Continuous Direction

[parcadei]

Continuous Cloude V3 is a cathedral. The new version is a system designed to keep up with 4.7. It work well for 4.6, as well other models (Codex, MinMax, Kimi) but I'm keen to understand how everyone else has been getting on.

Obviously I've been a tad AWOL between working on memory and security for agents - so what I want to basically ask is: how have you been?

I'll work through the open issues and pull requests tomorrow but there's a fork in the road because there's a lot here I don't use, generally because I'm dogfooding the 'harness experiment of the week' - but I know people get results from the agents, skills and memory setups here.

Over 3,000 people have enjoyed version 3, so I wanted to ask: what did you like? What did you not like? What changes have you made personally? What changes would you like to see? What is something that you struggle with? What infuriates you about current systems?

Let's open a forum so we can develop an idea of where to take this because Continuous is nothing without it's crew.

My aim is to release Continuous in a few months now that I've figured out a lot of problems and have a clear roadmap, it's something between a continuously learning agent harness designed to be security-first and token friendly, a multi-model management system and unlocking agent identity.

The last one sounds LLM psychosis-like, i know - but it's less about "it's conscious" and more about the interesting thing that happens when memory unlocks a kind of identity persistence.

But anyways, let's chat!

[stephenfeather]

You are welcome to pull inspiration(?) from my fork of OPC (https://github.com/stephenfeather/opc/).
Also some helper scripts and skills at https://gist.github.com/stephenfeather
It has drifted from CC3 a bit.

Cherry pick what you want.