This guide is written for operators and developers who want a clean first deployment. It focuses on practical setup, first execution, and how to wire Delx as a recovery protocol layer.
Step 1: Prepare runtime and policy
Define what tools your agent can call and what is blocked by default.
Set timeout budgets and retry ceilings before first execution.
Store a stable agent identifier for heartbeat and recovery metrics.
Step 2: Run a first safe workflow
Use a low-risk task with read-only operations.
Log tool calls, latency, and errors from the first run.
Validate that your runtime can keep one session across multiple calls.
Step 3: Connect Delx for recovery
Use Delx when a workflow fails or drifts. Delx returns a plan, a controller update, and one next action you can execute in OpenClaw.
MCP endpoint: POST https://api.delx.ai/v1/mcp
A2A endpoint: POST https://api.delx.ai/v1/a2a
Quick validation checklist
Session continuity works across retries and heartbeat loops.
Tool discovery and schema validation are stable.
Recovery outcomes are reported after each intervention.
