This project is a local AI assistant built with Python and Ollama.
Latest Development Log
Project Log — AI Assistant Development
Date
March 15, 2026
Progress Today
Today I worked on improving the memory system of my local AI assistant built with Python, Ollama, and PostgreSQL.
1. Git Workflow Update
- Created a new Git branch for database-related work:
git checkout -b feature/database_store
- This branch is dedicated to developing and testing database memory features without affecting the main branch.
2. PostgreSQL Memory Integration
Connected the assistant to a PostgreSQL database.
Created a
memorytable to store important user information.Implemented database helper functions:
store_memory()→ stores a memory in the databaseget_memories()→ retrieves stored memoriesclear_whole_database()→ clears the memory table
This allows the assistant to persist information between sessions.
3. Memory Trigger System
Added logic to detect when the user wants the assistant to remember something.
The assistant now looks for trigger words such as:
rememberstoresave
Example:
User input:
remember my name is Rohit Rajvaidya
The assistant detects the trigger and prepares the information for storage.
4. Memory Paraphrasing with LLM
Implemented a small LLM prompt that converts the user sentence into a clean factual memory before storing it.
Example transformation:
Input:
remember my name is Rohit Rajvaidya
Stored memory:
User name is Rohit Rajvaidya
This ensures the database stores structured and consistent information.
5. Output Cleaning
Added a cleanup step to remove unnecessary text returned by the model, such as:
Output:
Explanation:
This ensures only the final fact is stored in the database.
6. Assistant Improvements
The assistant now includes:
- Local LLM interaction using Ollama
-
Model fallback system (
tinyllama → phi3 → llama3) - Terminal commands (clear chat, switch models)
- Loading animation during model generation
- PostgreSQL memory storage
- Automatic detection of memory instructions
- Memory paraphrasing before database storage
Next Steps
Planned improvements:
- Inject stored memories into the system prompt so the assistant can recall user information across sessions.
- Prevent duplicate memory entries in the database.
- Improve memory extraction prompts.
- Introduce structured memory types (name, preferences, location, etc.).
- Implement memory retrieval during conversation to make the assistant more context-aware.
Previous Logs
See the full history in the ProjectLogs folder.
Here's Github Link :
Local AI Chatbot Project
This project is a local AI assistant built with Python and Ollama.
Latest Development Log
Project Log — AI Assistant Development
Date
March 15, 2026
Progress Today
Today I worked on improving the memory system of my local AI assistant built with Python, Ollama, and PostgreSQL.
1. Git Workflow Update
- Created a new Git branch for database-related work:
git checkout -b feature/database_store
- This branch is dedicated to developing and testing database memory features without affecting the main branch.
2. PostgreSQL Memory Integration
-
Connected the assistant to a PostgreSQL database.
-
Created a
memorytable to store important user information. -
Implemented database helper functions:
-
store_memory()→ stores a memory in the database -
get_memories()→ retrieves stored memories -
clear_whole_database()→ clears the memory table
This allows the assistant to persist information between sessions.
3. Memory Trigger System
Added logic to detect when the user wants the assistant…
Top comments (0)