Saarthi: A Local AI Companion That Remembers What Matters

Saarthi is a local AI companion designed to remember what matters to individual users and use that context to guide their decisions. Created for a close friend dealing with multiple responsibilities like exams, projects, personal goals, and deadlines, Saarthi goes beyond being just a chatbot. It understands the user's context to determine what to prioritize.

Users can communicate with Saarthi naturally, such as mentioning an upcoming exam or a confusing topic. Saarthi then assesses the importance of the information and stores it in MongoDB as part of the user's long-term memory. Later, when the user asks what to focus on, Saarthi retrieves relevant memories from MongoDB and provides a personalized recommendation from Gemma, an AI model running locally on the user's machine.

Saarthi allows users to approve or reject the information it remembers. It also supports updating existing memories instead of creating duplicates when context changes. The AI identifies useful information from natural conversations, tracks goals, preferences, and upcoming events, and even visualizes the user's context through timelines and graphs.

Open innovation played a significant role in building Saarthi. By using local AI through Gemma and Ollama, Saarthi provides users with greater control over the AI inference process. Users can run the model locally, use the app without internet, modify the model and prompts, and experiment with the AI behavior without relying on a third-party API. This approach allowed Saarthi to focus on the interaction between memory, context, and local AI, rather than complicated infrastructure.