Technology Architecture
Technology Architecture
Forge is deployed as a Kubernetes‑based micro‑service mesh split into four logical planes, giving us fast iteration for the product team and hard isolation for the AI workloads.
1. Experience Plane
Next.js / React SPA delivered via Cloudflare’s global edge.
OAuth2 Gateway hands out short‑lived JWTs (and wallet attestations) that flow through the mesh.
2. Application Plane
Brand‑API (FastAPI, Python) orchestrates brand kit generation (logo, palette, copy).
Site‑API (Go) streams incremental static builds to Vercel for sub‑10‑second site deploys.
Funnel‑API (Node/TS) templates multi‑step funnels and syncs configs to the user’s CRM.
3. AI Plane
Model Router chooses GPT‑4o for long‑form copy or a diffusion cluster for visuals.
Prompt Compiler injects brand context, safety filters, deterministic seeds.
Vector Store (pgvector) stores embeddings so every asset remains on‑brand across sessions.
4. Infrastructure & Web3 Plane
Event Bus (NATS JetStream) guarantees at‑least‑once delivery of user events and model outputs.
Token Gateway acts as a thin proxy to pump.fun’s immutable bonding‑curve program on Solana. During the curve phase it forwards mint calls; once the token auto‑migrates to pump.swap, it still uses the same SPL mint/transfer primitives. For every brand‑creation or subscription purchase inside Forge, the Gateway signs an SPL
burn()tx that permanently removes the required $FORGE from supply and logs the burn ID off‑chain for analytics.Secrets in HashiCorp Vault; OPA Gatekeeper enforces runtime policy.
Key Properties
Idempotent jobs – every AI task carries a UUID + checksum so retries never double‑charge tokens.
Cold‑start < 300 ms – functions are pre‑warmed in a KNative pool tuned by Prometheus.
Last updated