Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/jxnl/kura/llms.txt

Use this file to discover all available pages before exploring further.

​
ConversationSummary

Extends GeneratedSummary with conversation-specific fields for tracking and retrieval.
​
chat_id
str
required
Unique identifier linking to the original conversation
​
metadata
dict
required
Metadata from the original conversation
​
embedding
Optional[list[float]]
Vector embedding of the summary for similarity search. Defaults to None
​
summary
str
required
​
request
Optional[str]
​
topic
Optional[str]
​
languages
Optional[list[str]]
​
task
Optional[str]
​
concerning_score
Optional[int]
​
user_frustration
Optional[int]
​
assistant_errors
Optional[list[str]]

​
GeneratedSummary

Base class containing all LLM-generated summary fields.
​
summary
str
required
A clear and concise summary of the conversation in at most two sentences, avoiding phrases like ‘Based on the conversation’ and excluding proper nouns or PII
​
request
Optional[str]
The user’s overall request for the assistant
​
topic
Optional[str]
The main high-level topic of the conversation (e.g., ‘software development’, ‘creative writing’, ‘scientific research’)
​
languages
Optional[list[str]]
Main languages present in the conversation including human and programming languages (e.g., [‘english’, ‘arabic’, ‘python’, ‘javascript’])
​
task
Optional[str]
The task the model is being asked to perform
​
concerning_score
Optional[int]
Safety concern rating from 1-5 scale. Constrained to values between 1 and 5 (ge=1, le=5)
​
user_frustration
Optional[int]
User frustration rating from 1-5 scale. Constrained to values between 1 and 5 (ge=1, le=5)
​
assistant_errors
Optional[list[str]]
List of errors the assistant made

​
Example

from kura.types import ConversationSummary

summary = ConversationSummary(
    chat_id="chat-123",
    summary="User asked for help with Python async code. Assistant provided examples and explained event loops.",
    request="Help understanding Python asyncio",
    topic="software development",
    languages=["english", "python"],
    task="code explanation",
    concerning_score=1,
    user_frustration=2,
    assistant_errors=[],
    metadata={
        "model": "claude-3-sonnet",
        "token_count": 450
    },
    embedding=[0.1, 0.2, 0.3, ...]  # 1536-dimensional vector
)

​
String Representation

The ConversationSummary class implements a custom __repr__() method that formats the summary as XML tags: 
print(summary)
# Output:
# <summary>User asked for help with Python async code...</summary>
# <topic>software development</topic>
# <request>Help understanding Python asyncio</request>
# <task>code explanation</task>
# <languages>['english', 'python']</languages>
# <assistant_errors>[]</assistant_errors>
# <model>claude-3-sonnet</model>
# <token_count>450</token_count>
Metadata fields are included in the output with type checking for str, int, float, bool, and lists of these types.

Build docs developers (and LLMs) love

Get started for freeTalk to us