Tools I Use
Full transparency on the stack behind Agentic Up — everything I use to build, deploy, and run AI agents.
AI Agent Development
Claude (Anthropic)
Primary LLM for building agents. Claude excels at tool use, long context, and following complex system prompts. I use Claude 4 Sonnet for most agent loops and Opus for complex reasoning tasks.
Claude Code
My daily driver for coding. Terminal-native AI coding assistant that understands full codebases. Used for building agent systems, refactoring, debugging, and writing infrastructure code.
Python + TypeScript
Python for agent backends (LangGraph, CrewAI), TypeScript for frontends and simple agents. Python dominates the agent ecosystem — most frameworks and tooling are Python-first.
Git + GitHub
All agent code is version-controlled. GitHub for hosting, CI/CD via GitHub Actions. Every agent project gets its own repository with proper README, license, and documentation.
Agent Frameworks
LangGraph
My go-to framework for production agents. Stateful graphs with branching, human-in-the-loop, and persistent memory. Handles complex workflows that span multiple LLM calls and tool executions.
CrewAI
Used for multi-agent systems where specialised agents collaborate. Great for research pipelines, content generation workflows, and scenarios requiring role-based agent teams.
Custom Loops
For simple agents, I build custom loops from scratch. No framework overhead — just an LLM call, a tool list, and a while loop. Surprisingly effective for narrow-scope agents.
Infrastructure & Deployment
Docker
Every agent is containerised. Docker ensures consistent environments from development through production, simplifies dependency management, and makes deployment repeatable.
Railway
Primary deployment platform for agent backends. Easy Docker deploys, built-in monitoring, automatic SSL, and reasonable pricing for production agent workloads.
Cloudflare Workers
Used for lightweight agent APIs, webhook handlers, and edge functions. Great for stateless agents that need low latency and global distribution. Also hosts this blog.
Hetzner VPS
For agents that need persistent compute — long-running processes, scheduled tasks, resource-intensive operations. Hetzner offers the best price-performance ratio for European hosting.
Data & Storage
Chroma
Local vector database for RAG systems. Lightweight, easy to set up, and runs embedded in your application. Perfect for agent prototypes and internal tools.
PostgreSQL + pgvector
Production vector storage. PostgreSQL with the pgvector extension gives you a relational database and vector search in one system. My default for production RAG.
Supabase
Managed PostgreSQL + auth + storage for agent backends. Excellent developer experience, generous free tier, and built-in vector search via pgvector. My go-to for new projects.
Monitoring & Observability
LangSmith
Tracing and debugging for LangGraph agents. Traces every LLM call, tool execution, and state transition. Invaluable for understanding why an agent made a particular decision.
Plausible
Privacy-first analytics for this blog. Lightweight, no cookies, GDPR-compliant. Tracks page views and referral sources without compromising visitor privacy.
Custom Logging
JSON-structured logging for all agent systems. Every LLM call, tool result, and error is logged with timestamps, costs, and context. Debugging agents without logs is impossible.
This Blog
Astro 6
Static site generator. Ships zero JS by default, React islands for interactive components. Fast builds, excellent Markdown support, and great SEO integration.
Tailwind CSS
Utility-first CSS framework for styling. Keeps the design system consistent across components without writing custom CSS.
Cloudflare Pages
Hosts the blog. Automatic deploys from GitHub, global CDN, free tier. Zero server management for a static site.
Want an agent built with this stack?
→ See the offer →