The Brazilian PostgreSQL community is gearing up for one of the most anticipated events of the year: PGConf.Brasil 2026. Taking place in the city of Blumenau from September 2 - 4, this year’s conference promises to be another great gathering of developers, DBAs, and open-source enthusiasts. 

Our very own Euler Taveira, William Ivanski, and Charly Batista are part of the organizing team, and Barbara Leidens served on the Talk Selection Committee. Thanks to their hard work and a phenomenal response to the Call for Papers (CfP), a staggering 13 sessions featuring EDB talent have been selected—including practically the entire Barman team! Topics span from optimizing cloud-native deployments, deep dives into PostgreSQL internals, to bridging the gap between databases and Generative AI. 

A sneak peek at the lineup of EDB sessions you can look forward to in Blumenau. 

AI, LLMs, and the Future of Postgres 

While Large Language Models (LLMs) excel at generating text, they struggle with delivering discrete, dynamic, and always-accurate results. Enter the Model Context Protocol (MCP) - the new standard for giving generative AI models secure access to external data. In “Building an MCP Server Using Postgres”, Bruce Momjian will break down what MCPs are, how they are used, and showcase real-world database and non-database MCP servers. 

Niche vector databases had their moment, but PostgreSQL is proving it can natively handle heavy AI workloads. In “PostgreSQL como Motor de Inferência: Tokens, Embeddings e Gradients” (PostgreSQL as an Inference Engine: Tokens, Embeddings, and Gradients), Joao Detomini will take a deep technical dive into how Postgres manages tokens, handles giant context windows, and accelerates linear algebra calculations (like cosine similarity and Euclidean distance) to act as a core inference engine. 

Imagine troubleshooting database incidents by conversing directly with your localized knowledge base. In “Construindo um assistente de diagnóstico com pgvector e LLMs” (Building a diagnostic assistant with pgvector and LLMs), Raphael Max Brettas Vieira will demonstrate a 100% open-source, private framework using pgvector and local LLMs (via Ollama) to build a smart diagnostic assistant that runs on modest hardware without cloud APIs. 

Kubernetes and Cloud-Native Mastery 

Historically, adding or updating Postgres extensions in containers meant a painful cycle of rebuilding images and redeploying clusters. Gabriele Fedi, a major contributor to the postgres-extensions-containers project, will show how CloudNativePG’s new Image Volume feature connects with PostgreSQL 18's extension_control_path to load extensions dynamically on-demand right from your manifest. The title of his talk: “CloudNativePG: Rethinking PostgreSQL Extensions in Kubernetes”. 

Is CloudNativePG really unbreakable? Let’s find out! In “Implementando Chaos Engineering em clusters CNPG com LitmusChaos” (Implementing Chaos Engineering in CNPG clusters with LitmusChaos), Joao Cosme will put the operator logic under extreme stress, injecting compound failures like network latency, disk I/O exhaustion, and split-brain scenarios during failovers using LitmusChaos to prove why CNPG’s native Kubernetes architecture thrives under pressure. 

Database Internals & Performance Tuning 

Ever wondered why a Postgres backend continues to hold onto massive amounts of memory long after a heavy query finishes? Hint: It’s usually not a memory leak, but rather the way glibc interacts with long sessions. In “Conexões ociosas não são gratuitas” (Idle connections aren't free), William Ivanski will explore the system-level mechanics, evaluate alternatives like malloc_trim(), and explain why connection pooling with PgBouncer remains a pragmatic, cost-effective solution. 

Transaction ID wraparound remains one of the most critical operational risks for high-throughput Postgres databases, capable of bringing production clusters to a screeching halt. In “Entendendo transaction wraparound no Postgres” (Understanding transaction wraparound on Postgres), Israel Barth will build a clear mental model to help DBAs identify early warning signs, monitor the "xmin horizon", and take proactive remedial action before it's too late. 

Subtransaction overflow is a silent killer that slows down entire systems and lags replicas without obvious error messages. Barbara Leidens will peel back the layers on how Postgres manages SAVEPOINT, ROLLBACK TO, and PL/pgSQL exception blocks, exploring what happens when you breach the 64-subxid cache and how to diagnose it using wait events, in her talk titled “Subtransações no PostgreSQL: quando 64 SAVEPOINTs derrubam o seu banco” (Subtransactions in PostgreSQL: When 64 SAVEPOINTs Bring Down Your Bank). 

When performance dips, we usually blame individual query planners. But what about the lower system layers? Inspired by Brendan Gregg’s Systems Performance, Gustavo Oliveira will teach you how to look for "unknown unknowns"—such as CPU saturation, memory contention, and I/O bottlenecks—to build a truly data-driven optimization mindset. Look for “Análise de recursos e benchmarking com foco no PostgreSQL” (Resource analysis and benchmarking with a focus on PostgreSQL) on the schedule. 

What's Next for Postgres? (Features & Strategy) 

The Postgres backup landscape is at an inflection point. As a development engineering leader at EDB, Martin Marques will map out the current ecosystem, discuss recent shifts in open-source tool lifecycles, and debate the big question: which backup features belong in core Postgres, and which should stay in external tooling? His talk title: “Present and future of backups in Postgres”. 

Postgres connections are not stateless—session variables, temporary objects, and prepared statements accumulate quietly. When aggressively multiplexing or failing over, this creates chaos. Using Keel as a case study, Charly Batista will introduce Semantic State Virtualization: a protocol-aware proxy approach that tracks and migrates session state safely across backends. Bookmark his session: “Schrodinger's Connection: Mastering PG State with Semantic Virtualization”. 

With around 40 preinstalled data types in PostgreSQL 18, choosing the right one is an art form. In “Tour de Data Types: VARCHAR2 or CHAR(255)?” Andreas Scherbaum will move past the basic INTEGER and VARCHAR options to explore lesser-known data types, helping you understand how to efficiently store everything from IP addresses to geographic data. 

With “Postgres 19: O que temos de novo?” Matheus Alcantara will guide the technical audience through what is coming in Postgres 19. The session will focus deeply on architectural improvements inside the planner and executor—including features like pg_plan_advice and eager aggregation—so you can prepare your production systems ahead of time. 

Join us in Blumenau! 

PGConf.Brasil is the perfect place to level up your Postgres skills, meet the brains behind the database, and network with the community. Make sure to save these sessions, come say hello to the EDB team at our booth, and bring your toughest database questions. We can’t wait to see you there!