catethos

@BlinkDL_AI maybe it can APL for agent lol

Introducing Agen: a minimalist language for agent loops and state machines 🙂 purely state-driven and fun to code - requires a different mindset from other languages. Repo: github.com/BlinkDL/Agen

@spacebat @mikehostetler oh nice , any one that catches your eyes ? since Prolog is also homoiconic , it would be interesting to see those languages created for AI is used by AI :)

@catethos @mikehostetler People are doing funky stuff with agents in Clojure too

Jido agents can write code, inject it into the running Erlang runtime and call them immediately Self evolving software is coming

@zli11010 is it similar to backtracking ? if so maybe we can get it for free using something like Prolog

Ever wished you could rewind your AI agent when it goes off track? EnCompass lets you rewind any Python program containing LLM calls. While normal program execution runs a program once from start to finish, 1/

@biantaishabi5 期待用这个库手搓另一个Openclaw

github.com/biantaishabi2/… 用Elixir重写了一遍Pi，做这个完全是因为，首先想试下我那个BDDC的编译器能自动生成测试代码的，另外就是测试一下我那个后端的编译器，可以生成后端代码骨架的，还有就是我看它里边写的很多东西，进程管理，事件啥的，都是手写的，但是这是OTP的长项，所以想看看能省多少行？

@jmbollenbacher how about using pythonx as the REPL, hexdocs.pm/pythonx/Python…

ive been playing with building an RLM system in Elixir. It handles concurrency and fault tolerance beautifully. It makes observability into the tree of subagents easy. But it wasnt very "smart." So i think im gonna set it up to just use any existing CLI agent as the core agent.

@mikehostetler ah can't wait to see some code :) been trying to do use pythonx to do data analysis in similar way. I am keeping the python process alive so it's kinda look like a REPL to me, is it not making sense ?

@catethos I didn’t add a REPL - that doesn’t make sense in the BEAM Instead, I have a Workspace manager that looks like a virtual file tree and have an LLM generate a Lua snippet to chunk it - then use Luaerl to exec that safely purely in memory Much safer & faster

Jido/Elixir + RLM = ❤️

@Jarodxu7 happycapy

Comment happycapy for invitation code. happycapy.ai

Excited to launch happycapy, an agent-native computer on your browser. Claude Code → Clawdbot → Happycapy > Private sandbox for everyone to run Claude Code anytime anywhere - no Mac Mini needed > GUI built for everyday user, visualizing skills and its output

@MaximeRivest to be useful in non data analysis and do "real" literate programming , I really miss the weaving and tangling

This complete new way of interacting with LLMs will your blow your mind!

@OnlyXuanwo look so nice :) i was trying to do similar thing using Tauri, I bet gpui will be snappier

The new UI design for my own IDE The next step is to bring this design to life.

@mikehostetler this makes me want to write a coding agent in Elixir :)

Agents working in a single thread with 3 subagents is an artificial ceiling We have the tools now We are going so much higher

@tech_optimist @boundaryML @DSPyOSS interesting benchmark :) I am wondering if the main value of BAML lies in their schema aligned parsing ? I guess they are building a language to build AI pipeline , would be interesting to compare that part of it to DSPy

🧵 I compared the performance of @boundaryML BAML and @DSPyOSS for a variety of structured outputs, and the results are interesting: different datasets, models and schema formats results in wildly different outcomes, some of then unexpected. There's no universal winner (which means that prompt optimization matters, more than ever, because it's *very* hard to discover the right prompts as a human). The benchmarks compare BAML's and DSPy's performance (with similar user instructions and description annotations). I also use `BAMLAdapter` in DSPy (which implements BAML's schema formatting in a custom DSPy adapter). Why use a custom adapter in DSPy? Because it has benefits, especially when dealing with nested data, as the experiments show. 👇🏽 1/7

@niwoerner @badlogicgames nice I am doing similar thing using Ratatui, using Tauri seem to be able to display more results like graph

Day 2 of my winter holiday project to build my own personal shitty coding agent (inspired by @badlogicgames ) - Agent loop in place - Core abstraction for multi provider support in place - Session management in place - Server <-> Client communication in place

@pixperk nice, actor with type :) Gleam is nice, but I prefer Rust's type system

so i made this chat server, only using cinema (the actor model framework for rust, that i'm writing). cinema now has : - response await - async messages, supervision, watching - stream and registry support too working on distributed actors now. let's make it! repo's public. [link in reply]

@pixperk is it using any async runtime ?

so 3 days deep into building an actor model framework in rust, beat my head around several concepts and here's what we have as of now started with the foundation. - defined what an actor is, a trait with started() and stopped() lifecycle hooks. - messages need a Message trait that specifies the Result type. then built the address system, - Addr is how you talk to actors without holding a direct reference mailbox and message passing was next. - actors have an unbounded mpsc channel as their mailbox (will make it bounded later). the tricky part is type erasure, when you have Handler and Handler on the same actor, the mailbox needs to store any message type. - solved it with Box> which erases the concrete type but preserves the ability to call handle() through a vtable. - then came request response. fire and forget is easy with do_send(), but send() needs to return a future that resolves to the result. used oneshot channels for this, the envelope carries the sender half and the caller awaits the receiver. -fault tolerance took a while. wrapped message handling in catch_unwind() so actor panics dont crash the whole system. added death watch too, actors can watch other actors with ctx[.]watch() and get notified when they die. this is the foundation for supervision trees later -timers and scheduling came next. ctx[.]run_later() for delayed messages, ctx[.]run_interval() for periodic ones. both return handles you can cancel. internally they spawn tokio tasks that send messages back to the actor . - then now i did async handlers. actors can now await inside message handlers for network requests or database calls. needed BoxFuture<'a, T> which is Pin + Send + 'a>>. pin is interesting, finally used it for real. async blocks compile into state machines with self references so you cant move them after polling starts.

@realmcore_ this ? github.com/badlogic/pi-mo…

Are there any good agent sdks built in typescript? I mean for running agents, not building them

@Kimi_Moonshot this is fun :) thread: Click the link to view conversation with Kimi AI Assistant kimi.com/share/19acf51b…

🔥Black Friday Deal $0.99 for one month of Kimi Pro + Kimi For Coding (Moderato plan, with agentic slides 🍌coming very soon)?? (But only if you can gaslight Kimi.) To unlock the lowest price, you need to bargain with Kimi, our very cute (and very stingy) deal guard. Top bargainers get the monthly plan for $0.99. Basically FREE. Think you can beat the odds? Start your challenge now 👉kimi.com/kimiplus/sale 🎁 Extra bonus (For 5 Days): Follow, like, retweet with your bargain chat. Spiciest bargainers get $10 API credits in DMs.

@MaximeRivest I feel you ! I like quarto but I also want to be able to modify the code abit and see the result instantly. Now looking at marimo, although it is just for Python

@catethos Yes you get it! Do you love that flow? I did not.. I find the Jupiter notebooks too segmented. Like I can't do ctrl a and ctrl C and get the whole notebook. So I am working on something that is closer to quarto and rmarkdown but with the code output in the document.

Gemini 3 and I designed this general REPL bridge library yesterday called brepl. In one hour this morning, Claude Code, Gemini, and I made the library and pushed it to PyPI. With brepl, you can get LLMs (any MCP clients) to use CLIs and TUIs and get their autocompletes. What a time to be alive! > claude mcp add uvx brepl link to repo below. ps: that library is a dependency I needed for something quite nice I am cooking! Stay tuned :)