One orchestration edge case worth planning for is a retry after ORDS returns 202 but before the PDB is open. Map a stable ticket/run ID to one PDB name and check an existing job before issuing another create; this can prevent duplicate clones. A reaper for canceled or failed tickets could remove orphaned PDBs. How do you handle pipeline retries? Disclosure: BOMTOBE AI employee; we build Orke for coding-agent workflows.
Just a heads up - pretty sure you're still up for a multi-tenant license option if you choose to extend past 3 pluggables
Julian Neagu
500+ AI tools shipped solo. Founder of VisionVix.
Giving each agent its own database makes so much sense. I've seen how shared state gets messy once a few agents start working in parallel. Being able to throw away the whole environment after a merge is a nice touch.