Runtime Topology
This page summarizes the runtime identities that matter for developers, partners, and operators.
For canonical state, also review:
docs/CANONICAL_STATE.mddocs/ops/RUNTIME_TOPOLOGY.mddocs/internal/operations/API_REFERENCE.mddocs/migrations/DE_SUPABASE_MIGRATION.md
Current cloud-native runtime
IntegrateWise is 100% Cloudflare product plane for production runtime.
Primary bindings:
- D1 for structured state and metadata.
- KV for hot projections, cache acceleration, and reconstructable state.
- R2 for large immutable or versioned artifacts, evidence objects, and raw payloads.
- Vectorize for semantic retrieval vectors.
- Durable Objects for per-tenant runtime coordination and Twin session state.
- AI Gateway as the policy-bound inference runtime surface.
Supabase references remain only during migration. Do not add new Supabase logic, RLS talk, or dual-writes.
Core runtime services
The platform runtime is organized around bounded service responsibilities rather than OODA microservice boundaries.
Tier overview:
- Gateway: entry, auth, request routing, projection facade, and public API surface.
- Pipeline: ingestion, normalization, Spine hydration, delta sync, and state continuity.
- Connector: connection lifecycle, OAuth, transport adapters, webhook ingress, and capability health.
- Intelligence: signal generation, trigger policy, Twin orchestration support, and proposal shaping.
- Knowledge: chunking, embedding, semantic search, topic sync, and ingest routing.
- Governance: pre/post proposal gates, approval workflows, policy enforcement, and audit.
- Twin runtime: runtime-bound Twin session coordination.
Data substrate rules
These rules are hard:
- D1 owns structured state and metadata for current runtime.
- KV owns reconstructable hot projections and caches.
- R2 owns bytes, raw artifacts, large payloads, and evidence objects.
- Vectorize owns semantic retrieval vectors, not source truth.
- Embeddings are versioned derived representations and may always be rebuilt from source truth.
- Product logic talks to stable repository contracts, not directly to D1, KV, R2, or Vectorize.
- No product plane depends on a provider primitive.
Developer-facing boundaries
When building against IntegrateWise:
- Use the gateway routes in
docs/internal/operations/API_REFERENCE.mdas the stable integration surface. - Do not assume direct database access from custom code.
- Do not assume local filesystem or container-local persistence for production runtime.
- Assume multi-tenant isolation at the query layer.
- Assume secrets are resolved just-in-time and not exposed to config or logs.
Partner integration expectations
Partners should plan around async, governed, observable integration behavior.
Expect:
- webhook ingress under strict signature verification.
- capability availability changes surfaced through registry health.
- pay-as-you-go rate and token limits enforced at adapter or registry layer.
- failover behavior determined by configuration, not runtime guesswork.
- incident evidence packaged in audit events, approval records, proposal history, and outcome events.
Deployment orientation
IntegrateWise isolates compute capabilities from deployment environments. Core engines are packaged as immutable runtime units. Deployment manifests determine whether capabilities live in a unified container profile or mapped isolated instances.
Edge routing handles light read tasks and context projection updates. State isolation is enforced by tenant keys.
For deployment and rollback behavior, see docs/site/content/41-deployment.md and docs/internal/operations/DEPLOYMENT_RUNBOOK.md.