Documentation Index Fetch the complete documentation index at: https://mintlify.com/Evincere/klisk/llms.txt
Use this file to discover all available pages before exploring further.
Builtin tools are provider-hosted capabilities that extend your agent without writing custom code. They’re maintained and executed by the model provider (OpenAI).
All builtin tools require OpenAI models. They will be automatically disabled with a warning if you use a non-OpenAI model like Anthropic or Gemini.
Klisk supports four builtin tools:
web_search Search the web for real-time information
code_interpreter Execute Python code in a sandboxed environment
file_search Semantic search across vector stores
image_generation Generate images from text descriptions
Usage Pass builtin tools to define_agent() via the builtin_tools parameter:
from klisk import define_agent agent = define_agent( name = "Assistant" , model = "gpt-5.2" , builtin_tools = [ "web_search" , "code_interpreter" ], )
String shortcuts vs objects You can specify builtin tools in two ways:
1. String shortcuts (uses default configuration):builtin_tools = [ "web_search" , "code_interpreter" , "image_generation" ]
2. Configured objects (customize behavior):from klisk import WebSearch, CodeInterpreter, ImageGeneration builtin_tools = [ WebSearch( search_context_size = "high" ), CodeInterpreter( container = { "image" : "python:3.11" }), ImageGeneration( model = "gpt-image-1" , quality = "high" ), ]
file_search requires the object form because it needs vector_store_ids. You cannot use the string shortcut.
web_search Search the web for real-time information.
Basic usage define_agent( name = "Researcher" , model = "gpt-5.2" , builtin_tools = [ "web_search" ], )
Configuration from klisk import define_agent, WebSearch agent = define_agent( name = "Researcher" , model = "gpt-5.2" , builtin_tools = [ WebSearch( search_context_size = "high" ), ], )
Parameters: How much context to return per search result. Options: "low", "medium", "high"
When to use:
Answering questions about current events
Fact-checking with recent information
Research tasks requiring multiple sources
Looking up documentation or articles
Example conversation: User: What are the latest developments in AI agents? Agent: [Uses web_search] Based on recent articles...
code_interpreter Execute Python code in a sandboxed environment. Useful for data analysis, calculations, and file processing.
Basic usage define_agent( name = "DataAnalyst" , model = "gpt-5.2" , builtin_tools = [ "code_interpreter" ], )
Configuration from klisk import define_agent, CodeInterpreter agent = define_agent( name = "DataAnalyst" , model = "gpt-5.2" , builtin_tools = [ CodeInterpreter( container = { "image" : "python:3.11" , "memory" : "2gb" }, ), ], )
Parameters: Container configuration for the code execution environment. Optional.
When to use:
Analyzing CSV or JSON data
Performing complex mathematical calculations
Generating charts and visualizations
Processing files uploaded by users
Example conversation: User: Analyze this CSV and find the average sales by region Agent: [Uses code_interpreter to load CSV and compute statistics]
file_search Semantic search across vector stores. Query large document collections using natural language.
file_search cannot use the string shortcut. You must use the object form with vector_store_ids.
Configuration (required) from klisk import define_agent, FileSearch agent = define_agent( name = "DocumentAssistant" , model = "gpt-5.2" , builtin_tools = [ FileSearch( vector_store_ids = [ "vs_abc123" ], max_num_results = 10 , ), ], )
Parameters: List of OpenAI vector store IDs to search. Must be created via the OpenAI API or dashboard.
Maximum number of search results to return. Optional.
Creating vector stores Vector stores must be created separately using the OpenAI API:
import openai # Create a vector store vector_store = openai.beta.vector_stores.create( name = "Company Documentation" , ) # Upload files with open ( "docs.pdf" , "rb" ) as f: openai.beta.vector_stores.files.create( vector_store_id = vector_store.id, file = f, ) print (vector_store.id) # vs_abc123
When to use:
Answering questions from company documentation
Searching through large knowledge bases
Building RAG (Retrieval-Augmented Generation) systems
Querying uploaded documents
Example conversation: User: What is our refund policy? Agent: [Uses file_search to find relevant policy documents] According to the policy...
image_generation Generate images from text descriptions using OpenAI’s image generation models.
Basic usage define_agent( name = "Designer" , model = "gpt-5.2" , builtin_tools = [ "image_generation" ], )
Configuration from klisk import define_agent, ImageGeneration agent = define_agent( name = "Designer" , model = "gpt-5.2" , builtin_tools = [ ImageGeneration( model = "gpt-image-1" , quality = "high" , size = "1536x1024" , ), ], )
Parameters: model
string
default: "gpt-image-1"
Image generation model to use
Image quality. Options: "auto", "low", "medium", "high"
Image dimensions. Options: "auto", "1024x1024", "1536x1024", "1024x1536"
When to use:
Creating illustrations for content
Generating marketing materials
Visualizing concepts or ideas
Prototyping designs
Example conversation: User: Create an image of a sunset over mountains Agent: [Uses image_generation] Here's the image...
OpenAI-only restriction Builtin tools are implemented by OpenAI and only work with OpenAI models. If you specify builtin tools with a non-OpenAI model, Klisk will:
Issue a warning at agent creation time
Disable the builtin tools
Continue with only custom tools
Example: # ❌ This will warn and disable web_search define_agent( name = "Researcher" , model = "anthropic/claude-sonnet-4-5-20250929" , builtin_tools = [ "web_search" ], # Disabled ) # Warning output: # "Builtin tools (web_search) are only supported with OpenAI models # and will be disabled for 'anthropic/claude-sonnet-4-5-20250929'."
Solution: Either:
Switch to an OpenAI model (gpt-5.2, gpt-4o, etc.)
Remove builtin tools and use custom tools instead
This check happens in primitives.py:109.
You can mix builtin tools with custom tools:
from klisk import define_agent, get_tools, WebSearch agent = define_agent( name = "Assistant" , model = "gpt-5.2" , tools = get_tools( "format_response" , "save_data" ), builtin_tools = [ WebSearch( search_context_size = "high" ), "code_interpreter" , ], )
The agent can use all tools together:
User: Search for Python tutorials and save the best ones to a file Agent: 1. [Uses web_search to find tutorials] 2. [Uses code_interpreter to filter and rank results] 3. [Uses custom save_data tool to write to file]
Complete example from klisk import ( define_agent, get_tools, WebSearch, CodeInterpreter, FileSearch, ImageGeneration, ) # Research agent with web search researcher = define_agent( name = "Researcher" , model = "gpt-5.2" , instructions = "Search the web and analyze findings." , builtin_tools = [ WebSearch( search_context_size = "high" ), CodeInterpreter(), ], ) # Document assistant with file search doc_assistant = define_agent( name = "DocAssistant" , model = "gpt-5.2" , instructions = "Answer questions from company documentation." , builtin_tools = [ FileSearch( vector_store_ids = [ "vs_abc123" ], max_num_results = 5 , ), ], ) # Creative agent with image generation designer = define_agent( name = "Designer" , model = "gpt-5.2" , instructions = "Create visual content from descriptions." , builtin_tools = [ ImageGeneration( quality = "high" , size = "1536x1024" , ), ], ) # Full-featured assistant assistant = define_agent( name = "FullAssistant" , model = "gpt-5.2" , instructions = "A versatile assistant with all capabilities." , tools = get_tools( "custom_tool_1" , "custom_tool_2" ), builtin_tools = [ "web_search" , "code_interpreter" , "image_generation" ], )
Next steps
Tools Learn how to create custom tools
Models Understand model selection and multi-provider support