TOBER FEST

Description and use

SecondMind is a local‑first personal memory assistant that stores data on the user’s device and uses open‑weight models Qwen 2.5 1.5B and nomic‑embed‑text, avoiding cloud data transmission.

Problem

Users need a personal memory assistant that keeps data locally and does not rely on cloud AI services.

Outcome

SecondMind provides a local, privacy‑preserving personal memory assistant that can remember, recall, and forget information using open‑weight models, eliminating the need for cloud AI services.

Workflow

• 1. User inputs a query.
• 2. Query is embedded with nomic‑embed‑text.
• 3. Embedding is compared to stored memory embeddings using cosine similarity.
• 4. Relevant memories are ranked and passed to Qwen 2.5 1.5B for response generation.
• 5. The agent may invoke Remember, Recall, or Forget tools as needed.

Evidence

SecondMind stores memories locally and uses a local open‑weight model to understand conversations and retrieve relevant information when needed.

Source

The project runs locally with Django and Ollama.

Source

The entire system can run on a laptop without sending personal memories to a third‑party AI provider.

Source