AI TOOLS &
ENGINEERING STACK_
An unfiltered, battle-tested breakdown of the autonomous coding agents, custom local inference routers, private speech pipelines, and automation engines powering my daily engineering workflows.
The Production AI Arsenal
Categorized by functional role—spanning autonomous coding environments, proprietary proxy gateways, local offline models, and multimodal pipelines.
AI Coding Agents & Cloud Teammates
Interactive IDEs, terminal agents, persistent assistants, and remote cloud teammates used for software delivery and operational work.
T3 Code
Desktop coding-agent environment that pairs with the T3 mobile app through T3 Connect, keeping approved terminal and repository work accessible across phone and workstation.
Grok Bot
xAI cloud-agent desktop with named Bots, a shared cloud computer, persistent routines, connected apps, and phone approvals for work that continues beyond the local session.
Cursor
Primary visual coding workspace for repository-aware edits, multi-file changes, previews, MCP integrations, and human-in-the-loop engineering review.
Antigravity IDE
Google's agent-oriented IDE for long-running code changes, browser-assisted tasks, architecture work, and Gemini-backed planning across complex workspaces.
Antigravity CLI (agy)
Google's terminal counterpart to the Antigravity IDE, used for repository changes, command execution, and persistent agent sessions without switching to the desktop app.
OpenCode CLI
Terminal-first coding agent wired to OpenCode Zen, Kilo, OpenRouter, the Klytron router, and MCP tools for fast provider-independent implementation.
Pi Coding Agent
Extensible terminal agent with provider routing, reusable extensions, and session-aware workflows; managed through pi.dev and connected to the local inference router.
Hermes Agent
Self-hosted persistent agent with cross-session memory, browser automation, scheduled workflows, and chat, desktop, TUI, and messaging surfaces.
Codex CLI
OpenAI terminal agent available through a ChatGPT session and the local router, providing focused implementation, code review, and an independent second opinion.
GitHub Copilot CLI
Verified terminal lane for shell scripting, repository tasks, and GitHub-connected workflows when another agent or provider lane is unavailable.
Freebuff
Zero-configuration terminal and cloud fallback for continuing small and medium implementation tasks when preferred provider quotas are exhausted.
Devin
Installed specialist environment for isolated autonomous experiments and Markdown or IDE preview work, retained without depending on its unreliable free inference tier.
OpenClaw
Previous always-on personal assistant gateway for Telegram, Discord, canvas hosting, and clawhub skills; retired in November 2026 and superseded by Hermes Agent, with its setup guides and npm package kept for reference.
Agent Zero
Docker-isolated recursive agent that writes, runs, and tests its own code in sub-containers behind a local web UI; evaluated for autonomous coding isolation and not deployed, with that lane currently covered by OpenCode and Hermes in Docker.
Cursor Agent CLI
Terminal form of Cursor for repository work, code review, and print-mode automation outside the editor, running on the same subscription and also wired into Hermes as an ACP backend for unattended terminal sessions.
Custom In-House AI Infrastructure
Purpose-built routing, state-management, provisioning, and autonomous systems that make the wider AI stack reliable and recoverable.
Klytron Inference Router
Custom OpenAI-compatible gateway with 530 catalogued models, eight virtual failover combinations, 82 upstream integrations, 545 combo lanes, pre-first-byte circuit breakers, quota-aware rotation, and client setup.
Klytron Workstation Setup
Automated Linux provisioning that installs and versions AI CLIs, desktop agents, browsers, runtimes, and restore paths from one idempotent setup script.
Klytron AI Trading Bot
Multi-model crypto trading engine with dynamic position sizing, sentiment analysis, liquidation guards, and async timeout architecture in Docker.
AI Session Export Tool
Forensic export engine for conversations and decision traces across Antigravity IDE and CLI, Devin, OpenCode, Cursor, Claude Code, Freebuff, Copilot CLI, Pi, and Codex.
IDE Reset Tool
Registry-driven utility for safely resetting or clearing AI IDE and agent state while preserving binaries, credentials where required, and generating restorable backups.
Frontier & Local Model Access
Frontier model pools and self-hosted open-weight models selected for reasoning, multimodal work, tool use, and offline privacy.
Gemini 3 Flash
Google's fast multimodal model used through Antigravity and hosted inference lanes for architecture planning, long-context analysis, and agentic coding workflows.
Claude Sonnet 4.6
Anthropic model used through a pooled Antigravity lane for rigorous logic, architectural review, and high-precision code transformations.
Ollama (Self-Hosted)
Local model runtime currently serving Qwen 2.5, Qwen 3.5, and DeepSeek Coder models for private, air-gapped work with no external API dependency.
Browser Agents & Web Automation
Authenticated browser surfaces and automation protocols that let agents work safely inside real web applications.
BrowserOS Neo
Privacy-first agent browser with a native HTTP MCP server, persistent profiles, tab isolation, and signed-in app integrations for reliable browser work.
Chrome CDP & Playwright
Chrome DevTools Protocol and Playwright attach to live Chromium instances for deterministic DOM interactions, browser testing, and agent navigation.
Cursor + Playwright MCP
Playwright's MCP server exposes browser navigation, snapshots, clicks, and form controls directly to the coding workspace for editor-centered browser tasks.
Voice, Speech & Media Synthesis
Production pipelines for natural speech, private transcription, and programmatic media generation.
Edge-TTS
Twenty-two Microsoft neural voices with word-boundary SRT timestamps for video narration, automated demos, and podcast audio.
gTTS Regional Audio Routing
Seven verified regional acoustic models with phonetic preprocessing for technical acronyms such as PHP, CVE, and API.
faster-whisper
Local faster-whisper runtime for confidential audio, meeting, and voice-note transcription without sending source media to a third-party speech API.
fal.ai
API-driven generative-video pipelines for short-form media assets, visual presentations, and repeatable content production workflows.
Workflow Orchestration & Protocols
The protocols, durable engines, automation hubs, and Git-backed architecture connecting agents to real systems.
Model Context Protocol (MCP)
Standardized protocol connecting T3 Code, Cursor, Pi, OpenCode, Hermes, BrowserOS Neo, notes, filesystems, and databases through one tool interface.
Temporal
Distributed, stateful workflow engine powering long-running background tasks with deterministic retries and resilient error recovery.
n8n (Self-Hosted)
Self-hosted workflow automation orchestrating social publishing, webhook processing, AI services, and multi-system synchronization.
Zero-DB Markdown Engine
Laravel 13 engine powering this portfolio with zero database overhead, instant page loads, and fully version-controlled content.
Postiz
Self-hosted open-source social scheduling engine that plans, previews, and publishes across TikTok, Instagram, LinkedIn, and more, with durable schedules powering this site's Klytron Broadcaster stack.
Architectural Principles for AI Systems
A tool is only as good as the governance and architectural constraints surrounding it. Here is how I run high-velocity AI systems without chaos.
Circuit Breaker & Fallbacks
Never tie production workflows to a single model provider. My custom local router dynamically fails over across multiple providers if limits or outages hit.
Local-First Confidentiality
Confidential client data, credentials, and private intellectual property never touch third-party cloud APIs. Offline inference via Ollama and Whisper keeps data sovereign.
Git-Tracked Determinism
No fragile database states. Every prompt, agent soul, memory profile, and content piece is tracked in Git, making environments reproducible and disaster-proof.
Durable Agent Autonomy
Transitioning beyond naive prompt loops. Autonomous agents operate in sandboxed Docker containers with Model Context Protocol (MCP) tool integration.
Want an Autonomous AI Engine for Your Business?
I help startups, enterprises, and founders architect, deploy, and govern customized AI systems, agent swarms, and resilient backend pipelines that cut operational friction.