I DID IT, I SHIPPED THIS IN @anymo_ai
this is so fun lol, thanks for the idea @maria_rcks
If y'all like this I can ship a codex plugin to customise your codex UI with this (I think there's a way) what do you think @thsottiaux
I added model verifiers to @anymo's meta harness!
AI agents now check their own work and won't mark a task "done" if the output is actually wrong or incomplete, with the checker tuned against a real model so it catches mistakes without flagging good work.
it's a separate judge + reality-checks it can't fake + measuring whether the checker is any good. anymo now verifies agent work for real. This is important to maintain quality as an orchestrator.
Stay Tuned π
@prz_chojecki People don't know how much cool stuff can be done with open source models like Kimi k2.7 code. To solve that, I'm building @anymo_ai (opensource harness for the biggest unlocks that models are limited to)
Anymo AI just built and launched its own documentation!
It's now live. I'm close to launch this project, stay tuned and follow @anymo_ai for updates.
This documentation is still an early version and will continue to evolve. We have some exciting features planned, and as we move toward a stable release, weβll keep making the docs clearer, more complete, and more reliable.
Anymo will be writing loops for self-updating documentation as new releases follow!
Can't wait to see what you do with Anymo, We'e building the second computer for agents, we're getting there step by step. Initial launches are designed to gain feedback, ideas to ship better things, this will be completely open-sourced so devs can contribute and personalize!
Looking forward for the launch, if we see a good response, there are so many things we'd like to explore that people thought wasen't possible/limited internally. We've seen advancements in orchestration where mac users can call claude's frameworks for multi-step reasoning
3/ This will be post-beta. Once it's stable we'll start to introduce basic plans like CoLab Go, and a few interesting things where users will be able to use some of our served opensource models and use API keys as well. In CoLab, your team will be able to work and manage together
2/ The second will be an official launch, we'll be adding capabilities needed specially in the coding harness, base models, generous limits (basic) and some improved features including 3rd party providers that are helping us make this happen, in this phase we'll monitor prod.
π§΅We've planned Anymo's launch in phases.
1/ The first will be the opensource, you will be able to self-host it and use it yourself, helping us improve the software and fix any bugs and make it frictionless across Windows, Mac and Linux
HOLY, THIS IS CRAZY.
@alexalbert__ pushed me to try claude fable 5 for the most ambitious work in the announcement video, I thought let's try creating a launch video for @anymo_ai (opensource for kimi work, self learning capabilities)
I gave it context, it one shotted it. This is literally crazy and shocking, fable is really too good @ClaudeDevs
(Note: this is not the official launch video)
Seeing this model's capabilities and what it can achieve, I can't imagine with additional context and tweaks how good it can do, I'm really proud of the team @bcherny@trq212@karpathy@adocomplete@nc_frey and everyone who worked on this