The exact stack I use.
Recruiters ask, candidates ask, friends ask. Easier to write it down. No affiliate links, no aspirational tools, only what I actually open.
Last updated July 3, 2026 · 49 tools, grouped by job
AI & LLM stack
Where most of the daily tokens go. Model loyalty is a trap, so I switch per task.
Claude CodeMax plan, primary coding assistant, agents and MCP servers
Claude Code pluginsclaude-mem, graphify, gstack, apsolut-seshat, and my own apsolut-cortex on npm
OpenClawpersonal AI assistant gateway, self-hosted on an isolated mini PC so experiments stay contained
Hermes AgentNous Research's agent on the same isolated box, run from its dashboard; my bet is dashboards, not CLIs, become how we drive agents
ElevenLabsvoice on this site, the assistant's Listen button runs on their TTS API
Unstructureddocument parsing for LLM pipelines, PDFs and messy docs into clean chunks
Direct APIsAnthropic, OpenAI, Google, for production paths
Coding with AI
Prompt-to-app and hosted prototyping, where ideas earn the right to become real projects.
Bolt.newpremium since 2024, browser-based prompt-to-app when I want zero setup
ReplitPlatform Builder L3, prototyping in a hosted environment
LovableAI app builder, vibe-coding speed when the goal is to ship a working surface fast
v0.devVercel's UI generator, fast React/Tailwind scaffolding from a prompt
Grok BuilderxAI's builder, in the daily rotation for quick prototypes
Inference & routing
Where the models actually run, and what they cost. Provider loyalty is the same trap as model loyalty.
OpenRoutergeneralist router, broad model coverage, easy switching
Heliconecost and latency tracking, real eval numbers per model
Requestyalternate router, different cost profile, worth keeping in the rotation
Fireworks AIOSS model hosting, Llama, DeepSeek, Qwen without GPU babysitting
Basetenproduction deploys, custom models when generic hosting won't cut it
Ollamalocal model serving in the homelab, for offline, private, or token-free iteration
Web & framework
The stack I ship to production.
Next.jsReact 19 + TypeScript, default for new builds
TanStackQuery for data-fetching, Router and Table when the project earns them
Rechartsthe chart layer in my dashboards, Brand Intelligence included
Tailwind CSSdesign system implementation
Payload CMSTypeScript-native, code-first, the forward-CMS choice when greenfield
HubSpot CMSHubL, HubDB, current daily driver, at the level a real HubSpot dev ships
WordPressGutenberg, ACF, WooCommerce, WPML, decade-deep when the project calls for it
Vercelhosting for Next.js work
PostHogproduct analytics, wired into this site and my apps
Playwrightend-to-end tests and browser automation
Expo (expo.dev)React Native builds when the project needs to leave the browser, one app close to publish
Automation & integration
The glue between marketing, product, and the rest of engineering.
n8nprimary workflow tool, self-hosted
Scheduled agent pipelinesCIA's weekly runs, LLM jobs on Railway cron, briefs committed to git
MCP serverscustom-built, wiring my own tools into Claude Code and agents
Power Automatewhen the client lives in Microsoft 365
Chrome extensionscustom-built for repetitive browser work
Data & infrastructure
Boring tech I trust to not page me.
Tursoedge SQLite, fast, durable, cheap
Drizzle ORMTypeScript-first schema and queries, pairs with Turso and Postgres
PostgreSQL + pgvectorwhen relational + embeddings sit better together
Qdrantvector search for Brand Intelligence, when embeddings need their own store
Ablyrealtime data sync when polling won't do
SupabasePostgres + auth + realtime + storage when I want batteries-included
Railwaycron and small services
Coolifyself-hosted PaaS for the apps I want full control over
AWSS3 for voice/media/durable blobs, Route 53 for DNS, the bits that fill in when self-hosting isn't the right call
Daily drivers
Where I actually live in working hours.
Obsidiansingle source of notes, left Notion behind
Git + GitHubversion control, briefs-as-git for CIA
Figmadesign handoff, prefer it for delivery
VS Codestill the editor under Claude Code
Where & how
Starnberg home office, just outside Munich. Remote-first since 2017.
Windows · Ubuntu · Zorin OSdaily driver across the desktops, depending on the work
Self-hosted labthe boring rig that runs the agents and homelab services
Want this stack working for your team?
Based near Munich, open to senior in-house roles, remote in Germany, plus select Toptal engagements.
Prefer email? hi@aleksandarperisic.com.