We built an internal, remote claude code agent service that runs on Fly.io machines orchestrated through FLAME. It’s been a huge benefit to our whole team and was a really exciting project to work on from a tech and product perspective.

https://sylverstudios.dev/blog/2026/04/06/one-agent-one-machine

Episode 298 of Thinking Elixir. News includes the hex.pm website getting a fresh new redesign, the Hex 2.4 package manager upgrading to OAuth device flow authentication with two-factor authentication support for improved security, the official Elixir Expert LSP hitting its v0.1 release milestone, a new LiveStash library from SoftwareMansion that prevents state loss on LiveView reconnects, and José Valim himself stopping by to remind everyone that the ElixirConf US Call for Talks deadline is fast approaching, and more!

https://www.youtube.com/watch?v=jn4tmDrIMFo

Dima Mikielewicz shows Legion: agents that execute Elixir code in sandboxes, spawn other agents, and react in real-time. The BEAM advantage. Save your spot at ECEU 2026 https://www.elixirconf.eu/#register

Your application’s throughput is only as fast as it’s slowest bottleneck. https://blog.appsignal.com/2026/04/02/debugging-slow-ecto-queries-with-appsignal.html

I’m a big fan of FLAME, but recently, I’ve been having trouble properly testing it in a local environment. The FLAME.LocalBackend module does not properly mimic a FLAME cluster, and that means that if I want to test something that needs to ensure that the distribution works as expected, I’m stuck either locally testing with a real backend (such as Fly or EC2), or pushing to production and hoping for the best.

Worry no longer! Introducing FlamePeer: https://github.com/probably-not/flame-peer

Brought to you by the same guy who created SafeNIF and FlameEC2 (me), FlamePeer takes the things I learned from both of these libraries and puts them together! It provides a FLAME backend for Erlang :peer nodes, allowing you to properly test FLAME locally, without worrying about whether your code will actually work with real FLAME distribution in production.

Chris McCord – Creator of Phoenix José Valim – Creator of Elixir Quinn Wilton – Collector of dead languages One more to go 👀 Who’s next? https://elixirconf.com/ Haven’t submitted yet, CFT ends on April 12

Episode 297 of Thinking Elixir. News includes Quickbeam, an exciting new research project that brings a full JavaScript runtime inside the BEAM with OTP supervision, native DOM access, and a built-in TypeScript toolchain, plus a companion Volt asset pipeline for Elixir; José Valim highlights how Elixir’s type system work is already inspiring optimizations in Python’s Ruff project, and shares a new blog post on the latest BDD performance improvements coming in Elixir v1.20; LiveView Debugger v0.7 arrives with source code links, nested live view tree structure, and a closer path to v1.0; a new browser-based Elixir Language Tour expands its Processes chapter with hands-on GenServers, Supervisors, and more; Oban v2.21 lands with workflow tracking, rate limiting, and massive index improvements; and more!

https://www.youtube.com/watch?v=FfB08RlSZJc

Free tickets for underrepresented groups. Everybody welcome to apply. Apply now: https://www.elixirconf.eu/#diversity

This post is based on Chapter 3 of Build a LLM from Scratch by Sebastian Raschka, with one twist: all Python examples are rewritten in Elixir. We are building LLMs attention mechanism. Attention mechanism is second part of stage 1. We have already prepared input text data. Attention mechanism help LLM to predict next token. We will implement four attention mechanism:

• simplified self-attention
• self-attention
• casual attention
• multi-head attention https://karlosmid.com/2026/03/coding-attention-mechanism-in-elixir/

Only 15 days remain to submit your talk. The Program Committee will begin reviewing submissions soon. https://elixirconf.com/#cft

Open Telemetry Tracing via Erlang Tracing right?

But what if you didn’t prepare? Hot code updates to the rescue.

https://underjord.io/injecting-tracing-the-hot-way.html

The official Hologram extension for VS Code is now live! It brings full syntax highlighting for HOLO templates: ~HOLO sigils in .ex files and standalone .holo template files. As the template language has grown, writing HOLO in plain text started to feel increasingly painful - this extension fixes that and makes the editing experience much more pleasant. More here: https://hologram.page/blog/hologram-vscode-extension-released

We wrote about structuring authorization in Phoenix using Phoenix Scopes. The article focuses on keeping permission logic closer to the domain and avoiding scattered checks across plugs and controllers. It covers: How Phoenix Scopes approach authorization How to reduce ad-hoc permission checks How this pattern works in practice

https://www.curiosum.com/blog/phoenix-scopes-authorization-permit-phoenix

Noelia Lencina shares patterns for Postgres health, distributed orchestration, and rate-limiting at scale. Save your spot. https://www.elixirconf.eu/#register

Episode 296 of Thinking Elixir. Elixir v1.20.0-rc.2 and rc.3 arrive with a faster compiler, better type inference, and improved incremental compilation; José Valim drops a low-key bombshell with Distributed Python running on top of the Erlang distribution with full Livebook integration; Chris McCord wows the community with fly_deploy, enabling zero-downtime hot code upgrades on Fly.io using the BEAM’s ability to boot a peer VM inside a running VM; Discord shares a deep dive into how they added distributed tracing to their Elixir systems without melting everything; Popcorn v0.2 brings Elixir to the browser as a proper npm package; OpenAI’s Symphony project — built in Elixir — sparks conversation about vibe-coded BEAM code and whether TypeScript tools can truly replicate what the BEAM offers, and more!

https://www.youtube.com/watch?v=8-r6jDNbl2o

Latest webinar from Erlang Solutions: Traffic spikes are part of the deal for gaming and entertainment platforms. In this talk, Lee Sigauke looks at why systems struggle under sudden demand, and shares practical ways to build platforms that stay reliable and scale smoothly when things get busy: https://www.erlang-solutions.com/webinars/how-to-build-platforms-that-dont-let-audiences-down/

Barry O’Reilly, Mike Begley, Ingela Andin, Björn Gustavsson + 11 more speakers covering AtomVM, Gleam, AI, performance optimization & more. May 18 at Ericsson offices. https://codebeamstockholm.com/#register

Every protocol you’ve shipped assumes the spec doesn’t change during the connection. This one doesn’t. Code is typed, composable, content-addressed, runtime-loadable, and cryptographically verified.

The BEAM was built for phone calls that can’t drop.
(https://systemic.engineering/conversation/)

Now it runs conversations that can’t afford to.
(https://github.com/systemic-engineering/conversation)

Part 1 of 12 — This series documents a proof-of-concept Notification-Oriented Paradigm (PON) engine in Elixir, a hexagonal boundary around it, and a Smart Brewery digital twin used as a stress lab (simulation, LiveView, telemetry, TimescaleDB, and ML hooks). Here we set the vocabulary and motivation; later posts walk through OTP wiring, the metaprogrammed DSL, the brewery case study, and hard-won performance lessons. https://dev.to/matheuscamarques/notification-oriented-paradigm-pon-in-elixir-why-the-beam-fits-reactive-rules-2p9e

Robert Virding is teaching “Secure Coding in BEAM” on 22 May, 2026 in Stockholm. https://codebeamstockholm.com/trainings/Secure-Coding-In-BEAM/