Dream Auto

What Dream Auto does

Dream Auto indexes your Hermes sessions, scanning conversation history for topics, errors, open questions and complexity signals. An LLM then grades each session between 0.0 and 1.0 for how much it is worth thinking about again. When the machine is idle, a scheduler starts background dreams, and the troubleshooting table says it only does so while CPU and RAM are both under 70 percent.

The thinking engine is a Monte Carlo tree search with a two-tier agent. Tier 2 rollouts, which can use tools, fire when Tier 1 confidence on the best branch falls below 30 percent. The plugin registers six hooks, a scheduler with a queue manager and a wall-clock killer, and a command-line insights dashboard. Linux uses a systemd user service, macOS uses hermes cron through launchd, Windows works inside WSL, and native Windows is not supported. The plugin is enabled with DREAM_AUTO_ENABLED=1.

Key features

• Indexes past sessions for topics, errors and open questions
• LLM grading of each session for dream potential
• Monte Carlo tree search engine with a two-tier agent
• Resource-gated scheduler that waits for CPU and RAM under 70 percent
• Command-line insights dashboard and six plugin hooks

When to use it

• Revisiting unresolved problems from earlier Hermes sessions without prompting
• Learning from errors that were fixed and then forgotten
• Running background analysis only when the machine is idle

Who it is for: Hermes Agent users interested in autonomous background reflection on their session history.

How it fits with Hermes Agent

It is a Hermes Agent plugin, built on Hermes hooks, hermes cron and a bundled skill, for Linux, macOS and Windows WSL.

Requirements: Hermes Agent on Linux, macOS or Windows WSL (native Windows is not supported), the psutil and rich Python packages, and DREAM_AUTO_ENABLED=1

Note: The repository is archived and read-only.

FAQ