morpheus_ai

A tool for combining Large Language Models with an ElasticSearch document store for document retrieval and summarization Named for the AI from the original Deus Ex

Getting Started

• Install an ElasticSearch Local Instance
• Download a Summarization Model
• Set up an LLM API Service (OpenAI and Grok currently are known to work)
• Set up a config.json file, filling out the fields as described in config.json_example

Building and Using

• trigger mnemosyne.py to build a database from your ebook library (pdf and epub)
• once that's done, use morpheus.py as a command line interface tool and ask away!

Usage of Individual Components

mnemosyne.py

• --no-cache : do not use the persistent SQLite vector cache (recompute embeddings)
• --clear-cache : clear the vector cache table before running
• --vacuum-cache : run VACUUM on the morpheus.db SQLite file to reclaim space (useful after --clear-cache)
• --force-reindex : ignore the indexed_files bookkeeping and reprocess all source files
• --atomic-reindex : build the new index into a temporary physical index and atomically swap a logical alias to it when complete (prevents partial-index states)
• --drop-old-index : when used with --atomic-reindex, delete the old physical index(es) after the alias swap

Example (PowerShell)

# Build into a temporary index and atomically swap when finished
python .\mnemosyne.py --atomic-reindex

# Rebuild from scratch, clear cache and reclaim DB space
python .\mnemosyne.py --clear-cache --vacuum-cache --force-reindex

# Re-index but avoid using the persistent vector cache
python .\mnemosyne.py --no-cache --force-reindex

kerukeion.py

Small conversation summarizer and title generator. New CLI flags allow filtering which conversations are displayed.

• --query / -q : space-separated words to search for in the prompt+response text. Words are searched separately (AND semantics) and matching is case-insensitive.
• --min-size : only show conversations with more than this many messages (integer).
• --min-start : minimum conversation start date (inclusive). Accepts a unix epoch (seconds) or an ISO date string (e.g. 2025-08-11).
• --max-start : maximum conversation start date (inclusive). Same formats as --min-start.

Notes:

• Date parsing is flexible: the script will accept integer epochs or ISO-like date strings and will normalize them internally. Filtering uses inclusive comparisons (>= for --min-start, <= for --max-start).
• The displayed Start column is formatted as YYYY-MM-DD HH:MM:SS for readability.
• If you prefer OR semantics for the query or different date formats, the script can be adjusted — open an issue or request the change.

Examples (PowerShell):

# Show conversations containing both "invoice" and "error", longer than 10 messages, starting after 2023-01-01
python .\kerukeion.py --query "invoice error" --min-size 10 --min-start 2023-01-01

# Show conversations in a date window
python .\kerukeion.py --min-start 2024-01-01 --max-start 2024-12-31

Development Roadmap

• Conversation continuations
• Branching Conversations