Stack

The production AI stack we build on.

Model providers, agent orchestration, MCP, RAG, observability, and the app stack — what we use, and when we use each. Built for technical authority, not affiliate links.

Agent orchestration

LangGraph

Durable orchestration — checkpointing, retries, branching, human-in-the-loop

Read more

OpenAI Agents SDK

Tools, handoffs, and tracing for multi-step OpenAI agents

Vercel AI SDK

Streaming, RSC, and the modern AI app interface on Next.js

Read more

Custom orchestration

Bespoke control loops when off-the-shelf frameworks don't fit

MCP

Model Context Protocol

The universal standard for connecting agents to tools and data

Read more

Custom MCP servers

Purpose-built servers exposing your tools and workflows to agents

Tool registries

Catalogued, governed tools agents can discover and call

Private tools

Internal-only tools behind auth, scoped to your systems

RAG

Postgres / pgvector

Vector storage and similarity search inside Postgres

Supabase

Postgres, pgvector, and auth — the AI-native backend

Read more

Pinecone

Managed vector database for large-scale retrieval

Weaviate

Vector database with hybrid search and filtering

Hybrid search

Combining keyword and vector search for better recall

Reranking

Reordering retrieved passages for relevance before generation

RAG

Retrieval architecture over proprietary knowledge

Read more

Observability

Langfuse

Tracing, prompt management, evals, cost, and latency

Read more

LangSmith

Tracing and evaluation for LangChain and LangGraph apps

OpenAI tracing

Built-in tracing for OpenAI agent and tool runs

Custom dashboards

Bespoke views into cost, latency, and agent behavior

App stack

Next.js

The React framework for production AI apps

React

The component model the interface is built on

TypeScript

Type-safe end-to-end application code

Python

Data pipelines, evals, and ML tooling

Supabase

Postgres, pgvector, and auth backend for AI apps

Postgres

The relational core for application data

Redis

Caching, queues, and low-latency state

Cloudflare

Edge compute, storage, and delivery

Vercel

Deploy, preview, and run the Next.js app

Newsletter

One letter, every week. Working systems — not hot takes.

Build logs, agentic engineering decisions, agent failures, evals, and what survives real users. Sent weekly, never more.

Weekly. No spam. Unsubscribe anytime.