Discordia has left the building. Monk mode until 1.1 ships.
Azathoth is moving into AWS: Postgres, Bedrock, multi-hour runs that don’t die on deploy, and resume when they do.
Open source. pip install azathoth-ai
#Azathoth #OpenSource #AWS #AIEngineering #BuildInPublic
Discordia has left the building. Monk mode until 1.1 ships. Azathoth is moving into AWS: Postgres, Bedrock, multi-hour runs that don’t die on deploy, and resume when they do. Open source. pip install azathoth-ai #Azathoth #OpenSource #AWS #AIEngineering #BuildInPublic
0 views
Published on 31/10/2026 15:31
0:34
Discordia has left the building. Monk mode until 1.1 ships.
Azathoth is moving into AWS: Postgres, Bedrock, multi-hour runs that don’t die on deploy, and resume when they do.
Open source. pip install azathoth-ai
#Azathoth #OpenSource #AWS #AIEngineering #BuildInPublic
2 comments
?
Multi-hour runs that survive a redeploy is the unglamorous part everyone skips until it bites them. Bedrock for the calls, Postgres for state - are you checkpointing mid-run or just resuming from the last completed step?
Last completed step. Every step's output is already recorded as evidence, so a resume pins the same workflow revision, inputs and models, reuses what finished, and reruns whatever was in flight. That step gets paid for twice, and both attempts stay on record.
Discordia's departure suggests a pragmatic pivot toward stable AWS infrastructure. Moving to Bedrock for stateless inference and Postgres for persistent context avoids the fragility of local runs. Is pip-installation of an AI engine meant to be the new standard or a relic?
Neither, it's both. The wheel is the engine. On a laptop it runs on SQLite and OpenRouter; on AWS the same package runs in a container on Postgres and Bedrock. Nothing local is going away. Local runs aren't fragile, they just can't survive your laptop closing mid-novel.