n8n, not ZapierOrchestration
Zapier-per-task pricing inverts past $400/mo. n8n on a dedicated VM stays flat as volume scales. Visual workflow builder so your team can audit the logic. Full data sovereignty.
Buyer’s guides tell you to demand the tech stack upfront. So here it is. Not a Zapier shop. Not a single-platform reseller. Self-hosted orchestration and custom code where the edges break.
Stack rationale
Open-source where the cost curve flattens. Custom code where the edges break. Self-hosted where data sovereignty matters.
Zapier-per-task pricing inverts past $400/mo. n8n on a dedicated VM stays flat as volume scales. Visual workflow builder so your team can audit the logic. Full data sovereignty.
OpenAI is the default. Groq’s hosted Llama-3 hits sub-second response times on production workloads where users wait. We reserve Anthropic for reasoning depth, Groq for everything else.
Agent state, audit logs, retry queues. One Postgres on Neon for everything we control. Vector store added only when retrieval cost crosses the per-call SQL break-even.
If we vanish, your stack still runs. Every workflow exports. Every agent prompt lives in your repo. The retainer is operations, not lock-in.
Architecture
Inbound signal hits orchestration. The agent decides. The operator audits the edges. Postgres records every state change. Downstream systems read from one source of truth.