Local-first AI agent infrastructure, developer tools, and automation systems.
I'm Phnix, a self-taught developer in Toronto building practical AI tooling that runs close to the user: browser agents, local memory systems, multimodal analysis, MCP servers, terminal workflows, and evaluation tools.
I use AI as a coding accelerator, but I review, test, debug, and own the code I ship. My strongest work sits at the intersection of Python backend engineering, local ML workflows, developer tools, and agent infrastructure.
I care about tools that are:
- Local-first and privacy-aware
- Useful from the terminal, not just a dashboard
- Testable enough to trust during long-running agent work
- Built for real websites, real files, and real user workflows
- Small enough to run on personal hardware, but structured enough to grow
| Project | Focus | Signal |
|---|---|---|
| Blackreach | Autonomous browser and research agent with Playwright, DOM observations, memory, and resumable sessions | Python, FastAPI, Playwright, agent loops, tests |
| Huginn | Self-hosted scraping, crawling, and extraction API for local AI research workflows | Python, crawling, extraction, REST APIs, local-first design |
| Velqua | Local-first memory proxy for Ollama and LLM apps with transparent context injection | Python, proxy design, LLM memory, desktop workflows |
| JMD | Markdown format and tooling for LLM-assisted fiction, annotations, lore, revisions, TUI, and HTML export | Parser design, CLI/TUI, VS Code extension, creative tooling |
| Rigr | Agent evaluation and regression testing toolkit for production AI systems | Evals, test harnesses, baselines, audit-friendly output |
| commit-critic | AI-powered CLI for reviewing and generating Git commit messages from staged diffs | Python CLI, Git plumbing, LLM providers, challenge project |
| Claude Voice | Local TTS voice mode for Claude Code with word highlighting and no cloud APIs | Python, terminal UX, local TTS, developer workflow |
| deep-video-watcher | Video perception and editing intelligence for multimodal analysis | FFmpeg, Whisper, scene detection, structured comprehension |
Current upstream PRs:
- MemPalace/mempalace #1569: chunk-size enforcement before embedding upsert
- dimensionalOS/dimos #2092: AprilTag 3D detector with solvePnP TF transforms
- Making local agents more reliable across browser, memory, audio, and video workflows
- Turning project prototypes into polished, documented, installable tools
- Building evaluation loops so agent behavior can be tested instead of guessed
- Keeping the stack practical for a MacBook plus local Linux hardware setup
Languages: Python, TypeScript, Go, Rust
AI/ML: Ollama, OpenAI-compatible APIs, Whisper, local TTS, multimodal pipelines
Backend and tooling: FastAPI, SQLite, ChromaDB, Playwright, MCP, CLI/TUI apps
Media and automation: FFmpeg, Demucs, browser automation, GitHub tooling
Engineering habits: tests, reproducible local workflows, clean READMEs, narrow PRs
I'm looking for software engineering work where this background is useful:
- AI tooling engineer
- Python backend engineer
- Developer tools engineer
- Browser automation engineer
- Applied AI / agent infrastructure engineer
- Local ML or automation-heavy product engineering
Best fit: small teams building useful AI infrastructure, developer tools, automation systems, or local-first products.