English | 简体中文

Bilibili Follow

Export your Bilibili follow list, classify it locally, optionally let ChatGPT / Claude / Gemini / Codex review the difficult cases, then sync the final grouping plan back to Bilibili follow groups.

This project is for people who follow a lot of Bilibili creators and want something stronger than Bilibili's ordinary manual grouping workflow.

Important clarification:

• Bilibili follow groups do not provide a strong built-in AI grouping workflow.
• This repository does not rely on a built-in AI model inside the script.
• The real workflow is: export structured data -> run local rules -> optionally ask a stronger external model to review edge cases -> sync the final result back.

What this project is

This is a follow-management pipeline:

1. Fetch followed creators and their public signals.
2. Classify them with editable local rules.
3. Export a human-readable summary.
4. Optionally let a frontier LLM review suspicious cases.
5. Sync the final grouping plan back to Bilibili follow groups.

Do I need an API key?

No, not for the recommended workflow.

The intended beginner path is:

• run the local scripts
• open the exported Markdown summary
• send it to ChatGPT web/app, Claude web/app, Gemini, Codex, Claude Code, OpenCode, or another strong general-purpose model
• ask for review suggestions
• apply the final result locally and sync back

API integration is an advanced automation mode, not the default.

Beginner quick start

Step 1. Install dependencies

pip install -r requirements.txt

Step 2. Prepare the data folder

Copy data_example to data.

Step 3. Fill in Bilibili cookies

Edit data/config.json and fill:

• sessdata
• bili_jct
• buvid3
• dedeuserid

Step 4. Run the first fetch

python fetch.py all

You should then see:

• data/up主详细数据.json

Step 5. Run local classification

python classify.py

You should then see:

• data/分类结果.json
• data/分类结果.md

Step 6. Review and sync

1. Open data/分类结果.md
2. Give it to your preferred model
3. Ask for suspicious items, overly broad categories, or suggested manual overrides
4. Update your local rules
5. Preview:

python sync_groups.py --dry-run

6. Sync:

python sync_groups.py

Screenshot-style beginner walkthrough

What you edit first

Open:

• data/config.json
• data/classify_rules.json

You mainly need:

• cookies in config.json
• category names in classify_rules.json

What you run first

python fetch.py all

After that, imagine the first “result screen” as:

• a JSON file appears in data/
• it contains your followed creators
• nothing has been written back to Bilibili yet

What you run second

python classify.py

Now you should get:

• a machine-readable result file
• a Markdown summary for human/AI review

What you hand to AI

You do not need to build an API flow. Just open or upload:

• data/分类结果.md

Recommended questions:

• Which creators are obviously in the wrong group?
• Which manual overrides should I add?
• Which categories are too broad?
• Which keyword rules need improvement?

What you sync back

Only after your manual review and dry-run:

python sync_groups.py --dry-run
python sync_groups.py

That is the point where Bilibili groups are actually changed.

Typical maintenance flow

After the first full run, the normal long-term workflow is:

1. fetch newly followed creators
2. classify again
3. review edge cases with your preferred AI tool
4. dry-run
5. sync

Incremental add:

python add_new.py <mid>

Files that matter

| File | Purpose | | --- | --- | | data/config.json | local cookies and runtime config | | data/classify_rules.json | your category system | | data/up主详细数据.json | fetched creator data | | data/分类结果.json | machine-readable result | | data/分类结果.md | AI-friendly review file | | generate_info.py | regenerate readable summaries | | sync_groups.py | dry-run or real sync |

Common misunderstandings

“Does Bilibili already have good AI grouping for this?”

No. This project exists because the normal grouping experience is limited, and many users want a stronger workflow.

“Does this repo come with a hidden powerful AI model?”

No. The strength comes from:

• your structured export
• your editable rules
• whichever strong external model you choose for review

“Do I have to pay for APIs?”

No. Subscription-based web/app tools are the default path.

Safety notes

• Always run --dry-run before real sync.
• Cookies expire and need refresh.
• Very large fetches may hit throttling.
• Group names cannot contain /.

License

MIT