Tool
PLUR Engrams — Shared Episodic Memory Across Hermes Agents
Brain-like persistent memory system enabling shared episodic memory across multiple agents. Community project for cross-agent knowledge sharing.
Quick answer
PLUR Engrams is a community project that gives multiple Hermes agents shared, brain-like episodic memory, so knowledge written by one agent is retrievable by others. It targets multi-agent setups where Hermes' default per-agent memory would otherwise silo what each agent learns.
Hermes memory is per-agent by default. PLUR Engrams is a community answer to the multi-agent case: a shared episodic store so several agents can read from a common pool of experience.
Features
- ✓Shared episodic memory
- ✓Multi-agent support
- ✓Persistent storage
- ✓Cross-agent knowledge
Why this tool matters
Default Hermes memory keeps each agent's knowledge local to that agent. That is correct for a single assistant, but it silos learning the moment you run a team of agents that should share what they discover.
A shared episodic layer is exactly the gap the community keeps hitting with multi-agent setups: people report that naive shared-memory approaches fail because each agent only writes to its own store, defeating the automatic flow. A purpose-built shared store addresses that coordination problem.
Shared memory raises questions default memory does not — what is global versus private, and how to avoid one agent's noise polluting another's recall. Treat the shared store as deliberate infrastructure, not a drop-in, and decide what should be common knowledge.
As a community project layered on Hermes, PLUR Engrams complements rather than replaces the built-in memory. Single-agent users should stick with the default; the shared layer earns its complexity only when multiple agents genuinely need a common memory.
Best use cases
FAQ
Built-in memory is per-agent. PLUR Engrams adds a shared episodic store so multiple agents can read from a common pool — solving the silo problem that appears in multi-agent setups.
No. For one assistant, the default per-agent memory is the right choice. A shared layer earns its extra complexity only when several agents genuinely need common knowledge.
Deciding what is global versus private, and keeping one agent's noise from polluting another's recall. Treat the shared store as deliberate infrastructure and define what should be common knowledge.