Tencent Internally Tests Cheso, an AI Agent for PPTs, Testing Surprised Me
Tencent is internally testing Cheso, an AI Agent specialized in generating presentation slides (PPT), with additional capabilities in image generation, poster design, research analysis, and web page creation Cheso supports multi-source web search with cross-validation for content generation, producing structured markdown outlines before creating PPTs with user confirmation at key decision points The platform integrates multiple LLM providers (Tencent Hunyuan, Google Gemini, DeepSeek, GLM, Alibab
Analysis
TL;DR
- Tencent is internally testing Cheso, an AI Agent specialized in generating presentation slides (PPT), with additional capabilities in image generation, poster design, research analysis, and web page creation
- Cheso supports multi-source web search with cross-validation for content generation, producing structured markdown outlines before creating PPTs with user confirmation at key decision points
- The platform integrates multiple LLM providers (Tencent Hunyuan, Google Gemini, DeepSeek, GLM, Alibaba Qwen) with dynamic model routing, abstracting model selection away from users
- Cheso features a mature four-column UI layout, project-based organization, version history, template library (with custom template upload), and supports multiple export formats (editable PPT, PDF) with online presentation capabilities
- The product emphasizes a human-AI collaborative workflow where users focus on goals, decisions, and quality review while AI handles content structuring, visual design, and iterative editing
Why It Matters
Cheso represents Tencent's strategic push into the AI-powered productivity tool space, specifically targeting the high-demand presentation creation workflow that dominates enterprise and professional environments. Its approach of abstracting model complexity behind intelligent routing and emphasizing human-in-the-loop confirmation at key nodes reflects a mature product philosophy that could set a new standard for AI agent design in vertical productivity applications.
Technical Details
- Multi-model orchestration: Cheso dynamically routes tasks across Tencent Hunyuan, Google Gemini, DeepSeek, GLM, and Alibaba Qwen based on task requirements, availability, and capability factors, with no model selection exposed to end users
- Research and content pipeline: For content generation from scratch, Cheso performs web searches across authoritative sources (arXiv, official docs, GitHub, HuggingFace, news sites), applies multi-source cross-validation, and produces structured markdown before generating PPT outlines for user approval
- Multi-format input support: Accepts diverse document formats as context including txt, md, doc, pdf, ppt, and images, integrating them into the AI generation pipeline
- Iterative editing workflow: Supports both manual text editing (free) and AI-assisted style/content modifications (consumes credits), with a copy-and-compare feature enabling side-by-side visual evaluation of design variations
- Project-based architecture: Each PPT project encapsulates all related artifacts (source materials, outlines, final presentations, AI conversation history) with version control, online sharing with granular access controls (public, password-protected, user-specific), and export to editable PPT or PDF formats
Industry Insight
- The "invisible model" design pattern—where the AI agent handles model selection internally—demonstrates a user-centric approach that could become an industry best practice for vertical AI agents, reducing cognitive load and preventing users from making suboptimal model choices
- Cheso's emphasis on confirmation checkpoints (outline approval before generation, copy-compare for design iterations) reflects a growing recognition that AI productivity tools must balance automation with human oversight to ensure output quality and user trust
- Tencent's integration of multiple competing LLM providers within a single product signals the maturation of model-agnostic agent platforms, where competitive differentiation shifts from model ownership to workflow design, user experience, and domain-specific capabilities
Disclaimer: The above content is generated by AI and is for reference only.