Workspace Vector DB Settings

Key features

• Database information — View vector DB identifier and vector count
• Max context snippets — How many document chunks are sent to the AI per query
• Similarity threshold — Minimum relevance score for a chunk to be included
• Reset database — Wipe all embedded documents and re-embed from scratch

Database information

Vector DB identifier

Shows which vector database backend this workspace uses (e.g., Chroma, Pinecone, Qdrant). Determined at workspace creation or by instance configuration.

Vector count

Displays the number of embedded vectors. Use Reload to refresh the count after adding or removing documents.

Search settings

Max context snippets (top N)

This controls how many document chunks are retrieved and sent to the AI when answering a question.

• Range: 1–200
• Default: 4
• Higher values — More context, potentially better answers, but more tokens and cost
• Lower values — Fewer tokens, faster responses, but may miss relevant content

Recommended:

• Start with 4–8; increase if answers seem incomplete, decrease if responses are slow or off-topic.

Document similarity threshold

Determines how similar a chunk must be to the user's question to be included. Chunks below this score are filtered out.

| Value | Label | Use case | |-------|-------|----------| | 0.0 | None | Include all retrieved chunks (broadest) | | 0.25 | Low | Default; good balance for most workspaces | | 0.5 | Medium | Stricter; fewer but more relevant chunks | | 0.75 | High | Strictest; only closely matching chunks |

When to adjust:

• Raise — If answers cite irrelevant or loosely related content
• Lower — If answers miss clearly relevant information

Database management

Reset vector database

Resetting the database removes all embedded vectors for this workspace. Documents must be re-uploaded and re-embedded to be searchable again. Use this when:

• Switching embedding models
• Fixing embedding or indexing issues
• Starting fresh after major document changes

To reset:

1. Click Reset in the Database Management section.
2. Confirm in the dialog.
3. Re-upload or re-embed documents as needed.

{{DIMABLOCK0}}

Saving changes

1. Max context snippets and Similarity threshold — Click Update workspace when changes are made.
2. Reset database — Runs immediately after confirmation; no separate save step.

Troubleshooting

1. Vector count not updating — Click the reload/refresh control in the Vector count section.
2. No documents found in chat — Check that documents are embedded (Saved Items → Documents) and that the similarity threshold is not too high.
3. Slow or irrelevant answers — Try lowering top N or raising the similarity threshold.

Related topics

For more information, see:

1. Workspace Settings Introduction — Overview of all settings tabs
2. Documents Guide — Managing workspace documents and embeddings