The toolkit

A lab of tools that run every day.

Each tool below carries an honest status — live, internal, prototype or private. No launch theatre: what works you can open, what doesn't is labelled as such.

01
web dashboardInternal

Borg Cockpit

One screen over the whole toolkit.

The public front door to the lab: a dashboard that pulls together portfolio state, audit snapshots and agent activity in one place. Deployed and reachable at cockpit.borg.tools.

  • Portfolio and audit snapshots surfaced in one view
  • Live at cockpit.borg.tools — not a mockup
Open Borg Cockpit →
02
portfolio auditPrivate

Project Orchestrator

A reproducible read on a large personal portfolio.

Runs a self-built audit across the portfolio — code metrics, security scans and COCOMO cost modelling — then synthesizes the results with multiple LLMs. Intentionally not a public service; a redacted, read-only demonstration does not yet exist.

  • Self-built portfolio audit (not independently verified)
  • Code metrics and dependency hygiene
  • COCOMO-style cost modelling and multi-model synthesis
03
agent controlInternal

cmux Remote TUI

Drive coding agents on remote machines over SSH.

A fast terminal UI for orchestrating multiple coding agents across remote hosts. Built to run day to day on the machines that do the work.

  • Terminal UI over SSH to remote agent fleets
  • Worktree isolation per worker to avoid collisions
04
coding agentPrototype

AgentZero CLI

A TUI coding agent with a security interceptor.

An experimental TUI-based coding agent with a live activity feed and an interceptor that gates risky actions. Early and evolving — shared here as work in progress, not a product.

  • Security interceptor on agent actions
  • Live activity feed while the agent works
05
workflow intelligenceInternal

UniPro OS

Observe, analyze, and compile desktop workflows into governed execution.

A dedicated workflow intelligence platform that captures screen telemetry, extracts activity primitives, and compiles natural language intent into validated Plan JSON. Reachable at uniproos.theones.io.

  • Dedicated platform at uniproos.theones.io
  • Multi-tier architecture: Insight → OS → Rosea
  • Strict policy checks and syntax validation
Open UniPro OS →
06
agentic feedback loopPilot

Feedby

Feedback that ends in a pull request, not a ticket.

An AI-first feedback system in pilot: capture context from real users, triage signals with AI, and dispatch execution agents to draft pull requests. Not a completed, independently audited service.

  • Pilot at feedby.borg.tools
  • Automated AI triage and deduplication
  • Integration with coding agents and PR drafting
Open Feedby →
07
real-time generative audioLive demo

Synthapse

Google Lyria RealTime techno performance instrument.

A browser-based live music instrument powered by Google Lyria RealTime, dynamic DSP filters, loop banks, and audio-reactive WebGL visuals. Public demo is open at synthapse.theones.io — bring your own Gemini API key; it stays in your browser.

  • Open demo at synthapse.theones.io
  • Sub-bass mono & phase watchdog layers
  • Real-time prompt interpolation & MIDI controller support
Open Synthapse →
08
audio-to-3d-video pipelineInternal

Teledysk Studio

Audio analysis to 3D WebGL voxel music video generation.

Upload an audio track, and the pipeline extracts musical structure with Librosa/Whisper, directs a screenplay with LLMs, and renders a 3D voxel video in WebGL. Reachable at teledysk.theones.io.

  • Live platform at teledysk.theones.io
  • Multi-stage Librosa BPM / energy / beat extraction
  • In-browser Three.js voxel rendering and webm recording
Open Teledysk Studio →
How it's built

The principles behind the toolkit.

A lab, not a launch

These are the tools that run one developer’s operation daily. Some are live, some stay internal, some are still prototypes. The status label on each card is the honest state — nothing here is dressed up as a finished SaaS.

Evidence over claims

The portfolio audit is self-built and reproducible: a dated snapshot you can re-run, not an external certification. Where a tool is live you can open it; where it is internal, that is stated plainly.

Models are swappable

LLMs sit behind gateways with tiered routing, so a better or cheaper model is a config change rather than a rewrite. Deterministic cores, AI at the edges — the parts that must stay testable do.

Build-vs-buy is costed

Dependencies earn their place: managed auth over hand-rolled auth, pgvector over a separate vector DB, self-hosting only where data gravity demands it. Choices are made on cost and fit, not novelty.