Tools
The hardware, software and services behind the work.
Z790 board, 24 cores, 110 GB RAM shared with WSL — also runs the self-hosted GitHub Actions runners.
Disposable Linux/Windows VMs, one per project or per agent, reachable through VS Code tunnels.
One long canvas: editor, terminal columns and a browser side by side.
HyperHDR screen capture drives the USB strip and the case LEDs; agent events pulse the room.
Daily build, tunnels into every VM.
Agent mode with my own rules, skills, hooks and memory — see workspace-ai.
Same agents in the terminal, routed through codai in BYOK mode.
Default shell; scripts validated with the parser before they run unattended.
Workspaces + catalogs in every monorepo.
The only search tool that survives 2,000 node_modules folders.
Linux builds, CI runners and Docker host.
Cloud Run images and local Postgres.
Next.js web apps, previews per branch.
Cloud Run services and jobs, Cloud SQL Postgres, KMS, Secret Manager, Vertex AI for Claude and GPU training.
Serverless Postgres for dev branches and small products.
Workers for release updaters.
Primary IaC for GCP.
IaC in Brivio.
Self-hosted WSL runners on the workstation; ubuntu-latest as fallback.
Errors and traces, env-gated in every app.
One model name, Claude via Vertex AI and Azure AI Foundry behind it, with spend caps and prompt caching.
App-side LLM calls, streaming and tool use.
axiom (reasoning) on npm, plus the codai MCP memory server and Windows computer-use.
Speech in and out for metu and TikSee.