When I decided to turn OpenClaw into a SaaS product

# Goals * Use OpenClaw to automate the company's operational workflows. * Humans should only need to send a single instruction. # Version 1 * Python scripts are responsible for calling APIs. * OpenClaw is responsible for reasoning and executing the workflow by running those scripts. # Problem As business data volume increased, the context became too large. OpenClaw's execution began to drift away from the intended workflow. # Version 2 * Python scripts are responsible for both API calls and reasoning. * OpenClaw is only responsible for executing scripts. # Problem The scripts require about three hours to complete. OpenClaw has to continuously poll for progress, which creates several issues: * Tool timeouts * The system may terminate processes that generate excessive polling traffic # Attempts I tried using background tasks and heartbeat mechanisms, but neither approach was able to solve the problem. # Reflections I wanted the workflow to live in a document so that as the organization evolves, I could simply modify the text rather than rewrite code. In practice, however, OpenClaw is clearly not capable of reliably handling a task that runs for three hours. Perhaps I shouldn't expect OpenClaw to execute such a complex workflow from scratch. Instead, I can convert the workflow document into a scheduler script, and OpenClaw can execute that script successfully. At that point, the system becomes a SaaS product, and OpenClaw serves as the entry point to that SaaS.