Mark Ericksen

Excited to share the new CodeFlow course for individuals and teams learning Elixir! Coming from OOP, Elixir can be jarring, make the mental shift faster and love the process! @elixirlang #elixirlang @ThinkingElixir

@Bull_lion_aire It also provides an abstraction layer over the different models/services you call. So you can switch providers without changing your code. And it has built-in error handling and fallbacks to other models when a service is overloaded. Stuff like that.

@Bull_lion_aire It’s based on the idea of a “chain” of messages. Of built-in tools with override options for things like summarization. And of linking a native Elixir function to a %Function{} that the LLM can call, meaning a ToolCall. Some of the data structures that the same names.

Elixir LangChain v0.6.2 released! Includes the new chat model ChatReqLLM! If you're using ReqLLM but want a bit more on top of it, now it's easy! Now there's no excuses for trying out the Sagents library for awesome OTP agents. #Elixirlang #MyElixirStatus github.com/sagents-ai/sag…

@truechatdata This feels very AI. Is it a real question? You can do it either way you like. A custom Sagents middleware can be added to do observably however you want with any backend you use. There’s a doc for that.

@brainlid Nice release. Middleware and OTP supervision are a strong combo for agent work because they make orchestration less ad hoc. Curious how ChatReqLLM is positioned versus using ReqLLM directly when teams want observability, retries, and tool calling in production.

@Bull_lion_aire The Elixir library was “inspired by” but isn’t a copy. I plan to change the name a well.

@brainlid The fuck? Langchain was a mistake that was needed to be made. Why repeat same mistakes?

@davydog187 This 👆 should make things easier for experimenting

Elixir LangChain v0.6.2 instructions for using the new experimental ChatReqLLM module: hexdocs.pm/langchain/0.6.…

@davydog187 @cjbell_ You should check out Sagents. The middleware allows you to write a custom ability to summarize, compact or completely change the set of messages. The custom run modes let you have more control over tool calls and execution. Built in HITL too. More to share soon!

Another feature I'd love is first-class support for rewriting messages, e.g. tool call results. @cjbell_ calls this "hydratable" The idea being that when making subsequent calls to the model, you can modify previous messages such that they are smaller, by summarizing, a ref, etc

I sent some PRs to ReqLLM this morning to clean up the docs github.com/agentjido/req_… github.com/agentjido/req_… github.com/agentjido/req_… I like this project, but it feels very LLM generated, and there's a broad API surface area that I think needs some work. #myelixirstatus

@josevalim @hugobarauna Post-it note driven development! Now you just need to systematize it and start another consultancy. 🟨😆

When we started the Numerical Elixir effort, I had a huge post-it on my desk with all of the things I wanted to tackle. It is almost all done: only Nx v1.0 remaining! 🎉

News includes @JoseValim's deep dive into @ElixirLang's type system changes, a new Dashbit post on type systems as leaky abstractions, Oban Pro teasing Workflow UI improvements, MDEx v0.11.6, Livebook Desktop on Linux, and more! #ElixirLang youtube.com/watch?v=YJQokV…

Hire me! I want to help an Elixir company deeply embed advanced AI Agents in ways that make sense and add real value to your product and value for your users. linkedin.com/posts/mark-eri… #MyElixirStatus #ElixirLang #AI #LLM

The SagentsLiveDebugger was updated to raise visibility into the HITL state of your your agents and subagents. github.com/sagents-ai/sag…

Sagents v0.3.x is out! Lots to share! AI agents doing things on your behalf is great... until they do the wrong thing. Sagents lets you define exactly which operations need human approval. github.com/sagents-ai/sag… #MyElixirStatus #ElixirLang #LLM #AI #agents 🧵👇

@chris_mccord Dang! That is cool! Thank you for sharing. 🔥🔥🔥

You only need to rename your start/2 and then we boot your entire sup tree as a new peer node and broker the cutover. The outgoing and incoming peers are even clustered together so can handoff state. Again and again erlang continues to blow my mind 12+ years and counting

The Erlang VM has all these incredible things. Did you know you can boot a `:peer` node as a VM within the VM, fully meshed? Enter `FlyDeploy.BlueGreen` - "hot bluegreen" deploys where we spin up the new incoming VM, then cutover seamlessly: before/after code changes:

News includes Expert LSP releasing its first RC, Elixir v1.20 compile time improvements up to 20% faster, @LivebookDev Desktop moving to Tauri with #Linux support, a new erlang-python library for ML/AI integration, and more! @ElixirLang #ElixirLang #AI #ML youtube.com/watch?v=Ed83ck…

News includes Hackney v3.1.0 with pure @erlang_org HTTP/3 support, Hornbeam running #Python apps on the BEAM, the Easel Canvas 2D drawing library for LiveView, Hologram v0.7.0 reaching 96% Erlang runtime coverage, and more! #ElixirLang @ElixirLang youtube.com/watch?v=DBuH_P…

News includes Mark’s new Sagents AI agent library, @JoseValim on why @ElixirLang is best for #AI, LiveDebugger v0.6.0, Elixir salary analysis, new #MCP server implementations, #Lua for Elixir revamp, and more! #ElixirLang youtube.com/watch?v=H1iyBr…

@codestirring @elixirphoenix Please try it out and share your feedback!

@brainlid @elixirphoenix This is awesome Mark - I can already think of a few projects I could use this in

Introducing Sagents -- an open-source #ElixirLang framework for building #AI agents with human oversight, composable middleware, and real-time @ElixirPhoenix LiveView integration. Built on OTP. Each agent is a supervised GenServer. Watch the full demo: youtube.com/watch?v=r-14qc…