Skip to main content
POST
/
api
/
r2r
/
v3
/
retrieval
/
rag
RAG Query
curl --request POST \
  --url https://api.intelligence.io.solutions/api/r2r/v3/retrieval/rag \
  --header 'Authorization: Bearer <token>' \
  --header 'Content-Type: application/json' \
  --data '{
  "query": "<string>",
  "search_mode": "custom",
  "search_settings": {
    "use_hybrid_search": false,
    "use_semantic_search": true,
    "use_fulltext_search": false,
    "filters": "<string>",
    "limit": 10,
    "offset": "0",
    "include_metadatas": true,
    "include_scores": true,
    "search_strategy": "vanilla",
    "hybrid_settings": {
      "full_text_weight": 1,
      "semantic_weight": 5,
      "full_text_limit": 200,
      "rrf_k": 50
    },
    "chunk_settings": {
      "index_measure": "l2_distance",
      "probes": 10,
      "ef_search": 40,
      "enabled": true
    },
    "graph_settings": {
      "limits": [
        "<any>"
      ],
      "enabled": true
    },
    "num_sub_queries": 5
  },
  "rag_generation_config": {
    "model": "<string>",
    "temperature": 123,
    "top_p": 123,
    "max_tokens_to_sample": 123,
    "stream": true,
    "functions": [
      "<any>"
    ],
    "tools": [
      "<any>"
    ],
    "add_generation_kwargs": [
      "<any>"
    ],
    "api_base": "<string>",
    "response_format": [
      {
        "Base Model": {}
      }
    ],
    "extended_thinking": false,
    "thinking_budget": 123,
    "reasoning_effort": "<string>"
  },
  "task_prompt": "<string>",
  "include_title_if_available": false,
  "include_web_search": false
}'
{
  "key": "value"
}
The RAG Query endpoint executes a Retrieval-Augmented Generation (RAG) workflow by combining semantic search, optional knowledge graph integration, and large language model (LLM) generation. It returns contextually grounded, source-cited responses derived from your document corpus and external web content (if enabled). This endpoint is ideal for applications that require explainable AI answers, document-grounded responses, and real-time contextual reasoning.

​
Key Features

  • Combined retrieval and generation: Merges vector search, optional graph traversal, and LLM output generation in one request.
  • Automatic source citation: Each referenced document includes a unique citation identifier.
  • Streaming and non-streaming modes: Supports token-level updates or full-response delivery.
  • Provider flexibility: Compatible with OpenAI, Anthropic, Ollama, and other LiteLLM-supported models.
  • Web search integration: Optionally augments internal context with real-time external data.

​
Model Support

ProviderDescription
OpenAIDefault provider supporting GPT-based models (gpt-4o, gpt-4o-mini, etc.).
AnthropicSupports Claude models (requires ANTHROPIC_API_KEY).
OllamaEnables local model execution via Ollama runtime.
LiteLLMProvides access to additional supported model providers.

​
Request Body

The request body combines search configuration (for retrieval) and generation configuration (for LLM behavior).
All search parameters available in the /search endpoint can be reused here, including filters, hybrid search, and graph-enhanced retrieval.

​
Generation Configuration

Control model behavior using the rag_generation_config object. Example: 
{
    "model": "openai/gpt-4o-mini",
    "temperature": 0.7,
    "max_tokens": 1500,
    "stream": true
}
Parameters:
  • model: Specifies the model used for generation.
  • temperature: Controls output randomness (0 for deterministic, 1 for creative).
  • max_tokens: Sets maximum output length.
  • stream: Enables or disables token streaming for real-time responses.

​
Streaming Responses

When stream: true is enabled, the API emits Server-Sent Events (SSE) during processing.
Each event type corresponds to a distinct phase of the retrieval and generation workflow.
Event TypeDescription
search_resultsContains the initial search results from your documents.
messageStreams partial tokens as the model generates them.
citationEmits citation metadata when a source is referenced.
final_answerContains the complete, generated response with structured citations.
Example Response: 
{
	"generated_answer": "DeepSeek-R1 is a model that demonstrates impressive performance...[1]",
	"search_results": { ... },
	"citations": [
    	{
        	"id": "cit.123456",
        	"object": "citation",
        	"payload": { ... }
    }
]
}

Authorizations

​
Authorization
string
header
required

The access token received from the authorization server in the OAuth 2.0 flow.

Body

application/json
​
query
string

The user's question

​
search_mode
enum<string>
default:custom

Default value of custom allows full control over search settings. Pre-configured search modes: basic: A simple semantic-based search. advanced: A more powerful hybrid search combining semantic and full-text. custom: Full control via search_settings. If filters or limit are provided alongside basic or advanced, they will override the default settings for that mode.

Available options:
basic,
advanced,
custom
​
search_settings
object

The search configuration object. If search_mode is custom, these settings are used as-is. For basic or advanced, these settings will override the default mode configuration. Common overrides include filters to narrow results and limit to control how many results are returned.

​
rag_generation_config
object

Configuration for RAG generation

​
task_prompt
string

Optional custom prompt to override default

​
include_title_if_available
boolean
default:false

Include document titles in responses when available

​
include_web_search
boolean
default:false

Include web search results provided to the LLM.

Response

200

​
key
string
Example:

"value"