Welcome to GigaElixir Gazette, your 5-minute digest of Elixir ecosystem news that actually matters.
This week in Elixir: Mob brings OTP natively to iOS and Android apps, NervesConf 2024's full talk archive is finally public, and Ex_graphblas ships GraphBLAS-standard sparse matrix algorithms to Hex.
Plus: 13 CLAUDE.md rules that stop AI assistants from writing case/if pyramids where function heads belong - and other habits that age badly under a real OTP supervisor.
. WEEKLY PICKS .
🔥 The BEAM Now Runs Inside iOS and Android Apps
Mob is the story of episode 302: OTP running natively inside iOS and Android apps - not wrapped in a webview, but actual BEAM processes inside the app. The episode also covers Folio for print-quality PDF generation via Typst Rustler NIFs; Oban v2.22 and Pro 1.7 with a rate-limiting overhaul; LiveVue v1.1 with Node-less server-side rendering via QuickBeam; and erlang_quic v1.3.0 - a production-ready pure-Erlang HTTP/3 stack for OTP 27. A lot shipped this week.
⚡ NervesConf 2024 Full Talk Archive Is Finally Online
The complete NervesConf 2024 video set is now public after amclain tracked down missing recordings with help from John Cotton and Todd Resudek. Talks include what's new with Nerves from Jon Carstens, Ryota Kinukawa on the Nerves-Linux relationship, and the full conference lineup. If your team works in embedded Elixir, this is the reference library you were missing - all talks, all sessions, one place.
📊 GraphBLAS Sparse Matrix Algorithms Land on Hex
Ex_graphblas v0.2.0 brings the GraphBLAS standard to Elixir - the same mathematical foundation behind MATLAB's sparse operations and Python's python-graphblas. Express graph algorithms as sparse matrix operations using semiring algebra. Ships with seven built-in algorithms, including BFS, shortest path, PageRank, and connected components, plus a pure-Elixir reference backend for development and a SuiteSparse NIF backend for production. Same application code, no migration when you swap backends.
🎯 One BEAM Node Solves Kubernetes' Noisy Neighbor Problem
Kubernetes scales compute. It does not solve fairness. When an agentic burst hits, pods get hammered and the autoscaler takes two minutes to respond while requests are already failing. EZThrottle Local queues inbound jobs in memory on a single BEAM node and drains them at the rate the upstream dictates via response headers. On 32GB, it buffers 3-32 million jobs. Hot code reloads preserve the in-memory queue across deploys - no state lost on updates.
🚀 Hologram Builds a Component Library for Browser-Compiled Elixir
Hologram compiles Elixir to run directly in the browser and is now building a maintained component library. The waiting list is open and the team is asking which components to ship first. The argument: components need accessibility, responsive design, and consistency across the full set - attributes that individual AI-generated components reliably skip. If you are watching the full-stack Elixir space, Hologram UI is the project to follow.
💡 Pro Tip
Drop a CLAUDE.md, Reclaim Idiomatic OTP
AI assistants have a specific problem with Elixir: they write code that runs but fails the smell test. They default to case/if chains for multiple function heads, reach for try/rescue as if it were Java exception handling, and treat GenServers as stateful objects with methods.
The code compiles, passes basic tests, and ages badly the moment a supervisor restart exposes an implicit assumption. The cause is not model quality - it is the absence of Elixir-specific conventions the model was never taught.
A CLAUDE.md (or .cursorrules or AGENTS.md) at the project root closes 13 specific failure modes. The rules cover multiple function heads over conditionals, with/1 for fallible operation chains instead of nested case pyramids, always using Ecto changesets for validation, pattern matching on data shape not type, and never leaving IO.inspect in returned values.
Each rule ships with a before/after code example showing exactly what went wrong. Drop the file in any project and the AI starts producing output that passes a senior Elixir developer's code review.
Remember, for AI-assisted Elixir development:
Function heads beat case/if - Pattern matching at the function head is Elixir's signature move; a single function with internal conditionals is a Ruby reflex that OTP supervisors will eventually punish
with/1 is your fallible chain - Nested case pyramids are the second most common AI mistake; with/1 flattens the happy path and returns early on the first failure
Supervisors are your error boundary - Let OTP restart processes instead of wrapping call sites in try/rescue; crashes are information, not failures to hide
CLAUDE.md works for every AI tool - Whether you use Claude, Cursor, or Codex, the same rules file teaches house conventions before the first line is written
. TIRED OF DEVOPS HEADACHES? .
Deploy your next Elixir app hassle-free with Gigalixir and focus more on coding, less on ops.
We're specifically designed to support all the features that make Elixir special, so you can keep building amazing things without becoming a DevOps expert.
See you next week,
Michael
P.S. Forward this to a friend who loves Elixir as much as you do 💜
