A 7-layer memory operating system for Hermes Agent, Memory OS addresses the critical issue of AI agent amnesia by providing a persistent, local memory infrastructure. Designed for power users who require agents to evolve and retain context, it offers API-agnostic memory capabilities, enabling agents to act as true long-term collaborators without vendor lock-in or subscription fees.

How It Works

This system employs a sophisticated 7-layer architecture, integrating flat files, SQLite databases (with FTS5 and trust scoring), a heavily forked Icarus plugin for cross-session recall, and a Qdrant vector database with a 4-level fallback cascade. A key innovation is the "Ground Truth Hierarchy" (Layer 7), which explicitly instructs the agent to treat injected memory as authoritative, preventing token-inefficient rediscovery. Memory is captured, stored, and injected surgically, gated by relevance and deduplication, ensuring only necessary context reaches the LLM.

Quick Start & Requirements

• Primary Install/Run: Requires Hermes Agent and Docker for its local memory stack (Qdrant, Redis, ARQ Worker).
• Prerequisites: Python 3.11+. Compatible with any LLM provider supported by Hermes Agent, including OpenRouter, OpenAI, Anthropic, Ollama, and local models.
• Infrastructure: Runs entirely on the user's machine, offering local, private memory.
• Links: Setup guide, Layer deep-dives, Infrastructure docs.

Highlighted Details

• Comprehensive 7-layer memory system featuring Qdrant hybrid search, structured facts with trust scoring, and an auto-curated wiki pipeline.
• The critical "Ground Truth Hierarchy" (Layer 7) ensures injected memory is utilized by the agent, overcoming "memory-zero" behavior.
• Surgically token-efficient context injection through gated retrieval, per-session deduplication, and social-closer filtering.
• Local, API-provider-agnostic infrastructure with no memory subscription or vendor lock-in.

Maintenance & Community

Developed by an individual user focused on practical application within the Hermes Agent ecosystem. No explicit community channels (e.g., Discord, Slack) or details on corporate sponsorship are provided.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Designed for local production use with Hermes Agent and compatible with various LLM providers.

Limitations & Caveats

Requires Hermes Agent and Docker. The bundled Icarus Plugin is a non-upstream-compatible fork. The project's strong emphasis on local-only infrastructure may not suit users seeking cloud-based memory solutions.