Yeah conversational assistant and productivity assistant are quite different.
Regarding long-term context, the fallback behavior is writing all of them to a file and let codex use tool to retrieve them. And the harness give the agent full flexibility to manipulate the memory structure by itself
How does this compare to nanoclaw, which is based on claude code?
nanoclaw closer to an actual project that ships with many pre-built functionality (ignoring the legal risk of using claude code).
seedbot is trying to be as minimal as possible - getting rid of everything except coding, and let agent self evolve according to user instruction
[dead]
Yeah conversational assistant and productivity assistant are quite different.
Regarding long-term context, the fallback behavior is writing all of them to a file and let codex use tool to retrieve them. And the harness give the agent full flexibility to manipulate the memory structure by itself