MemoBase
maximskorohod/memobase
Hermes plugin for a local knowledge base over files, sites, YouTube and Obsidian, with verified citations
MemoBase is a plugin for hermes-agent that turns PDFs, DOCX files, web pages, YouTube videos and Obsidian notes into a local searchable knowledge base. It answers only from those sources, with verbatim citations, or says the answer is not in the base.
What MemoBase does
MemoBase ingests PDF, DOCX, HTML, MD, TXT and CSV files, web pages, YouTube videos and whole channels, audio recordings and Obsidian notes or vaults. Text is normalized, split into chunks, scanned for secrets and embedded. Anything containing a secret is quarantined and not indexed. Everything is stored in one local SQLite file, memobase.db, using FTS5 and vec0.
Questions go through hybrid search that combines FTS5 keyword matching (BM25 with Russian stemming) and vector KNN, merged with RRF and reranked with Cohere rerank-v3.5. A sufficiency gate decides whether the retrieved fragments can answer. If not, MemoBase replies that the knowledge base has no answer. Otherwise a model without tools answers from the fragments only, and each quotation is checked against the source verbatim. Video and audio answers carry timestamps. MemoBase runs alongside Hermes's own memory rather than replacing it, and the README describes a fully local mode with a local ONNX embedder (fastembed) that needs no API keys.
Key features
- Ingests PDF, DOCX, HTML, Markdown, TXT, CSV, web pages, YouTube videos and channels, audio and Obsidian vaults
- Hybrid search with FTS5 keyword matching and vector KNN, merged with RRF
- Verbatim citations checked against the source, or an explicit refusal when the base has no answer
- Secret scanner that quarantines chunks containing secrets
- Whisper transcription with segment timestamps for audio and video
- Local ONNX embedder (fastembed), so search works without paid keys
When to use it
- Asking questions about a folder of PDFs and getting quoted answers with page or section references
- Searching the transcripts of a YouTube channel from a Hermes chat
- Querying an Obsidian vault through your agent with answers tied to the original notes
Who it is for: Hermes Agent users who want a NotebookLM-style, source-grounded knowledge base running inside their agent.
How it fits with Hermes Agent
Built for Hermes Agent as a plugin that targets hermes-agent 0.18 or newer and runs alongside its built-in memory.
Requirements: hermes-agent 0.18 or newer
FAQ
What is MemoBase?
MemoBase is a Hermes Agent plugin that builds a local knowledge base from your files, web pages, YouTube videos, audio and Obsidian notes. It answers questions strictly from those sources, with checked quotations.
Does MemoBase replace Hermes memory?
No, MemoBase is a separate module for documents. The README says both systems can stay enabled at once without conflict, and describes MemoBase as a librarian rather than a memory.
Is MemoBase free and open source?
Yes, the repository is released under the MIT license.
Similar memory for Hermes Agent
All memoryLocal memory provider for Hermes Agent backed by an embedded Ladybug graph database
electricsheephq LCM-XLossless context memory for Hermes Agent using a SQLite-backed summary DAG
cobibean Agent Memory WikiObsidian daily logs, session logs and /prepforreset handoffs for Hermes and other AI agents
jiangyuxue666 SOUL.md Governance FrameworkSOUL.md governance framework for Hermes Agent with structured memory and skill management
maximskorohod MemoHoodDialogue memory provider for Hermes Agent with auto-recall, fact extraction and supersede versioning
westsource Agent Context SyncSelf-hosted session sync that shares Hermes and other agent conversations across devices
Related guides: SOUL.md for Hermes Agent: what it is and how to write one · Run multiple Hermes agents with profiles