Mnemosyne + OpenWebUI

Add persistent memory to OpenWebUI via a native @tool.

Setup

1. Install Mnemosyne:

pip install mnemosyne-memory

2. Create a bridge file in OpenWebUI's data/tools/ directory:

echo "from openwebui_tool import MnemosyneTool as Mnemosyne" \
  > /path/to/open-webui/data/tools/mnemosyne.py

3. In OpenWebUI, go to Workspace > Tools. The mnemosyne tool appears automatically (hot-reloaded).

4. In a chat, select the Mnemosyne tool from the model's tool dropdown.

Configuration

The tool provides a settings panel in OpenWebUI with these Valves:

Setting	Default	Description
db_path	~/.hermes/mnemosyne/data/	Path to memory database directory
bank	default	Memory bank name for isolation
top_k	5	Max results returned by recall
vec_weight	null (env default)	Vector search weighting
fts_weight	null (env default)	Full-text search weighting
importance_weight	null (env default)	Importance boost weighting
show_citations	true	Show source citations in output

These settings are persisted per-user in OpenWebUI's database.

Usage

The tool provides these operations:

Operation	Description
mnemosyne_remember	Store a memory
mnemosyne_recall	Search memories by semantic similarity
mnemosyne_forget	Delete a specific memory
mnemosyne_stats	View memory statistics
mnemosyne_sleep	Run consolidation

The LLM decides when to use each tool based on the conversation context.

Tips

• Memory isolation: Use different banks per chat/workspace for privacy
• Data location: Override MNEMOSYNE_DATA_DIR env var to change storage path
• Testing: Import the tool in Python directly to verify it works:

  import asyncio
  from openwebui_tool import MnemosyneTool
  tool = MnemosyneTool()
  result = asyncio.run(tool.mnemosyne_remember("test memory"))
  print(result)