this is really cool
hadn’t thought enough about agent auth this way
sometimes the agent should act as itself. sometimes it should act as me
making that a choice in the runtime makes a ton of sense
been really enjoying watching openwiki evolve
first self correcting memory and now making the work itself durable
finish something → checkpoint it → fail somewhere else → pick back up
really like where this is going
really really like this
memory is way more interesting when it doesn’t just remember, but knows when something it remembers might not be true anymore
tying claims back to evidence and rechecking when that evidence changes feels like the right direction