I've been running the original BabyAGI for a few weeks to automate some basic customer feedback categorization, but I've encountered significant instability—tasks getting stuck in loops, the agent losing context, and occasional crashes. This led me to the community fork 'BabyAGI-plus'.
My initial tests show it's addressing some core issues. Specifically, the changes to the task execution chain and the enhanced memory management seem to reduce loops. However, I'm looking for more community data before fully migrating my workflow.
Has anyone here conducted a longer-term stability comparison between the original and the plus version? I'm particularly interested in:
* **Uptime/reliability:** For those running it persistently, what's the average session length before a reset is needed?
* **Task complexity handling:** Does it manage multi-step tasks (e.g., "analyze this churn segment, then draft an email template, then log the insight") more reliably without dropping subtasks?
* **Configuration changes:** Were there any specific .env or agent settings you had to adjust for stability, beyond the documented ones?
My priority is a predictable, stable agent for processing NPS comments and generating health score triggers, not necessarily the most cutting-edge features. If the plus fork simply makes the core loop more robust, that's a significant win.