Welcome to GigaElixir Gazette, your 5-minute digest of Elixir ecosystem news that actually matters.

This week brings LiveFlow's interactive flow editor, FusionFlow's workflow engine, critical OTP security patches, and a clever approach to PDF generation that ditches the Chrome dependency. Plus a real-time query engine that's actually interesting.

After 18 months of building and one complete rewrite, Jido 2.0 proves the BEAM is the best runtime for agent systems. The core insight: agents are just data, a struct with state, actions, and tools that runs inside a GenServer. Everything flows through one function: `cmd/2` takes actions in, returns updated agent plus directives out. Side effects become typed data structures the runtime executes. This makes every agent decision unit-testable without touching networks, databases, or LLMs. The 1.0-to-2.0 journey matters here. Original version was overengineered - too many abstractions fighting OTP instead of embracing it. Version 2.0 strips ceremony, adds pluggable strategies (Direct for sequential, FSM for state machines), and makes AI integration just another strategy implementation. ReAct and Chain-of-Thought reasoning plug in without changing the core model. Single-threaded event loops managing concurrent agents with promises is architectural malpractice. Python does slightly better but can't stay up long-term. The BEAM was built for exactly this: lightweight processes, supervision trees, message passing. After 18 months in production, that bet is paying off. Available on Hex now.

Read more →

GES233 shipped a working proof-of-concept: complete DiffSinger (Singing Voice Synthesis) inference pipeline in Elixir, no Python required. The core engine published as `Orchid` and `OrchidSymbiont` on Hex. This isn't a toy - it generates mel-spectrograms, F0 pitch curves, and phoneme boundaries, all rendered through VegaLite. The Livebook example requires downloading a 500MB model, then runs the full pipeline locally. The long-term goal challenges current AI workflow tools directly to build better, more maintainable AI workflow engines. Current Python-based node tools (like ComfyUI) suffer from spaghetti graphs and poor abstraction. The BEAM's process model maps naturally to pipeline stages, with each node becoming a supervised process; data flows through message passing, and failures are isolated without cascading. Right now the pipeline is constructed programmatically via manual DSL with no visual node editor yet. But the hard part is done: proving the BEAM can handle real ML inference workloads with proper abstractions. This is the kind of project that shows Elixir's reach beyond web apps - ML pipelines, audio synthesis, concurrent data processing. The BEAM keeps surprising.

Read more →

Early Bird pricing for ElixirConf EU 2026 ends in one week, then prices jump. The conference hits Málaga, Spain in April with 600+ attendees and content that matters: José Valim keynoting on type inference in Elixir 1.20, Chris McCord unveiling DurableServer (processes that survive node failures and migrate between machines), plus 40+ talks covering Phoenix at scale, LiveView innovations, Nx/ML, Nerves embedded systems, and production architecture. Pre-conference training on April 22nd includes Andrea Leopardi teaching Advanced Concurrency Patterns, Bruce Tate on AI + Elixir, and Chris Beck on RAG Systems. This is Europe's largest Elixir gathering; if you're building anything serious on the BEAM, this is where the community concentrates knowledge transfer. The DurableServer announcement alone justifies attendance: processes that survive crashes AND move between nodes rewrites the rules for stateful distributed systems. Málaga in April beats another virtual conference. Face-to-face debugging sessions in the hallway track, late-night architecture discussions over Spanish wine, watching José live-code type system improvements. Early Bird ends next week. Don't wait.

Read more →

Ten years ago, the question was: what would Chef/Ansible look like built entirely in Elixir using Erlang distribution instead of agents? The project got 90% done: SSH tunnels, remote BEAM bootstrap, transparent clustering. Then life happened. A few weeks ago, the final 10% (bytecode pushing and dependency resolution) shipped in a day with AI assist. Now it's on Hex as Fusion. Connect to any remote server over SSH, push your modules, run them. No deployment pipeline. No pre-installed app. Just SSH access and Elixir installed. When you call `MyApp.Worker` remotely, Fusion reads BEAM bytecode, walks the dependency tree, pushes everything needed. Three SSH tunnels bridge Erlang distribution through firewalls. Remote BEAM bootstrap starts Elixir with flags routing all traffic through tunnels. Zero runtime dependencies. ~700 lines of Elixir. Use cases: ad-hoc tasks across server fleets, provisioning without Python, one-off diagnostics, interactive remote exploration. The architecture is elegant - two reverse tunnels, one forward, automatic dependency resolution, transparent clustering. This is what happens when you stop fighting SSH and embrace what the BEAM does naturally: distributed computing. The three-part deep-dive series covers tunnel architecture and bytecode transfer internals.

Read more →

Ashkan spent two years building Alpen, a personal planner that challenges the "big list of todos" model. The philosophy: until a task enters your calendar, it's in purgatory. The app implements the A.L.P.E.N method, time blocking, and the Eisenhower Matrix, not as features you configure, but as opinionated workflows baked into the UI. Planning becomes a flow, not a fight. The learning curve exists - this isn't another Todoist clone with familiar patterns. But after the curve, you have a framework with certainty. Life is chaotic; Alpen provides mechanisms to jot things down quickly, then give intention to when you do them. The business model matters: subscription-based but affordable, leveraging Elixir's efficiency so a solo dev keeps costs low. Roadmap is ambitious: personal project management that co-exists with Linear/Jira. Managers create tickets there, Alpen makes them yours. You break them into pieces, schedule them, execute. It's live at alpenplanner.com.

Read more →

You're training Nx models in IEx, stepping away for coffee, coming back to a session with zero command history. You retype the same data loading pipeline. Again. The BEAM's shell doesn't persist history by default. One environment variable fixes this permanently: `export ERL_AFLAGS="-kernel shell_history enabled"`. Add it to your `.bashrc` or `.zshrc`. Now the Erlang shell remembers everything across sessions. Ctrl+R works like your terminal - reverse search through weeks of commands. Training a model overnight? Your entire experiment setup is still there in the morning. This isn't just convenience for ML workflows. Any long-running IEx session benefits: debugging production issues, testing GenServer interactions, exploring APIs. The history file lives in your home directory (`.erlang-hist-*`), survives reboots, and works across all BEAM languages. You're already using Ctrl+R in bash. Now use it where you actually write Elixir. Your IEx session becomes your lab notebook. Scroll back through yesterday's experiments, copy the exact sequence that worked, iterate. No Jupyter overhead, no JSON diffs in git. Just the BEAM shell doing what terminals did in 1975, but better.

Remember:

Source →

Stay opinionated, stay building.

Want to build agent systems that don't make you hate your life? Check out what Jido 2.0 is doing differently and question everything you know about AI frameworks.

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 💜