Create Autocorrection

POST /v1/autocorrect

Creates a new autocorrection job to detect and fix hallucinations, factual errors, or quality issues in AI-generated responses.

Authentication

• Type: Bearer Token

Request Body

Field	Type	Required	Description
prompt	string	Yes	The original user prompt or question
response	string	Yes	The AI-generated response to check and potentially correct
context	string	Yes	Reference information to verify against
top_k	number	No	Number of top chunks to consider (default: 3)

Configuration Parameters

Parameter	Type	Required	Description
risk_threshold	number	No	Threshold for detecting issues (0-1, default: 0.7)
enable_tools	boolean	No	Allow tools usage for verification (default: false)
web_search	boolean	No	Enable web search for fact verification (default: false)
correction_strategy	string	No	Strategy for corrections: "conservative" or "aggressive"

Example Request

curl -X POST https://sandbox-api.pegasi.ai/v1/autocorrect \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "What were Apple\'s Q1 2025 earnings?",
    "response": "Apple reported $150 billion in revenue for Q1 2025, a 12% increase year-over-year.",
    "context": "Apple Inc. (NASDAQ: AAPL) today announced financial results for its fiscal 2025 first quarter ended December 28, 2024. The Company posted quarterly revenue of $123.9 billion, an increase of 8% year over year.",
    "top_k": 3,
    "risk_threshold": 0.65,
    "enable_tools": true,
    "web_search": true,
    "correction_strategy": "conservative"
  }'

Response

{
  "id": "autocorr_8f7d6e5c4b3a2f1e",
  "status": "processing",
  "created_at": "2025-06-11T22:34:30Z"
}

You can check the status and results of the autocorrection using the Get Autocorrection endpoint with the returned ID.

Tool Usage

When enable_tools is set to true, the autocorrection system can use various verification tools:

• Web Search: When web_search is true, searches the web for factual verification
• Calculator: Verifies numerical calculations and comparisons
• Date Checker: Validates temporal claims and date references
• Entity Resolver: Disambiguates and verifies entity references

Correction Strategies

• Conservative: Only corrects clear factual errors while preserving the original response style and structure
• Aggressive: Makes more extensive corrections including style, clarity, and completeness improvements