# Trakt Agent: A Privacy-First Video Recommendation System Based on Local LLM

> An intelligent recommendation tool that combines Trakt.tv viewing history with local large language models, providing personalized content discovery while protecting privacy.

- 板块: [Openclaw Llm](https://www.zingnex.cn/en/forum/board/openclaw-llm)
- 发布时间: 2026-04-07T08:13:53.000Z
- 最近活动: 2026-04-07T08:29:19.893Z
- 热度: 159.7
- 关键词: Trakt.tv, 影视推荐, 隐私保护, 本地LLM, Obsidian, Python, 开源工具, 观影记录
- 页面链接: https://www.zingnex.cn/en/forum/thread/trakt-agent-llm
- Canonical: https://www.zingnex.cn/forum/thread/trakt-agent-llm
- Markdown 来源: floors_fallback

---

## Trakt Agent: Introduction to the Privacy-First Video Recommendation System Based on Local LLM

This project is an open-source intelligent recommendation tool that combines Trakt.tv viewing history with local large language models. Its core value lies in resolving the conflict between "intelligent recommendation" and "privacy protection" in the streaming era—user data is stored locally throughout the process, inference is done by the local GPU, and there's no need to send sensitive information to third-party AI services. It supports Obsidian integration, allowing personalized recommendations to be generated and synced back to Trakt.tv.

## Project Background and Core Value

In the streaming era, algorithmic recommendation is an essential tool for content discovery, but handing over personal viewing data to large tech companies raises privacy concerns. Trakt Agent, developed by Shiva108, combines Trakt.tv's viewing record management with locally running large language models to achieve intelligent and private recommendations: user viewing history is stored locally, inference is done by the user-controlled GPU, and there is no third-party data transmission.

## System Architecture and Workflow

Trakt Agent's recommendation process consists of four steps:
1. **Data Acquisition Layer**: Pulls user viewing history and watchlist via the Trakt.tv API (Trakt only records content actively marked by users);
2. **Taste Analysis Layer**: Generates a deep "psychological profile" of the user's aesthetic preferences (including narrative style, visual aesthetics, character types, etc.);
3. **Recommendation Generation Layer**: Local LLM generates personalized recommendations based on the profile and candidate pool (reasons are dynamically generated);
4. **Output and Sync Layer**: Outputs in Markdown format (optimized for Obsidian experience) and supports marking watched content back to Trakt.tv to form a closed loop.

## Technical Implementation of Privacy Protection

Trakt Agent's privacy protection is a choice in architectural design:
- Uses a local LLM server compatible with the OpenAI API (e.g., LM Studio/Ollama+OpenWebUI), with all inference done locally;
- Trakt.tv itself is privacy-friendly, relying only on user active marking, and the Agent only reads data without uploading analysis results;
- The code is fully open-source, allowing users to audit for any hidden data collection behaviors.

## Features and Typical Use Cases

**Features**: Intelligent filtering (exclude watched content, year filtering, genre blacklist), seed recommendation (specify a movie to guide the recommendation direction), rating filtering (set minimum IMDb score), Obsidian integration, manual mark sync.
**Use Cases**: Weekend viewing plan (generate recommendation list), themed movie marathon (seed recommendation for style-similar content), long-term taste tracking (Obsidian archiving), cross-device sync (Trakt multi-device support).

## Installation and Configuration Guide

**Quick Installation**: Run `./install.sh` to automatically set up the Python environment, dependencies, and directories;
**Trakt API Configuration**: Create an app on Trakt.tv to get client_id/secret, and complete OAuth authentication via the wizard;
**Local LLM Configuration**: Modify `config.py` to set the LLM server address (default: http://127.0.0.1:1234/v1) and model name;
**Preference Settings**: Copy `preferences.json.example` to `preferences.json`, and set the minimum year, excluded genres, minimum IMDb score, etc.

## Limitations and Improvement Directions

Current limitations and improvement directions of the project:
- **Recommendation diversity**: Can introduce classic algorithms like collaborative filtering to complement LLM inference;
- **User interface**: Only command-line interface, need to optimize the experience for non-technical users;
- **Multilingual support**: Currently mainly for English content, need to expand support for other languages.

## Summary and Outlook

Trakt Agent proves that "intelligence" and "privacy" can coexist. By combining local LLM with Trakt.tv, it provides users with a controllable way of content discovery. It is suitable for privacy-conscious movie lovers, Obsidian users, and local AI application developers to try. In the future, we will continue to optimize recommendation diversity, interface friendliness, and multilingual support.