Build Your First AI Knowledge Base with ima: A Step-by-Step Tutorial

The article introduces ima (Tencent's AI Knowledge Manager) as a workbench that integrates storage, understanding, generation, and sharing into a closed loop for knowledge assets. It uses a scenario where a content creator named Alin opens ima but feels lost among its four entry points: New Chat, Knowledge Base, Discovery Square, and Q&A History.

Core Concepts: Three Capabilities Around the Knowledge Base

ima is defined as an AI workbench centered on the knowledge base. Its three core capabilities form concentric circles:

Knowledge Base (Warehouse) : Stores PDFs, Word docs, WeChat articles, images, web links. It automatically chunks content semantically, converts to vectors, and builds a meaning-level index.

AI Q&A (Machine) : Answers questions by retrieving semantically relevant content from the knowledge base and synthesizing answers, not by keyword search.

Discovery Square (Exhibition Hall) : Allows publishing your knowledge base publicly or subscribing to others'. Visitors gain query access without copying files.

The analogy: a private library (knowledge base), a research assistant (AI Q&A), and a book club (Discovery Square). The loop works only when all three connect: build the library, train the assistant, then optionally open the club.

Practical Steps: Create Your First Knowledge Base in 5 Minutes

3.1 Download and Login

Official site: https://ima.qq.com. As of July 2026, supports Windows, Mac, mini-program, iOS/Android apps, and web with real-time sync. Desktop recommended for bulk uploads and permission management.

3.2 Create a Shared Knowledge Base

Click "Knowledge Base" → "+" → choose "Shared Knowledge Base" (not Personal). Shared bases allow inviting members and later publishing to Discovery Square; personal bases cannot be converted easily. Fill in:

Name: e.g., "My Test Knowledge Base"

Cover: style-consistent image

Description: e.g., "Collects my WeChat articles, paid column notes, reading notes for writing and Q&A"

Recommended Questions: 3 specific, content-driven prompts like "Which articles did I write about Redis?" or "Summarize my views on RAG"

3.3 Upload Files

Supports PDF, Word, PPT, Excel, images, Markdown, TXT, web links, and direct WeChat article/file import. Two pitfalls:

File size limit : unofficial reports vary (100–200 MB). Safest: split files >100 MB into ~30 MB chunks. Large files risk timeout during parsing, chunking, and vectorization.

PDF page count : stable parsing under 1,000 pages. Over that, split by chapter for better parsing and precise source citation.

Wait for parsing status to show "Completed" before querying.

3.4 Set Sharing Permissions

Options: Private (only you), Invite-only (via link/QR), Public/Discovery Square (anyone can subscribe). For testing, use Invite-only. Publishing to Square has benefits: content doesn't count against personal cloud storage (since March 2025 rule) and supports up to 1 million members per base. Trade-off: platform content review and recommendation criteria apply (title clarity, content density, update frequency).

3.5 Test with AI Q&A

After parsing, ask questions in the knowledge base's input box. Example: "@My Test Knowledge Base, list all my views on AI knowledge bases." The @ syntax targets a specific base when managing multiple. Answers are strictly grounded in uploaded content, not model training data.

Advanced Case & Common Pitfalls

4.1 Alin's "WeChat Article Material Library" Prototype

Alin uploaded 3 PDFs: Redis persistence selection, 7 RAG pitfalls, knowledge management for creators. She asked 5 questions; first three answered accurately, fourth needed cross-article synthesis, fifth triggered a self-comparison impossible in traditional search. She discovered inconsistent heading styles ("1,2,3" vs "一、二、三") caused mis-citations. Fixes: re-export as Markdown with unified headings, add tags ( #Redis, #RAG, #知识管理), split long articles into Background/Problem/Solution/Pitfalls/Summary sections. Result: accuracy improved. Lesson: AI knowledge base is an amplifier, not a magician. Clearer structure yields better amplification.

4.2 Pitfall: Treating ima as a Cloud Drive

Uploading hundreds of mixed files (tech, life, meetings, shopping) leads to chaotic answers. Solution: one theme per base (writing material, interview prep, reading notes). Use multiple bases and @ switching.

4.3 Pitfall: Chasing Perfect One-Time Upload

Spending weeks organizing 5 years of articles before any query wastes time. Value comes from being queried, not just organized. Start with 3–5 most-used recent articles, ask 10 questions, iterate.

4.4 Pitfall: Ignoring Recommended Questions

They force you to define core use cases: how will others (or future you) query this base?

4.5 Pitfall: Publishing Without Permission Review

Check for personal info, company secrets, copyright issues before going public. Test with Invite-only first, then review content again before publishing.

Key Takeaways

ima's three capabilities are interlocked: storage, Q&A, sharing — missing one breaks the knowledge asset loop.

Don't wait for perfection: upload 3 frequent articles, ask 10 questions, improve iteratively.

Public bases don't consume personal storage but require content audit and permission management before publishing.

This base will be reused in later chapters (writing with knowledge base, resume/interview library). The author has already imported their Knowledge Planet column into ima.

Exercises

Hands-on : Create "My Test Knowledge Base", upload 3 articles, ask 3 diverse questions. Share one insight and one pitfall.

Judgment : For a 300 MB PDF, how to split — by chapter or page count? Why?

Transfer : Besides WeChat articles, what other work/study content fits ima? Give 2 examples with suitable base themes.