Advanced setup

Let AI save knowledge automatically

Three safeguards help preserve knowledge from every conversation.

Start with zero configuration, then add explicit project instructions and manual control.

Three Layers

Three layers of protection

Use the combination that matches how much control you need.

Before you begin: Connect KnowMine to your AI through MCP.Not connected yet? Open the connection guide →

Layer 1: Automatic AI detection

Active after connecting · no configuration

How it works

After an AI client such as Claude Code or Claude Desktop connects to KnowMine MCP, it can recognize valuable content and call save_memory to save it. KnowMine uses five knowledge types:

🎯Decisiondecision

"Choose pgvector instead of Pinecone because it costs less and PostgreSQL is already in use."

💡Lessonlesson

"Do not configure a proxy in the VS Code extension because it breaks the OAuth flow."

Insightinsight

"AI memory benefits from Trace, Memory, and Soul layers with different lifecycles."

🎨Preferencepreference

"The user expects asynchronous work to include retries and observable status."

📚Domain knowledgedomain_knowledge

"MCP can use Streamable HTTP transport and support stateless deployment."

Note: This layer is probabilistic. AI may not save something every time because the decision depends on the model and conversation context. Layers 2 and 3 provide explicit safeguards.

Layer 2: CLAUDE.md instructions

Check at important moments · Claude Code only

What is CLAUDE.md?

CLAUDE.md is Claude Code’s project-level instruction file. Put it in the project root and Claude Code reads it at the start of every conversation, giving AI a reusable working guide.

Setup (about two minutes)

01Open or create CLAUDE.md

Find or create CLAUDE.md in the root of your project.

02Append these instructions

The block includes automatic checks and a /distill shortcut. Copy both together:

## KnowMine automatic knowledge saving

At these moments, proactively check whether the conversation contains knowledge worth saving:
1. After completing a feature
2. After resolving a bug
3. After making a technical decision
4. After discussing a complex topic

Save one of these five types:
- **decision**: What was decided, and why was A chosen over B?
- **lesson**: What went wrong, what was the root cause, and how was it fixed?
- **insight**: What valuable insight or idea emerged?
- **preference**: What working preference or habit was discovered?
- **domain_knowledge**: What specialized knowledge was discussed?

When something is worth keeping, call the save_memory tool.
Do not save temporary debugging details, raw code snippets, or intermediate output.

## Shortcut
When the user says /distill or /沉淀, review the current conversation, extract the 1–3 most valuable pieces of knowledge, and call save_memory to store them in KnowMine.

Or copy each part separately —

03Save the file

The instructions take effect in the next Claude Code conversation. No restart is required.

What changes after setup

After completing a feature AI checks for decisions or lessons worth saving

After fixing a bug AI captures the cause and fix so the team can avoid repeating it

After a technical choice AI records the reasoning for future reference

Say /distill or /沉淀 AI reviews the conversation and saves 1–3 high-value items

Layer 3: Manual trigger

Save on demand · works on every platform

How to trigger a save

In any AI conversation connected to KnowMine, use one of these prompts:

/distill

English shortcut

/沉淀

Chinese shortcut

Save this to my knowledge base

Natural-language trigger

Save the decision we just made to my knowledge base

Save a specific part of the conversation

When to use it

A useful idea appears and you do not want to lose it → /distill

You finished a complex discussion and want to preserve the conclusion → /distill

The conversation is ending and you want a final knowledge check → /distill

Other AI platforms

Not using Claude Code? Other platforms can still use Layer 1 (automatic) and Layer 3 (manual):

Claude Desktop

Layer 1 + 3

ChatGPT

Layer 1 + 3

Cursor / VS Code

Layer 1 + 3

Claude Code

Layer 1 + 2 + 3

Layer 2 (CLAUDE.md instructions) is specific to Claude Code. On other platforms, connect through MCP and use a prompt to trigger saving.

Start building reusable knowledge

After setup, each thoughtful AI conversation can become knowledge you can reuse.

AI Knowledge Auto-Save — KnowMine Three-Layer Guide