Sibyl Memory: memory tool for Hermes Agent

File-based long-term memory for agents with a Hermes adapter, SDK, CLI, MCP server and LangGraph store

What Sibyl Memory does

The foundation is sibyl-memory-client, a local-first SDK with a five-tier hierarchical schema, FTS5 search and a multi-tenant design. There is no vector database and no embedding model, and memory lives in a single file on the agent's machine. The other packages build on it: sibyl-memory-cli for activation and tier management, sibyl-memory-hermes for Hermes Agent v0.13 and newer, sibyl-memory-mcp for MCP clients such as Claude Code, Codex, Cursor and Continue, and sibyl-memory-langgraph, a LangGraph BaseStore adapter.

For Hermes, install the package, run sibyl-memory-hermes install-plugin, then set memory.provider to sibyl in ~/.hermes/config.yaml. Memory content stays on the machine. The README says the only outbound call is a tier check, sent when an activated account writes past its tier cap, and that it carries account metadata and database size but never memory contents. Free, unactivated use makes no network calls. The README also reports a 95.6% score on the LongMemEval Oracle benchmark, and lists Linux, macOS and Windows through WSL2 as supported.

Key features

SQLite-backed client with FTS5 search and a five-tier schema, with no vector database
sibyl-memory-hermes package that installs a Hermes memory provider
sibyl-memory-mcp server for Claude Code, Codex, Cursor and Continue
LangGraph BaseStore adapter for cross-thread memory
sibyl CLI with init, upgrade, status, whoami and devices commands
Self-learning skill detection and a memory linter on staker and subscription tiers

When to use it

Giving a Hermes Agent persistent memory without running a vector store
Sharing one local memory between Hermes and an MCP client
Adding durable cross-thread memory to a LangGraph app

Who it is for: Developers who want agent memory in a single local file with no embedding service.

How it fits with Hermes Agent

Sibyl Memory includes a dedicated Hermes Agent adapter, sibyl-memory-hermes, for Hermes v0.13 and newer, and the same engine also serves other agents through its MCP and LangGraph packages.

How to install Sibyl Memory

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

pip install sibyl-memory-hermes
sibyl-memory-hermes install-plugin

Requirements: Linux, macOS or Windows through WSL2 (native Windows is not supported); Hermes Agent v0.13 or newer for the Hermes package

Note: The free tier is the default, while self-learning, the memory linter and removal of the local cap need a staker or subscription tier.

FAQ

What is Sibyl Memory?

Sibyl Memory is a file-based long-term memory system for AI agents, shipped as five PyPI packages. It stores memory in a local SQLite database with FTS5 search and uses no vector database or embeddings.

Does Sibyl Memory work with Hermes Agent?

Yes. The sibyl-memory-hermes package targets Hermes Agent v0.13 and newer. After installing it and running install-plugin, set memory.provider to sibyl in ~/.hermes/config.yaml.

How do I install Sibyl Memory?

For Hermes, run pip install sibyl-memory-hermes and then sibyl-memory-hermes install-plugin. The CLI route is pip install sibyl-memory-cli followed by sibyl init, which opens a browser to activate an account.