Agent

Knowledge Base Management

AE Knowledge Base Management

Create, upload to, compile, and query AE knowledge bases, building structured knowledge assets from documents and URL sources. Supports deterministic retrieval (index to grep to read) and LLM Q&A (ask), so team knowledge is no longer scattered.

IndustriesAll industries

Agent · Skills Live

skill · loadedindustry+100

Run this Skill on my data

Tracking schema

events · properties

event · payment3 props

user_idstring

pay_amountnumber

channelstring

Analysis readyevidence attached

The AE Knowledge Base Management Skill is a knowledge-asset management capability in ThinkingAI's industry Skill library. It helps teams create AE knowledge bases, upload documents and URL sources, generate compile schemas, compile knowledge bases, and query knowledge content, supporting both a deterministic retrieval path (index to grep to read) and an LLM Q&A path (ask), upgrading from documents scattered everywhere and impossible to find to one-click retrieval from a structured knowledge base. This is ThinkingAI's decade of industry experience, packaged into ready-to-use Skills.

Live walkthrough

AgentSave as AgentExport session

ae-knowledge-base-management

Our team's engineering docs are scattered across 5 Feishu docs and 3 Google Drive pages, and a new hire asking about sandbox config spends 20 minutes just finding it. Build an engineering-handbook knowledge base, pull these docs in and compile them, then verify you can actually retrieve something.

ae-cli kb +create to create the company-scope knowledge base engineering-handbook

+add to bulk-upload the Feishu-exported docs and the local config directory

All 12 source files accepted

+url to add 3 Google Drive page sources

Auto-converted to markdown

+schema to generate the compile schema: organized into 'Environment Setup / Release Process / Troubleshooting Manual'

+compile the knowledge base

15 sources compiled into 38 knowledge pages

Retrieval check: +index to navigate → +grep to locate sandbox → +read to read closely

Hit the latest config page, about 10 seconds total, zero tokens

All three steps are done: build, compile, and a live retrieval test. 15 knowledge sources compiled into 38 searchable pages, and looking up 'sandbox config' went from 20 minutes on average to 10 seconds: over the zero-token deterministic retrieval path.

1. Knowledge base status

Knowledge base	Scope	Source files	Compile status	Last updated
engineering-handbook	company	15 (12 files + 3 URLs)	Compiled · 38 pages	Today
product-faq	company	23	Compiled · 51 pages	6 days ago
sdk-integration-notes	personal	7	Not compiled	32 days ago

← Scroll to see more

2. This run's upload and compile results

All 12 Feishu-exported source files and 3 Google Drive URL sources were accepted, and the URL pages were auto-converted to markdown

The schema is organized into three top-level sections ('Environment Setup / Release Process / Troubleshooting Manual'), and compilation produced 38 knowledge pages with no failed sources

The retrieval check used the deterministic path: +index to navigate → +grep to locate the keyword (3 pages hit) → +read to read the latest sandbox config steps closely, at zero token cost

For synthesis-style answers that need summarizing or comparing (like 'what should a new hire read in week one'), use +ask for LLM Q&A, consuming tokens as needed

Heads-up

One thing worth flagging: sdk-integration-notes hasn't been updated in 32 days and is not compiled: query results from an uncompiled knowledge base are unreliable. Either run +schema + +compile on it, or merge the useful content into engineering-handbook and delete it, so new hires don't retrieve an outdated doc and follow stale config.

The engineering-handbook knowledge base is compiled and live; team members can search it immediately.

On your data

That was a simulated run

Leave your work email and we will run a live walkthrough on your real business data.

The problem

Team knowledge management is a pain point for every organization, but over 60% of teams have no searchable knowledge assets. Internal docs are scattered across Lark Docs, Google Drive, local folders, Wikis, code repos, and more, and finding the answer to a specific question (such as how to configure the sandbox environment) takes searching 3-4 tools and over 15 minutes on average, and often turns up nothing. An even more common problem is stale knowledge: a doc is updated but the old version is still being cited, and new hires follow the old doc and misconfigure things. LLM Q&A is powerful but expensive, consuming platform tokens on every question, with answer quality depending on the knowledge base's compile quality, and query results from an uncompiled knowledge base are unreliable. Deterministic retrieval (keyword grep + precise page read) is a zero-token alternative, but most teams do not know this path exists.

What it does

Full lifecycle management: create knowledge base to upload source files/URLs to generate schema to compile to query/Q&A, one chain covering everything from zero to ready

Deterministic retrieval path: index navigation to grep location to read precise reading, zero token consumption, ideal for automated retrieval by Agents

LLM Q&A path: the ask command calls a large model for a synthesized answer, suited to complex questions requiring summary, reasoning, or comparison

Multi-source upload: local files (markdown/Office/PDF/images) + URL sources (web pages auto-converted to markdown) + local directories, aggregated in one place

Strictly no guessing: knowledge base names, page paths, and source-file display names all come from query results, never from guesswork

When to use it

Creating a project knowledge base and uploading internal docs, Wikis, code repos, and other knowledge sources

Deterministically retrieving a configuration step or API description from the knowledge base

Getting a synthesized answer from the knowledge base via LLM Q&A

Uploading a URL page as a knowledge base source (web page auto-converted to markdown)

Generating a knowledge base compile schema and compiling it to make it queryable

Checking a knowledge base's compile status or deleting outdated source files

In the field

Case

A game development team · engineering knowledge base build-out

Internal docs were scattered across 5 Lark Docs and 3 Google Drive folders, and new hires took 20 minutes on average to find the "sandbox environment configuration" and often could not find the latest version. With the AE Knowledge Base Management Skill, the system created the company knowledge base engineering-handbook, batch-uploaded Lark Doc URLs and local config files with +add, added Google Drive pages with +url, generated a compile schema with +schema, and compiled the knowledge base with +compile. New hires used the deterministic +index to +grep to +read path to find answers within 10 seconds with zero token consumption.

FAQ

What is the difference between deterministic retrieval and LLM Q&A?

Deterministic retrieval (index to grep to read) is keyword location + precise reading, with zero token consumption, suited to finding specific steps or configs; LLM Q&A (ask) calls a large model for a synthesized answer, consumes tokens, and is suited to questions requiring summary, reasoning, or comparison.

Which file formats are supported for upload?

Markdown/text, Office documents (Word/Excel/PPT), PDF, images, local directories (non-recursive), and HTTP(S) pages (auto-converted to markdown).

What is knowledge base compilation?

Compilation turns the uploaded source files into a queryable knowledge index. Query results from an uncompiled knowledge base are unreliable; you must first generate a schema with +schema, then compile with +compile, before it can be queried normally.