What I built

SNOWAI is a bring-your-own-LLM workspace that turns any AI coding assistant into a capable ServiceNow senior administrator. You point it at an instance, and the assistant works under a ServiceNow-specific operating discipline, makes platform-native decisions, and connects to the live instance for read-only reconnaissance — with the team's discoveries accumulating in shared, versioned memory.

  • Live, read-only instance recon — a Day-1 onboarding routine plus read helpers that inventory the automation layer (flows, subflows, actions, triggers, playbooks, decision tables, business rules), snapshot a flow's current state, and search the platform knowledge base. All HTTP GET — no writes, no stored credentials.
  • 16 ServiceNow-specific skills — spanning scripting, ACL/security, business rules, Flow Designer, Workflow Studio, CMDB, catalog admin, change management, import sets, notifications, reporting, decision tables, update-set promotion, and knowledge management.
  • A platform-native decision framework — steers the assistant toward the right tool for the job (Business Rule vs Flow vs Script Include, scoped vs global) so it doesn't pick up bad platform habits.
  • Knowledge that compounds — shared, git-backed memory means an instance-specific discovery made once is available to the whole team afterward.

Enterprise-ready by default

Because this connects an AI worker to a real ServiceNow instance, I built it to enterprise security expectations from the start: OAuth bearer-token auth instead of plaintext passwords, a structured audit log of every instance operation, credentials kept out of the codebase (OS keychain, gitignored environment file, pre-commit secret scanning), SSL failures that fail hard instead of silently downgrading, and a container hardened to the CIS Docker Benchmark.

Why it matters

ServiceNow rewards people who know the platform deeply and punishes those who don't. SNOWAI packages that senior-admin judgment into a reusable worker — one that reads an instance safely before it recommends anything, explains its choices in platform terms, and gets sharper as the shared memory grows. It's LLM-agnostic, so it isn't tied to any single model or vendor.