BestBlogs Documentation

The BestBlogs Article Analysis Workflow is a Dify-based process designed to ingest a website article ID and produce a comprehensive analysis. The workflow uses HTTP and code nodes to fetch article metadata (title, source, link, language) and full text via the website's API.

Key features include:

• Adaptive Processing: Articles longer than 6,000 characters are automatically segmented for processing, while shorter articles are analyzed in full.
• Rich Content Extraction: Generates a one-sentence summary, abstract, keywords, main viewpoints, and golden quotes.
• Categorization: Uses LLMs to assign domains and tags (e.g., themes, technologies, products, companies, trends) to improve search and recommendation.
• Multi-dimensional Scoring: Evaluates content depth, writing quality, utility, and relevance.
• Self-Correction Loop: Includes a 'Check & Reflect' node where an LLM acts as a technical reviewer to ensure comprehensiveness, accuracy, and consistency, followed by an 'Optimization' node to refine the final output.

Developers can integrate this workflow into website applications by calling the Dify Workflow API with an article ID.

BestBlogs is composed of several specialized components that define its system boundaries. Developers interacting with the platform should identify which component matches their use case:

• bestblogs-app: Public website and user Web interface.
• bestblogs-admin: Administrative backend.
• bestblogs-api: User-facing API, authentication, Pro features, reading, and personalization capabilities.
• bestblogs-admin-api: Administrative API, content processing, and background task hosting.
• bestblogs-common: Shared domain models, configurations, Adapters, and infrastructure.
• bestblogs-worker: Independent background task modules.
• bestblogs-mobile: iOS / Android client modules (currently on hold).
• dujian-app: 'Dujian' (读见), the localized version for domestic deployment and Chinese presentation.
• bestblogs-cli: CLI and Agent-friendly entry point.
• bestblogs-skills: Agent Skills wrapper layer.

BestBlogs provides a unified quality judgment system for high-quality information input across multiple platforms, media types, and languages.

Key Capabilities:

• Multi-source Access: Aggregates content from RSS, Newsletters, Twitter, YouTube, and Podcasts.
• Multi-media Support: Handles articles, podcasts, videos, and tweets.
• Unified Content Processing: Performs extraction, cleaning, summarization, tagging, scoring, and translation.
• Cross-language Support: Provides immersive translation, summaries, terminology explanations, and AI companion reading to allow users to access primary source content in multiple languages.
• Public Quality Pool: A continuously calibrated content pool built through AI initial scoring and human calibration, accessible via Daily Newsletters, Weekly Digests, Topic Interpretations, Content Squares, and external interfaces like OpenAPI, RSS, or CLI.

BestBlogs is an AI-driven personal reading assistant designed to help users establish a stable, credible, and personalized high-quality information input. It focuses on three core capabilities:

1. Determining what is worth reading: Identifying signals and filtering out noise from massive content streams.
2. Assisting in comprehension: Using summaries, translations, explanations, Q&A, chapter jumping, and AI companion reading to lower the cost of understanding long or cross-language content.
3. Learning user interests: Continuously adapting to user preferences through follows, interest tags, reading, bookmarks, highlights, and feedback.

Core Philosophy: BestBlogs does not read or judge for the user. AI handles filtering, ranking, explanation, and association, but the user always decides what to read, what to believe, and how to form judgments. The goal is "Better content for you," not "More content."

Based on recent insights, the BestBlogs product strategy is evolving from a simple 'recommendation tool' to an 'AI Content Harness'. The goal is to provide a high-quality consumption experience that protects users' 'free attention' by using minimal 'bounded attention' to capture critical signals.

Key strategic pillars include:

• User Context & Data Flywheel: Leveraging user behavior (six-dimensional interest tags, reading completion rates, and bookmarks) to create a feedback loop for content scoring models.
• L4 Experience Standard: Aiming for a level of service where the BestBlogs morning report effectively replaces hours of manual RSS/social media scanning.
• Skills-based AI Integration: Converting operational expertise (scoring standards, editorial norms, tagging systems) into structured Skills that can be injected into Agent workflows.
• Content Expansion: Moving beyond LLM/Coding topics to cover 'Physical World AI' (Robotics, Embodied AI, Autonomous Driving).

BestBlogs is designed as an 'Agent Native' platform, meaning its capabilities are structured to be easily called, composed, and orchestrated by both humans and AI agents (like Claude Code, Cursor, or OpenClaw). Instead of just providing a web interface, BestBlogs provides three layers of access to integrate reading workflows into larger automation pipelines:

1. OpenAPI: The foundational layer providing stable, structured contracts for authentication, interest profiling, content discovery, deep reading, collection, and note-taking.
2. CLI (bestblogs-cli): A lightweight command-line interface designed for shell scripts and automation. It organizes capabilities into a logical reading workflow: intake → discover → read → capture.
3. Skills (bestblogs-skills): The highest layer, providing high-level primitives specifically for AI agent orchestration. It currently includes 5 skills and 25 stable primitives covering profiling, discovery, deep reading, accumulation, and explanation.