This is one of the approaches im considering for my own, Roder.
The approach mostly being communicating over json rpc which has become the standard for MCP so it makes it more approachable to agent developers.
Obviously its very much NOT mcp, its a low level events based rpc system for registering capabilities and extending low level primitives of the agnet itself not the model
I'm actually very close to being ready to release exactly that also in rust. I completely agree with your statement, extensibility is the most importnat feature.
I've been experimenting with providers offering similar functionality for the last year and it is really vastly superior experience this codex like approach than cursor, devin etc
I posted about these guys last year when they published at the time what was a sota swebench score in summer 2024. Does anyone have insight into what they are doing to achieve this?
I posted a blog post by these guy's last year where they also claimed to have a SOTA swe-bench score in summer 2024. They seem to be reporting even better scores than o3. Does anyone have insight into what they are doing? Is this legit? Or are they eval maxing?