eze-is/ai-partner-chat
Overview
This skill provides personalized, context-aware conversations by combining a user persona, an AI persona, and vectorized personal notes. It lets the AI remember and reference prior thoughts, preferences, and knowledge so responses stay coherent and tailored over time. The system indexes markdown notes into a local vector database and uses the top relevant chunks to inform each reply.
How this skill works
On initialization the skill reads your persona files and scans the notes directory to generate chunking logic per file, then produces validated chunks and indexes embeddings into a local ChromaDB. For each user query it retrieves the top-k semantically similar note chunks, merges them with the user and AI persona definitions plus recent chat history, and synthesizes a response that naturally references relevant notes while following the configured AI persona style. The chunking is generated dynamically per note to adapt to varying formats and ensure chunks conform to a defined schema.
When to use it
- When you want the AI to remember and cite your previous notes or ideas
- When you need context-aware replies that respect your communication preferences
- When working on long-running projects where continuity and recall matter
- When you maintain mixed-format notes and need adaptive chunking
- When you want a locally stored, privacy-friendly vector database for personal context
Best practices
- Keep user and AI persona files specific and example-driven to shape tone and behavior
- Store substantive markdown notes (not just titles) to improve retrieval quality
- Add new notes to the notes/ folder and re-index when many changes accumulate
- Use top_k tuning to adjust how much note context the model considers
- Review generated chunking code if chunk sizes or labels need refinement
Example use cases
- Personal study assistant that recalls prior learning goals and past summaries
- Long-term project collaborator that references previous decisions and meeting notes
- Creative partner that builds on earlier brainstorming captured in notes
- Coaching or mentoring assistant that remembers preferences and prior progress
- Research assistant that surfaces previous literature notes related to a new query
FAQ
Notes are chunked dynamically per file, embeddings are generated, and chunks are stored in a local ChromaDB under ./vector_db for persistent, private storage.
What do I need to set up before first use?
Create config/ with user-persona.md and ai-persona.md, populate notes/ with markdown files, create the project directories, set up a Python virtual environment, and run the provided chunk-and-index script to initialize the vector DB.
