first storage, portable formats and practical APIs.
  • Measure before claiming: latency, recovery and acceptance gates, written down.
  • Credit upstream: forks and integrations name the projects they build on.

📈 Activity

Rémi Pelloux's GitHub contribution calendar, showing daily activity over the last year.

Contribution calendar, refreshed daily by a GitHub Action

Building useful software across AI, products and systems. Thanks for stopping by.

All animations are self-contained SVGs with no scripts or network requests, and they stop when your system asks for reduced motion.

Pinned Loading

  1. Multi-Threaded-Prime-Number-Calculation Multi-Threaded-Prime-Number-Calculation Public

    Multi Threaded Prime Number Calculation

    Python

  2. BlackJackAI BlackJackAI Public

    The BlackPlayer class is an implementation of a reinforcement learning algorithm known as Sarsa(λ)

    Python

  3. EatIt EatIt Public

    EatIt Maquette

    PHP

  4. TextRPG TextRPG Public

    A simple text RPG

    Python