We spent the last 6 months building our take on a chat agent SDK. It’s open source and uses the stateful compute that powers our existing tasks and workflows.
On 22-23 June our us-east-1 and eu-central-1 regions had periods of downtime and the overall incident lasted 18 hours.
This falls way short of the reliability we aim for. I’m really sorry to all our customers, who will be receiving credits for that period and more.
What
Switching from Node to @bunjavascript gave us a 5x speed boost (28x better max latency) but there was a gotcha…
Pending response promises don't resolve on client disconnect. @nodejs handles this automatically. Fix: resolve disconnects with 499.
