Knowledge Base Maintenance

A CLI-managed RAG index goes stale when source content changes. Use ragrails ingest to rebuild sources, ragrails edit to replace known chunk IDs, and ragrails delete to remove stale IDs.

Refresh a Source

For simple refreshes, run ingest again against the changed source. New chunks are upserted by ID.

ragrails ingest --docs files/refund-policy.pdf \
  --vector-db qdrant \
  --collection support \
  --url http://localhost:6333

For long-lived indexes, keep your own manifest of source files and chunk IDs so you can delete IDs that are no longer produced.

Edit Known Chunks

edit reads chunk dictionaries from JSON files, re-embeds them, and replaces stored chunks by exact ID.

ragrails edit \
  --input-dir files/updated-chunks/ \
  --vector-db qdrant \
  --collection support \
  --url http://localhost:6333 \
  --provider voyage \
  --model voyage-3

Output:

Requested  : 1
Edited     : 1
Failed     : 0
Provider   : qdrant
Collection : support

Delete Stale Chunks

ragrails delete \
  --id refund-policy-0 \
  --id refund-policy-1 \
  --vector-db qdrant \
  --collection support \
  --url http://localhost:6333

Deletes are exact-ID operations. They do not perform semantic matching.

Practical Manifest Pattern

Save extraction/chunking output during ingest when you need auditability.
Store the produced chunk IDs in your own manifest by source path or URL.
On refresh, ingest or edit the changed source.
Delete IDs from the old manifest that no longer exist in the new output.
Run ragrails query to verify the expected chunks are retrieved.

Options

`edit`

Option	Default	Description
`--input-dir`	required	Folder containing edited chunk JSON files.
`--vector-db`	`qdrant`	Vector DB provider.
`--collection`	config or none	Collection, index, or class name.
`--url`	config or none	Vector DB URL.
`--provider` / `--model`	`voyage` / `voyage-3`	Embedding provider/model for re-embedding.
`--batch-size`	`64`	Chunks per edit batch.

`delete`

Option	Default	Description
`--id`	required	Chunk ID to delete. Repeatable.
`--vector-db`	`qdrant`	Vector DB provider.
`--collection`	config or none	Collection, index, or class name.
`--url`	config or none	Vector DB URL.

Storing

Use store, edit, and delete directly.

Retrieval

Verify updated chunks.

Get Started

Workflows

Pipeline

Maintenance

Knowledge Base Maintenance

Refresh a Source

Edit Known Chunks

Delete Stale Chunks

Practical Manifest Pattern

Options

`edit`

`delete`

Storing

Retrieval

​Refresh a Source

​Edit Known Chunks

​Delete Stale Chunks

​Practical Manifest Pattern

​Options

​edit

​delete

Storing

Retrieval

Refresh a Source

Edit Known Chunks

Delete Stale Chunks

Practical Manifest Pattern

Options

`edit`

`delete`