Hey, thanks for creating Autolith! I tried it out on my projects, and love granular self-improving capabilities, they match precisely the style I enjoy in my other favourite tools. For example, Emacs =) I previously used Pi Coding Agent, and integrated it into Emacs via Pi's RPC protocol, it worked like a charm. Pi allows self-improvements by exposing knowledge about harness to the agent and an ability to write extensions, but it's not as granular and smooth as being able to patch the harness at runtime.
My question is: what do think is the best way to approach driving Autolith externally, for example from Emacs? To clarify, the idea is not to run AL in a shell or comint buffer relying on AL's clinedi-based UI, but rather have a dedicated Emacs mode as an UI for AL, while talking to the process running in the background.
Do I just invent something ad hoc and keep it in my image, or do you plan a canonical RPC?
Hey, thanks for creating Autolith! I tried it out on my projects, and love granular self-improving capabilities, they match precisely the style I enjoy in my other favourite tools. For example, Emacs =) I previously used Pi Coding Agent, and integrated it into Emacs via Pi's RPC protocol, it worked like a charm. Pi allows self-improvements by exposing knowledge about harness to the agent and an ability to write extensions, but it's not as granular and smooth as being able to patch the harness at runtime.
My question is: what do think is the best way to approach driving Autolith externally, for example from Emacs? To clarify, the idea is not to run AL in a shell or comint buffer relying on AL's clinedi-based UI, but rather have a dedicated Emacs mode as an UI for AL, while talking to the process running in the background.
Do I just invent something ad hoc and keep it in my image, or do you plan a canonical RPC?