Cognee CLI - delete, add breakdown of affected data

[daukadolt]

Delete Command Preview (Enhancement)

NOTE: This issue is part of Contribute-to-Win. Please comment first to get assigned. Read the details here

Problem Statement

The cognee delete command currently removes data as requested (all data, by dataset, or by user). While it works as intended, users don’t get a clear overview of what will be affected before confirming the action.

Adding a preview step would improve usability and safety by helping users understand the scope of deletion before proceeding.

Proposed Solution

Enhance the delete command to include a preview step that summarizes what will be deleted.

The preview should display key metrics so the user can make an informed decision before confirming.

Preview metrics

When the user runs cognee delete (with any argument), we should show a breakdown like:

1. Datasets: How many datasets will be removed
2. Data entries: How many individual data items are affected
3. Other relevant metrics (e.g., users affected, chunks, embeddings)

Example:

You are about to delete:
- 3 datasets
- 152 data entries
- 1 user

~~~~~

Then confirm before proceeding:

Proceed? (y/N)

Cases

• Delete all
  • Show counts across the entire system
• Delete by dataset
  • Show counts scoped to that dataset
• Delete by user_id
  • Show counts of all datasets and data entries owned by that user

Non-Functional

• If nothing is found to delete → show No data found to delete. and exit
• Must run efficiently (counts only, not full data scan)
• If metrics can’t be retrieved → log warning and fallback to direct delete
• Default confirmation = No unless explicitly confirmed
• Add --force flag to skip preview and delete immediately (useful for automation/CI)

Integration

• Update cognee delete command in CLI
• Add preview logic before actual delete execution
• Extend tests to cover preview outputs for different argument combinations

Acceptance Criteria

1. Running cognee delete shows preview metrics before deletion
2. Preview correctly reflects counts for all, dataset, and user-specific deletes
3. Empty or invalid targets return No data found to delete.
4. --force skips preview and deletes immediately
5. Confirmation prompt defaults to safe (abort unless explicitly confirmed)
6. Metrics are logged before deletion begins

[github-actions]

Hello @daukadolt, thank you for your interest in our work!

If this is a bug report, please provide screenshots and minimum viable code to reproduce your issue, otherwise we can not help you.

[daukadolt]

A useful pointer - #1197 (comment)

[shehab-badawy]

Can i get assigned to this?

[hande-k]

hey @shehab-badawy, thanks for picking it up!

[Devashish2112]

allow me to participate in this

[hande-k]

hi @Devashish2112 the issue is already taken but I'd love you to pick another one!

Also, based on the challenge rules we shared , this task might be available again after 48h if not touched by the assignee yet. So you can wait if you like.

[shehab-badawy]

@daukadolt

I'm making progress on this issue and wanted to provide a quick update and ask for a pointer.

So far, I've set up my local environment and have been testing the current cognee delete command to establish a baseline. I discovered that after running cognee delete -d <dataset_name>, the dataset still appears in the Cognee UI, which suggests the command is failing silently.
To implement the preview, I need to write the counting logic, but I'm having trouble locating the core function responsible for deleting an entire dataset (not just a single document). My investigation has led me from delete_command.py to cognee.delete, but I'm unsure where the logic branches off from there to handle a full dataset deletion.

Could you please point me to the file containing the primary logic for delete_dataset? Knowing where that is will help me understand how to correctly query the database for the preview counts.

Thanks for the help!

[daukadolt]

Hey @shehab-badawy

First and foremost - thanks for the detailed update and walkthrough of your approach!

Apologies for the delay, currently in a bit of a crunch, but let me share details + some feedback on PR here:

1. Cognee.delete() does indeed need to be fixed separately, just wanted to confirm to help remove some of that confusion
2. I like the idea of creating the preview method

Some action points:

1. Rebase on the latest dev - -ui command is more stable there, let's not change anything about it
2. get_deletion_counts is there already, nice! Let's write test cases to validate the function output:

• single data deletion
• dataset_deletion
• data by user deletion

i.e. one per each delete option we support:

 if args.all:
                confirm_msg = "Delete ALL data from cognee?"
                operation = "all data"
            elif args.dataset_name:
                confirm_msg = f"Delete dataset '{args.dataset_name}'?"
                operation = f"dataset '{args.dataset_name}'"
            elif args.user_id:
                confirm_msg = f"Delete all data for user '{args.user_id}'?"
                operation = f"data for user '{args.user_id}'

Code organization / formatting also needs addressing, but we can do that once we'll have validated code.