Skip to main content
POST
https://app.artanis.ai
/
api
/
v1
/
feedback
Submit Feedback
curl --request POST \
  --url https://app.artanis.ai/api/v1/feedback \
  --header 'Authorization: Bearer <token>' \
  --header 'Content-Type: application/json' \
  --data '
{
  "trace_id": "<string>",
  "feedback_id": "<string>",
  "rating": {},
  "comment": "<string>",
  "correction": {},
  "labeller": {
    "labeller.name": "<string>",
    "labeller.email": "<string>"
  },
  "timestamp": "<string>"
}
'
{
  "status": "<string>"
}

​
Overview

The /api/v1/feedback endpoint allows you to record user feedback for traces. This is essential for:
  • Linking user satisfaction to specific operations
  • Building evaluation datasets from production data
  • Identifying problematic patterns
  • Measuring improvement over time
Feedback can be binary (positive/negative), numeric (0.0-1.0), or include corrections showing what the right answer should have been.
Feedback is a powerful way to understand which traces are working well and which need improvement. Corrections are especially valuable for building eval sets.

​
Request Body

​
trace_id
string
required
The trace ID to provide feedback for. Must match a trace created via /api/v1/traces.Example: "trace_a3f9c2e1b8d4f7a6c9e2"
Don’t have a trace ID? Use the Trace Search endpoint to find traces by your own identifiers (conversation ID, criterion ID, etc.).
​
feedback_id
string
Your own ID for this feedback record. Enables upsert behavior:
  • If omitted, creates a new feedback record each time
  • If provided, updates the existing record with this ID (or creates if not found)
Useful for auto-save scenarios where you want to update the same feedback record as the user edits.Example: "qa_evaluation_result_bgNzRGaJjv8T4obgJEWy38vA"
​
rating
string | number
User rating for the trace.Binary ratings:
  • "positive" — User found the result helpful/correct
  • "negative" — User found the result unhelpful/incorrect
Numeric ratings:
  • Any number between 0.0 and 1.0 (e.g., 0.85 for 4.25/5 stars)
Example: "positive" or 0.85
​
comment
string
Optional user comment providing context for the rating.Useful for understanding why users gave certain feedback.Example: "The response was too technical and difficult to understand."
​
correction
object
Optional correction data showing what the right answer should have been.Can have any structure matching your use case. Commonly used fields:
  • answer — The correct answer
  • category — Correct classification
  • explanation — Why the original was wrong
Example: {"answer": "We offer a 60-day guarantee, not 30 days."}
​
labeller
object
Optional information about who provided the feedback/correction.Example: {"name": "Jane Smith", "email": "jane.smith@example.com"}
​
timestamp
string
ISO 8601 timestamp. Defaults to current time if not provided.Format: YYYY-MM-DDTHH:mm:ss.sssZExample: "2025-12-22T10:31:00.000Z"

​
Response

​
status
string
Returns "success" when the feedback is successfully recorded.
After submitting feedback, use the Attribution endpoint to analyze the root cause of any disagreement.

​
Binary Feedback

Simple positive or negative feedback is the easiest way to track user satisfaction: 
import requests
from datetime import datetime

requests.post(
    "https://app.artanis.ai/api/v1/feedback",
    headers={
        "Authorization": "Bearer ak_...",
        "Content-Type": "application/json"
    },
    json={
        "trace_id": trace_id,
        "rating": "positive",
        "timestamp": datetime.utcnow().isoformat(timespec='milliseconds') + 'Z'
    }
)
Binary feedback is perfect for thumbs up/down, like/dislike, or similar UI elements.

​
Numeric Feedback

Record granular feedback using a numeric score between 0.0 and 1.0: 
# 5-star rating converted to 0.0-1.0
stars = 4  # User gave 4 out of 5 stars
rating = stars / 5.0

requests.post(
    "https://app.artanis.ai/api/v1/feedback",
    headers={
        "Authorization": "Bearer ak_...",
        "Content-Type": "application/json"
    },
    json={
        "trace_id": trace_id,
        "rating": rating,  # 0.8
        "timestamp": datetime.utcnow().isoformat(timespec='milliseconds') + 'Z'
    }
)
Numeric ratings allow for more nuanced feedback analysis. Convert your rating system (e.g., 1-5 stars) to the 0.0-1.0 scale.

​
Feedback with Comments

Add optional comments to provide context: 
requests.post(
    "https://app.artanis.ai/api/v1/feedback",
    headers={
        "Authorization": "Bearer ak_...",
        "Content-Type": "application/json"
    },
    json={
        "trace_id": trace_id,
        "rating": "negative",
        "comment": "The response was too technical and difficult to understand.",
        "timestamp": datetime.utcnow().isoformat(timespec='milliseconds') + 'Z'
    }
)
Comments help you understand why users gave the feedback they did. This is valuable for identifying patterns and improvement opportunities.

​
Feedback with Corrections

When users provide the correct answer, capture it as a correction: 
requests.post(
    "https://app.artanis.ai/api/v1/feedback",
    headers={
        "Authorization": "Bearer ak_...",
        "Content-Type": "application/json"
    },
    json={
        "trace_id": trace_id,
        "rating": "negative",
        "comment": "Wrong refund period",
        "correction": {
            "answer": "We offer a 60-day money-back guarantee, not 30 days."
        },
        "timestamp": datetime.utcnow().isoformat(timespec='milliseconds') + 'Z'
    }
)
Corrections are incredibly valuable for building evaluation datasets. They show exactly what the right answer should have been.

​
Correction Data Structures

Corrections can have any structure that matches your use case: 
{
  "trace_id": "trace_...",
  "rating": "negative",
  "correction": {
    "category": "billing",
    "subcategory": "refund",
    "priority": "high"
  },
  "timestamp": "2025-12-22T10:31:00.000Z"
}

​
Best Practices

​
1. Collect Feedback Immediately

Capture feedback as soon as users provide it. Don’t batch or delay - this ensures you don’t lose valuable data.

​
2. Make Feedback Easy

Use simple UI elements:
  • Thumbs up/down buttons
  • Star ratings
  • Like/dislike buttons
  • Optional comment box for details

​
3. Request Corrections

When users give negative feedback, ask what the correct answer should have been. This is gold for building eval sets. If you have user IDs in your trace metadata, you can analyze feedback patterns by user segment, customer, or use case.

​
5. Monitor Feedback Rates

Track what percentage of traces receive feedback. Low rates might indicate:
  • Feedback UI is not prominent enough
  • Users don’t understand what feedback means
  • Feedback process is too complex

​
Error Responses

Status CodeErrorSolution
400Invalid request formatCheck JSON syntax and required fields
401Invalid API keyVerify your API key is correct
404Trace not foundEnsure trace_id matches an existing trace
429Rate limit exceededImplement exponential backoff retry

​
Next Steps

Search Traces

Find trace IDs using your own identifiers

Complete Example

See a full end-to-end workflow with all endpoints

Slack Integration

Automatically match Slack feedback to traces