The top community-built memory and context management tools for Hermes Agent, ranked by GitHub stars. Memory is what makes Hermes self-improving — these tools extend its built-in recall with long-term semantic memory, graph-based retrieval, and cross-session persistence.
For the architecture behind these tools, read The Hermes Agent Memory Guidebook.
This official Hermes Agent memory provider offers a universal memory layer available as a managed cloud or Apache-2.0 self-hosted solution. It supports multi-level memory for User, Session, and Agent states through temporal reasoning and multi-signal retrieval.
An official Hermes Agent memory provider that organizes context as a virtual filesystem using viking:// URIs. It utilizes a three-tier L0/L1/L2 loading system for on-demand retrieval and supports session archiving via background memory extraction and merging.
This open-source platform provides persistent long-term memory across multiple agent sessions using a self-hosted knowledge graph engine. It builds knowledge graphs from text and code and supports automatic routing for graph, vector, and code retrieval.
This brain for Hermes Agents stores facts with provenance, allowing for corrections and memory withdrawal. It builds a knowledge graph with typed edges for entity relationships and generates synthesized answers featuring citations and gap analysis.
An official Hermes Agent memory provider that maintains user profiles with stable facts and recent activity. It features a custom vector graph engine with sub-300ms recall and provides connectors for Google Drive, Notion, GitHub, and OneDrive.
This memory system focuses on long-term retain, recall, and reflect workflows. It supports over 25 LLM providers and is available via Python and Node.js SDKs or through deployment via Docker, Kubernetes, pip, or managed cloud.
This memory library serves as the upstream component for the elkimek/honcho-self-hosted Hermes wrapper. It is implemented as a FastAPI server that provides reasoning-first memory by extracting conclusions from conversation history.
An official Hermes Agent memory provider that manages a markdown context tree using Git-like version control. It features a 5-tier retrieval system with sub-100ms latency and includes an interactive TUI and REPL interface.
A zero-dependency memory system that requires only one pure-Python dependency and a SQLite database. It features episodic compression for sub-linear storage growth and supports MCP, Python SDK, and native tool integrations.
This seven-layer memory system includes vector search, structured facts, and an auto-curated wiki. It utilizes a Ground Truth hierarchy to ensure agents prioritize injected context within a local infrastructure of Docker, Qdrant, Redis, and SQLite.
This recursive context harness helps agents succeed on complex tasks by executing tasks through iterations to retain lessons and artifacts. It stores immutable context bundles and outcome-gated promotion data in the filesystem.
A lossless context management plugin that uses a hierarchical DAG and SQLite store to ensure no message is lost. It provides specific recall tools, such as lcm_grep and lcm_recall, for detailed message recovery.
This tool indexes existing local history from over 20 coding agents to enable millisecond lookups across months of session data. It automatically strips keys and tokens during indexing to ensure security while searching gigabytes of past session history.
This self-hosted backend provides private cross-session memory for Hermes Agent using Docker, PostgreSQL, and Redis. It supports any OpenAI-compatible LLM provider or local inference server for flexible, persistent memory management.
Plur utilizes an open engram YAML format to store memory as plain-text files for local ownership and multi-agent use. It employs a hybrid of BM25 and embedding search to provide local-first retrieval for MCP tools like Claude Code and Cursor.
This local-first layer manages identity, memory, and secrets with 97.6% answer accuracy on LongMemEval benchmarks. It constructs memory automatically from Discord, Obsidian, and office documents, integrating directly with Hermes Agent and Claude Code.
ClawMem is an on-device context engine that operates locally without API keys or cloud dependencies. It utilizes hybrid retrieval, including BM25 and vector search, to integrate with agents via MCP and Claude Code hooks.
This five-package plugin family provides durable, file-based memory using a hierarchical SQLite schema and FTS5 search. It functions without vector databases or embedding models, supporting Linux, macOS, and Windows environments.
This project optimizes context through block reordering and deduplication to reduce prefill latency by 1.5–3×. It offers native plugin support for Hermes Agent to save up to 36% in token usage during inference.
This plugin provides explainable recall by tagging every retrieved result with a 'why_retrieved' marker and resolving memory contradictions. It supports an embedded in-process backend, making it suitable for serverless deployments.
Agentcairn uses an Obsidian vault as a source of truth, storing durable memory as inspectable Markdown files. It employs a local DuckDB cache for hybrid search, ensuring memory remains daemonless and user-owned.
This CLI serves as a machine interface for agent-driven knowledge maintenance, preserving citations and temporal memory without vector databases. It supports portable archives, allowing agents to autonomously recall and evolve knowledge across sessions.
Brainstack integrates Hindsight, Graphiti, and MemPalace into a single memory owner for Hermes Agent. It manages state and retrieval using SQLite, Kuzu, and Chroma to provide agents with purpose-built evidence packets for each query.
This anticipatory memory system prefetches context by watching agent session logs and storing data locally via SQLite-vec. It is an MCP-native server that uses GGUF models to provide RAG and vector search capabilities.
This memory layer resolves live workspace states into verified context files to reduce prompt token usage by up to 94%. It provides local-first session history tracking and supports MCP-native integration via stdio or SSE servers.
This reflective memory system utilizes semantic search with BM25 full-text ranking and graph traversal to manage agent knowledge. It features automated indexing for codebases and multimedia with OCR support, using edge tags to create bidirectional relationships between notes.
Designed to reduce schema overhead, this tool uses local BM25 ranking to select relevant keywords for Hermes Agent v0.14.0. It includes a dedicated dashboard for tracking estimated token savings and supports native tool search functionality.
This Git-backed context layer manages workflows for coding agents like Claude Code and Codex using hash-checked proposals and receipts. It facilitates reviewed context updates and provides a source-neutral migration path for chat and document exports.
This local-only plugin uses a single .lbdb embedded graph database to store memories with a 1-10 importance ranking. It prioritizes high-value information during context prefetch using BM25 keyword search and optional GLiNER2 entity extraction.
This episodic memory plugin implements a pull-model that supports immediate event deletion and an on-demand summarizer. It maintains a trace.jsonl audit log to provide a transparent record of every prefetch and memory operation.
This SDK and MCP server enables agents to handle x402 payments using USDC and stablecoins across EVM and Solana chains. It functions as a budget-bound wallet that operates without the need for a backend, database, or API keys.
This plugin indexes local artifacts such as scripts, cron jobs, and documentation to provide capability routing for Hermes Agent. It registers five native tools for searching and fetching artifacts while automatically generating routing notes.
This provider enables durable cross-session memory through pre-turn context injection and session-end snapshots. It utilizes a local SQLite reliability layer and a write outbox to manage tools for searching and updating agent state.
Integrating Exabase M-1, this plugin provides self-organizing long-term memory with tools for manual storage and retrieval. It supports configurable query expansion and result reranking to optimize how the agent accesses stored information.
This plugin provides private local code memory for repositories, requiring a local SourceVault server for indexing and vector storage. it adds deterministic slash commands and LLM-callable tools for semantic search and grounded Q&A.
This AI-native memory solution offers AES-256-GCM encryption at rest and supports HNSW ANN, sparse, and exact-text recall. It can be executed via native Rust FFI or an HTTP daemon, featuring MinHash deduplication and metadata filters.
This provider offers 17 tools for memory management and knowledge graph traversal, utilizing a hybrid search approach that combines BM25, vector, and graph methods. It supports automatic context checkpointing during message compression to preserve information integrity.
This file-based kit manages memory hygiene through a git-backed directory of versioned markdown files rather than a traditional database. It features 'rot' and 'lint' curators to identify stale or conflicting data, allowing users to review evidence-cited diffs via five specific slash commands.