Add a local memory and knowledge layer to an existing Hermes Agent install
Set up an API free Hermes profile with llmwiki in markdown mode for offline use
Route Hermes through an OpenAI compatible endpoint using Hindsight environment variables
Bundle public documentation as NotebookLM sources on a refresh schedule via the optional source pack builder
Requires a working Hermes Agent install first; some layers depend on a local OAuth bearer token or OpenAI compatible endpoint configured in .env.
Hermes Memory Stack is an optional add-on for an existing tool called Hermes Agent. The repository's installer layers in extra memory, knowledge, and skill components onto a fresh Hermes setup. The project insists that none of these components ship with API keys, OAuth tokens, Google cookies, or personal data; user secrets stay in a local .env file under the user's home directory. The installer is run with ./install.sh after cloning, or as a one-shot curl | bash line. It asks the user which layers to enable. The options listed in the README include reusable skills for memory-stack workflows, the Axiom Perception cache for repeated UI tasks, an llmwiki knowledge layer in several modes, Hindsight routing helpers for durable memory, an optional NotebookLM workspace, and an optional semble_rs code and log helper. Three install profiles are offered. Minimal adds only the skills and Axiom. API-free adds llmwiki in markdown mode and optional semble_rs, with no external network calls. Full is a guided wizard that lets the user pick everything piece by piece, including NotebookLM and the auxiliary model routing template. NotebookLM is off by default. If turned on, the installer adds a CLI and prints manual login steps. After login, a sync job can write a compact list of notebook names, IDs, source titles, and short summaries to a routing file under the user's wiki folder. A second opt-in path installs a source-pack builder from a separate GitHub repository, which can bundle public documentation as NotebookLM sources on a refresh schedule. For memory routing, the installer can keep the existing Hermes provider, ask whether Hindsight should reuse a local GPT OAuth bearer token, point Hindsight at any OpenAI-compatible endpoint via environment variables, or skip the routing change entirely. The Hermes config file is copied as a backup before any patching.
Generated 2026-05-22 · Model: sonnet-4-6 · Verify against the repo before relying on details.