Start asynchronous assistant generation and return its request ID

post

https://BASE_PATH/v2/gpt/generate/async

Recent Requests

Time	Status	User Agent
Retrieving recent requests…

Loading…

Body Params

V2 GPT Generate Command.

Key differences from V1:

conversationId is required (was optional)
Uses documentProcessingMethod enum ("raw", "rag", "truncate") instead of ragSkipEnabled boolean
Removed deprecated conversation and summarize fields
Removed task field from messageHistory items

content

string

The prompt/question to send to the LLM. Optional when messageHistory ends with a tool result.

conversationId

string

required

Required unique identifier for the conversation. Generate a UUID if not tracking conversations.

messageHistory

array of objects

Previous messages in the conversation

messageHistory

sources

array of objects

Source documents for context

sources

model

string

The LLM model to use

documentProcessingMethod

string

enum

Defaults to rag

Method for processing documents. Replaces ragSkipEnabled boolean.

rag: Use RAG (default). Documents are chunked and retrieved based on relevance.
raw: Process documents without RAG. Fails if sources exceed context window.
truncate: Truncate documents that exceed the context window.

Allowed:

maxTokens

integer

Maximum tokens in the response

temperature

float

Temperature for response generation (0.0 to 2.0)

topK

integer

Top-k sampling parameter

topP

float

Top-p (nucleus) sampling parameter

frequencyPenalty

float

Frequency penalty for token generation

llmSeed

integer

Seed for reproducible LLM outputs

logprobs

integer

Number of log probabilities to return

stream

boolean

Enable streaming response

transientData

boolean

Defaults to false

If true, data will not be persisted

embedder

string

The embedder to use for vector generation

generateCitations

boolean

Generate citations in the response

ragQuery

string

Custom query for RAG retrieval (overrides content)

retriever

object

RAG retriever configuration

prompts

object

Custom prompts for generation

searchOperator

string

enum

Defaults to AND

Logical operator to combine search conditions

Allowed:

getMetadata

array of strings

Metadata field names to request from certara-gpt (get_metadata). Returned as top-level properties on each item in chunksUsed.

getMetadata

searchFilters

array of objects

Filters to apply to RAG search

searchFilters

tools

array of objects

Tools the model may choose to invoke. Tool execution remains the caller's responsibility.

tools

Responses

400

Bad Request

401

Unauthorized

403

Forbidden

Language

Credentials

OAuth2

Loading…

Response

Click Try It! to start a request and see the response here! Or choose an example:

application/json

202Accepted