Two agents that handle daily ops — and heal themselves when something breaks.
A solo-operated company still has a company's worth of operations: orders, fulfillment, integrations, monitoring, decisions. I needed that work to run without me babysitting it — and to survive redeploys, model outages, and edge cases.
So I built a two-agent workforce on OpenClaw. KAVO handles strategy and operations; Rex is a Shopify orders & fulfillment specialist. Each runs on Claude with a DeepSeek fallback — on an Anthropic error or timeout the agent fails over to DeepSeek and reverts on the next clean call — carries long-term memory for continuity, and reconfigures itself from environment variables on every deploy — no setup wizard, no manual steps. They reach real systems through MCP tools and report into Slack, where a human can step in. The unglamorous parts — fallbacks, auto-heal, an exec allowlist, device pairing — are what make it safe to leave running.
Slack ◀───────────────▶ Operator (human-in-the-loop) │ │ ▼ ▼ ┌───────────────────────┐ ┌───────────────────────┐ │ KAVO · strategy + ops│ │ Rex · Shopify ops │ │ ops · integrations │ │ orders + fulfillment │ └───────────────────────┘ └───────────────────────┘ primary: Claude primary: Claude fallback: DeepSeek fallback: DeepSeek │ │ └────────────┬─────────────┘ ▼ MCP tools · long-term memory (RAG) auto-heal · exec allowlist · self-config from env │ ▼ boot: onboard → configure → gateway → run every redeploy, zero manual setup