agent memory keeps pulling me back

i keep coming back to agent memory, code intelligence, trajectories, evals, and why helpful models start agreeing too much.

questions still open

i keep asking what an agent should remember or forget, how a repository map changes its coding path, and when helpfulness becomes plain agreement.

where i test it

atlas tests structural code context, weave handles memory, and mayspace, aiproxy, orchestrator, and weppi test routing, tools, and interaction from different sides.