Retrieval Pipeline Tool

Runs only the retrieval stage — queries the knowledge base and returns matched chunks. No LLM generation is performed.

Endpoint

POST /components/{chatbot_id}:retrieval-pipeline/run

Example Request

curl -X POST "https://glchat.glair.ai/components/{chatbot_id}:retrieval-pipeline/run" \
  -H "Content-Type: application/json" \
  -H "X-API-Key: {your_user_api_key}" \
  -d '{
    "inputs": {
      "standalone_query": "What is our refund policy?"
    },
    "config": {}
  }'

Response

{
  "chunks": [
    {
      "id": "54652883-dc0e-4e08-928e",
      "content": "Returns are accepted within 30 days of purchase...",
      "metadata": { "source": "policy-doc.pdf", ...},
      "score": 0.71
    }
  ]
}

Field	Type	Description
chunks	list	Retrieved chunks sorted by relevance score (descending)

Inputs Schema

Field	Type	Required	Default	Description
standalone_query	string	Yes	—	Query used for knowledge base retrieval
retrieval_params	dict	No	{}	Low-level parameters forwarded to the vector store query
steps	list[dict]	No	[]	Custom processing steps to override default retrieval flow
attachment_chunks	list[Chunk]	No	[]	Pre-loaded chunks from file attachments
search_chunk_results	list[Chunk]	No	[]	Pre-fetched search chunks to inject alongside retrieval
memory_results	list[Chunk]	No	[]	Memory chunks from the user's memory store
context	string	No	""	Additional context string