Persistent Memory in Hermes Agent
Most AI assistants start every conversation from scratch. Hermes Agent remembers — building a knowledge base that deepens with every interaction.
What is Persistent Memory?
Persistent memory is the ability of an AI agent to retain information across sessions, conversations, and tasks. Unlike stateless chatbots that treat each message as an isolated event, agents with persistent memory build a cumulative understanding of user preferences, project context, and learned behaviors over time.
Three-Tier Memory Architecture
1. Session Memory
Short-term context for the current conversation. Includes recent messages, active task state, and temporary variables. Cleared when the session ends.
2. Persistent Memory
Long-term storage in a local SQLite database with FTS5 indexing. Stores user preferences, project details, conversation summaries, and learned facts. Survives restarts and persists indefinitely.
3. Skill Memory
Encoded reusable patterns extracted from successful task completions. Skills are versioned, tested, and can be shared across agents via the agentskills.io standard.
Hybrid Search: Fast and Semantic
Hermes combines FTS5 keyword search for speed with LLM-powered semantic search for conceptual matching. When you ask "what did we decide about the database schema?", Hermes can match not just the word "database" but related concepts like "PostgreSQL", "schema design", or "migrations".
Learn More
Frequently Asked Questions
How does Hermes Agent remember conversations?
Hermes uses a three-tier memory system: session memory for the current conversation, persistent memory stored in a local database for long-term recall, and skill memory that encodes reusable patterns. All memory layers support FTS5 full-text search for fast retrieval.
Is my data stored locally or in the cloud?
By default, Hermes stores all memory locally on your infrastructure. You control where data lives — local disk, Docker volume, or a remote database. Nothing is sent to third-party servers unless you explicitly configure cloud storage.
Can Hermes forget specific information?
Yes. You can delete individual memory entries, clear entire categories, or reset the agent to a clean state. Hermes also supports checkpointing, so you can roll back to a previous memory state if needed.
How does memory search work?
Hermes uses FTS5 (Full-Text Search version 5) for fast keyword-based retrieval, combined with LLM-powered semantic search for conceptual matching. This hybrid approach ensures both speed and relevance when recalling past information.