Core services and supporting infrastructure for Premortem v0.1.0, including trust boundaries and release guardrails.
Who this is forEngineers integrating with or extending the platform.
Core stack
- Supabase / Postgres for product data, auth-adjacent storage, and RLS-oriented multi-tenant ownership.
- Prisma for application data access and typed repositories.
- Cloudflare Workers via Wrangler for API edge entrypoints.
- GitLab as the primary issue publishing and repository provider.
- MCP Toolbox for Databases for safe SQL-oriented agent database access.
- Gemini as the default LLM path for audits and agent flows.
- Neo4j as the graph persistence layer for repository structure and risk context.
Supporting services (next)
- Cloudflare Queues for async audit fan-out.
- Cloudflare R2 for graph snapshot exports and evidence bundles.
- Upstash Redis or Valkey for idempotency and short-lived orchestration state.
- OpenTelemetry + Grafana/Tempo/Loki for traces, logs, and metrics.
- Microsoft Entra ID if enterprise SSO becomes required.
Operational guardrails
- Production mode requires real auth, real provider keys, and explicit release checks before publish.
- Audit generation uses a shared workflow contract: grounded evidence, base-rate checks, and an empty envelope when no critical risk is present.
- Graph payloads fail closed when mandatory nodes or relations are missing.
- Agent output is validated, clustered, and reviewer-approved before GitLab publish.
- Prompt evaluation gates run before deployment and block regressions below the accepted threshold.
Repository deep dives
- ADR 0001: canonical product and system design (full architecture authority).
- Graph strategy: docs/architecture/graph-strategy.md.
- Observability: docs/architecture/observability.md.
- Queue runbook: docs/runbooks/queueing.md.
- Enterprise readiness: docs/architecture/enterprise-readiness.md.
View source on GitHub
Was this article helpful?
