Building a Personal-Memory RAG with Ollama

AetherMind gathers personal data such as notes, Git commits, calendar entries, and location history and turns it into a searchable AI memory.

Example questions include:

• "What was I doing in March?"
• "When was I most productive?"
• "What are my work habit patterns?"

Ollama is used as the core inference engine for the RAG synthesis step. It finds relevant events from Qdrant to build context, then generates answers based on that content.

qwen2.5:7b was chosen as the model.

• Its 32k context allows many events to be included at once
• It's fast enough to use interactively even on consumer GPUs
• It follows the JSON schema for daily retrospectives relatively reliably

The configuration is also simple.

• model: qwen2.5:7b
• temperature: 0.3
• timeout_seconds: 120

Any Ollama-compatible model can be used as-is just by changing the name, and it's introduced as installable in about 5 minutes based on setup.py.