Personal AI Agent

What it is for

A local, memory-aware assistant built with Python and Ollama for personal conversation, decision support, and continuity across sessions.

A personal assistant feels generic when every session starts from zero. This project experiments with lightweight local memory so the assistant can recognize a user and recover relevant themes without sending the conversation to a hosted application.

Run a local conversation

Use an Ollama-backed assistant from a terminal on your own machine.

Separate personal histories

Choose a user identity before chatting so stored exchanges stay scoped.

Experiment with memory

Study keyword summaries and retrieval of earlier context across sessions.

Get started

Follow the repository setup, start Ollama with the documented model, then launch the terminal assistant. Select or create a user identity before beginning a conversation.

• Python and the dependencies listed in the repository.
• A running Ollama installation with a compatible model.
• Local storage for per-user history and summaries.

Features and workflow

Local inference

Run supported Llama models through Ollama on the user’s own machine.

Multiple users

Give each person a unique identity and keep histories separated.

Conversation history

Store prior exchanges so later sessions can build on earlier work.

Keyword summaries

Create monthly summaries that compress recurring subjects and interests.

Relevant recall

Retrieve matching earlier context instead of injecting the full history every time.

Voice roadmap

Text-to-speech and emotion-aware voice behavior are active areas of development.

1. Install Python dependencies and a compatible Ollama model.
2. Start the terminal assistant and select or create a user identity.
3. Chat locally while the application stores that user’s history.
4. Monthly keyword summaries condense recurring topics.
5. Relevant summaries and messages are recalled for future conversations.

Architecture

The project is a Python terminal application connected to a locally running Ollama model. Its memory layer stores per-user conversation records and derived summaries, then retrieves relevant context before a response.

Privacy and limitations

Local inference reduces third-party exposure, but privacy still depends on the machine, Ollama configuration, stored history, and any future external integrations. Retrieved memory can be incomplete or misleading and should not replace user judgment.

Status

Open-source work in progress with no public hosted deployment. Run it locally from the repository; voice and emotion-aware interaction remain ongoing work.

Availability

Run locally

Interface

Terminal

Inference

Ollama

Status

Work in progress

Questions