ENGINEERING BLOG
The technical decisions behind xopc, and what we learn while building a personal AI assistant.
After you approve an email, can an agent change the recipient and send it anyway? How xopc binds approval to an account, action, and arguments—and why one-time approval cannot solve every retry problem.
Read article →
After an interrupted tool call, a conversation can look perfectly normal while the next model request fails. How xopc pairs tool calls with results, trims browser screenshots, and rebuilds context after compaction.
Read article →
What happens inside an agent when you say, “Do it this way from now on”? A look at how xopc identifies facts, handles corrections and expiry, and selects what belongs in the next answer.
Read article →