AI agent tools

AI agent tools I use to build production agents

This is the practical stack behind my production AI agent work: frameworks, coding agents, MCP servers, models, databases, infrastructure, monitoring, and deployment patterns I trust when an agent has to do real work.

Build an AI agent with me Read the production checklist

Agent frameworks

Frameworks I use when the agent needs memory, tools, messaging, scheduling, and production ownership.

Hermes Agent

Self-improving, model-agnostic agents from Nous Research with memory, scheduled crons, subagents, and messaging gateways.

OpenClaw

Self-hosted, messaging-first agent framework with SOUL.md identity, plugin pipelines, and simple ownership of the runtime.

LangGraph and CrewAI

Useful for graph-based workflows, role-based agent teams, experiments, and prototypes that need explicit orchestration.

Coding agents and development workflow

Tools I use to move from architecture to working code without losing review, tests, and deployment discipline.

Claude Code

Daily driver for repository-aware coding, refactors, test loops, debugging, and agent workflow design.

Zed agents and Codex-style coding agents

Useful for parallel implementation, codebase audits, documentation, and fast iteration on focused tasks.

Next.js and TypeScript

My default web layer for dashboards, admin surfaces, agent control panels, and marketing pages.

Models and AI APIs

I keep agents model-agnostic so the workflow can switch providers as quality, latency, or cost changes.

OpenAI and Anthropic

Frontier models for reasoning-heavy tasks, complex synthesis, tool planning, and quality-sensitive automation.

OpenRouter and Nous Portal

Provider flexibility for routing, fallback chains, cost control, and access to multiple open and hosted models.

Local and open-weight models

Used when privacy, cost, or infrastructure ownership matters more than using a single hosted vendor.

MCP, tools, and data access

The useful part of an agent is usually the tools it can safely call and the context it can reliably retrieve.

MCP servers

Model Context Protocol servers for connecting agents to files, databases, SaaS tools, browsers, and internal APIs.

Supabase and Postgres

Structured memory, application data, auth, admin dashboards, audit trails, and reporting workflows.

Vector search and RAG

Semantic retrieval for long-term knowledge, support docs, internal processes, and domain-specific memory.

Infrastructure and operations

Production agents need boring infrastructure: isolation, logs, retries, approvals, secrets, and network boundaries.

Docker and VPS deployments

Containers on Hetzner, DigitalOcean, Hostinger, or client-owned infrastructure for self-hosted agents.

Tailscale

Private mesh networking so agent services, dashboards, and SSH are reachable without exposing public ports.

Cron, queues, logs, and evals

Scheduled work, background jobs, full audit trails, regression checks, and observability from day one.

How I choose tools for production AI agents

I do not pick tools because they are trendy. I pick them based on whether they make an agent easier to own, inspect, secure, and improve over time.

A useful production agent needs narrow scope, durable memory, safe tool access, human approval where it matters, clear logs, fallback behavior, and a cost model that still works at 10x usage. The tools above are the pieces I use to make that real.