CausalIQ Knowledge API Reference

API documentation for causaliq-knowledge, organised by module.

Import Patterns

Graph generation classes are available from the graph submodule:

from causaliq_knowledge.graph import (
    # Network context (main model)
    NetworkContext,
    NetworkLoadError,
    # Variable specification
    VariableSpec,
    VariableType,
    VariableRole,
    # Supporting models
    ViewDefinition,
    Provenance,
    LLMGuidance,
    Constraints,
    CausalPrinciple,
    GroundTruth,
    PromptDetails,
    # Filtering
    ViewFilter,
    PromptDetail,
    # Generation
    GraphGenerator,
    GraphGeneratorConfig,
    GeneratedGraph,
    ProposedEdge,
    GenerationMetadata,
    # Parameters
    GenerateGraphParams,
    # Prompts
    GraphQueryPrompt,
    OutputFormat,
    # Cache
    GraphCompressor,
)

Cache infrastructure is available from causaliq-core:

from causaliq_core.cache import TokenCache

LLM clients should be imported from the llm submodule:

from causaliq_knowledge.llm import (
    # Abstract base interface
    BaseLLMClient,
    LLMConfig,
    LLMResponse,
    # Vendor clients
    GroqClient,
    GroqConfig,
    GeminiClient,
    GeminiConfig,
    OpenAIClient,
    OpenAIConfig,
    AnthropicClient,
    AnthropicConfig,
    DeepSeekClient,
    DeepSeekConfig,
    MistralClient,
    MistralConfig,
    OllamaClient,
    OllamaConfig,
)

Modules

Graph Module

LLM-based causal graph generation from network context specifications:

• Graph Generator - Generate complete causal graphs
• GraphGenerator, GraphGeneratorConfig
• GeneratedGraph, ProposedEdge, GenerationMetadata
• Network Context - Pydantic models for network context
• NetworkContext, NetworkLoadError
• VariableSpec, VariableType, VariableRole
• PromptDetails, ViewDefinition, Provenance, Constraints
• View Filter - Extract context levels
• ViewFilter, PromptDetail (MINIMAL, STANDARD, RICH)
• Graph Prompts - Prompt builders
• GraphQueryPrompt, OutputFormat
• Response Models - Response parsing
• ProposedEdge, GeneratedGraph, GenerationMetadata

LLM Client Interface

Abstract base class and common types for LLM vendor clients:

• BaseLLMClient - Abstract interface all vendor clients implement
• LLMConfig - Base configuration dataclass
• LLMResponse - Unified response format

Vendor API Clients

Direct API clients for specific LLM providers. All implement the BaseLLMClient interface.

• Groq Client - Fast inference via Groq API
• Gemini Client - Google Gemini API
• OpenAI Client - OpenAI GPT models
• Anthropic Client - Anthropic Claude models
• DeepSeek Client - DeepSeek models
• Mistral Client - Mistral AI models
• Ollama Client - Local LLMs via Ollama

CLI

Command-line interface for graph generation and cache management.