OpenClaw Update 4.9: Dreaming Actually WORKS Now!

Summary

OpenClaw 4.9 introduces a completely rebuilt dreaming system that fundamentally changes how your AI agent remembers and prioritizes information about you. Instead of treating all memories equally like a basic database, the new dreaming feature works like a high-tier personal assistant by promoting important recent information over older patterns. For example, if you tell your agent you're starting a keto diet, it will prioritize that new dietary restriction over your past 40 burger orders and automatically order you bunless meals without you having to remind it every time. The update adds a dream dashboard where you can monitor different types of dreams: short-term, long-term, grounded signals, and promoted memories. The system models human sleep cycles with light sleep, REM sleep, and deep sleep phases. Light sleep tags conversations, while deep sleep identifies weighted signals and boosts their priority. This means when you mention being hungry, your agent connects that to your keto diet without explicit prompting. The dreaming feature works alongside the LLM wiki feature from previous updates, which builds a living Wikipedia about your life and preferences. While the feature is still raw and requires some manual oversight through debug files, it represents a major leap toward truly intelligent agents that understand context and changing priorities. OpenClaw now competes directly with Hermes agent, which also includes dreaming through Honcho integration (though Honcho requires payment while OpenClaw's dreaming is free). The video creators are currently using both platforms, keeping established workflows on OpenClaw while testing new projects on Hermes agent due to its self-improving skills and more verbose communication style.