Deployment Modes

Choose your deployment mode

Mode 1 — Hosted API

The fastest way to start. Your AI service calls https://api.datavibe.cc/v1/gate/outboundover HTTPS. Policy evaluation runs on DataVibe's shared Vercel Edge infrastructure. No deployment, no containers, no ops.

Tradeoff:The payload body travels to DataVibe's infrastructure. If your AI outputs contain PII or regulated data that must not leave your environment, use the sidecar or edge gateway instead.

import { DataVibeClient } from "@datavibe.cc/sdk";

// Hosted API — fastest onboarding, zero infrastructure
const dv = new DataVibeClient({
  apiKey: process.env.DATAVIBE_API_KEY,
  baseUrl: "https://api.datavibe.cc", // default, can omit
});

const result = await dv.intercept({ recipient, subject, body_html });

Mode 2 — Sidecar mesh (guardrail-mesh)

Deploy the guardrail-mesh Fastify service as a sidecar container or init container alongside your AI service. Policy evaluation runs in-process at localhost:3005. Payload never leaves the pod.

Policies are synced from the dashboard to the sidecar via Redis (bundle:v1:active:{orgId}). No database queries during evaluation — p50 latency is sub-5ms.

# 1. Run the guardrail-mesh sidecar alongside your AI service
docker run -p 3005:3005 \
  -e DATABASE_URL=$DATABASE_URL \
  -e UPSTASH_REDIS_REST_URL=$UPSTASH_REDIS_REST_URL \
  -e UPSTASH_REDIS_REST_TOKEN=$UPSTASH_REDIS_REST_TOKEN \
  datavibe/guardrail-mesh:latest

# 2. Point the SDK at localhost — payload never leaves the pod
const dv = new DataVibeClient({
  apiKey: process.env.DATAVIBE_API_KEY,
  baseUrl: "http://localhost:3005",
});

Kubernetes pattern: Add guardrail-mesh as a sidecar container in your pod spec. Both containers share the same pod network, so localhost:3005 is always reachable.

Mode 3 — Edge gateway (apps/edge-gateway)

Deploy the apps/edge-gateway Next.js application as a standalone Vercel project (or on Cloudflare Workers via OpenNext). You control the deployment region, the data residency, and the URL. The gateway has the same API surface as the hosted option — only the baseUrl changes.

When to use: Enterprise customers with data residency requirements (EU, GovCloud), or teams that need a customer-branded endpoint URL.

# apps/edge-gateway — deploy as a standalone Vercel project
# Required env vars:
# DATAVIBE_API_KEY (workspace API key for auth validation)
# DATABASE_URL     (Neon Postgres — same DB as dashboard)
# UPSTASH_REDIS_REST_URL
# UPSTASH_REDIS_REST_TOKEN
# INTERNAL_SECRET  (≥ 32 chars — for dashboard notification webhook)
# NEXT_PUBLIC_APP_URL (your dashboard URL)

# After deployment, point the SDK at your edge URL:
const dv = new DataVibeClient({
  apiKey: process.env.DATAVIBE_API_KEY,
  baseUrl: "https://gate.your-company.com",  // your Vercel deployment
});

The edge gateway shares your Neon Postgres DB with the dashboard (same DATABASE_URL). API key validation uses an Upstash Redis cache (L1) with Postgres fallback (L2). All gate decisions are written to gate_submissions and appear in your approval queue in real time.

Mixing modes across environments

You can run different modes in different environments with no code changes — only DATAVIBE_GATEWAY_URL differs:

• Local dev: Hosted API (no containers needed)
• Staging: Sidecar mesh (validates your Docker compose setup)
• Production: Edge gateway (customer-controlled, regional)