Support response median 57minutes, down from 1h 33minutes a day ago.
We just wrote yesterday about the Memory System, and the memory system is already better than yesterday. New deterministic, and cheap fuzzy search layer AND vastly improved performance on the first cheap layer. Shows up long term.
Everyday, new challenge arises. Today it was GET TICKET API timeouts, without researching the API timeouts were set to 30seconds. We added force searching other similar tickets, ie. the tool just returns similar tickets from past without asking. The search tooling, fetching etc. caused the get ticket request time to balloon.
Yesterday, it was noted that something flipped back to sequential single ticket processing. That ticket happened to have a fail, and backoff curve was flaky. Other tickets did not get processed.
The day before yesterday, it was trying to launch parallel jobs for multiple tickets, but all did FLOCK on single global lock for single ticket.
A week ago there was some hallucinated circuit breakers in there stopping all ticketing.
Originally the agent coded everything in bash, it was thousands of lines of code for simple launch "handle ticket #483743" command, codex rewrote and simplified for PHP. But these complexities creep in.
Please bear with us while we solve these. Yes, this is chaotic, but fast paced development. Structure arises slowly but surely, and things adapt. This is evolution! (Say it in The 300, Spartan style voice!)
Drop a ticket, and see for yourself how efficient is. Maybe ask it to deliver some ice cream for your seedbox?
Make the funniest ticket, at end make a human request on it linking this announcement, and we'll give you a little bit of credit if it makes the meatspace agents (humans) chuggle, might even use for marketing material. Join us on Discord for live heartbeat and chuggles too.
Friday, March 13, 2026
