Skip to content

CausalIQ Knowledge - Development Roadmap

Last updated: January 18, 2026

This project roadmap fits into the overall ecosystem roadmap

🎯 Current Release

Release v0.3.0 - LLM Caching [January 2026] In Development

SQLite-based response caching with CLI tools for cache management.

Scope:

  • TokenCache - SQLite storage with WAL mode for concurrent access
  • CacheEncoder system - Extensible serialization (JsonEncoder, LLMEntryEncoder)
  • LLMCacheEntry - Structured model for cached responses
  • BaseLLMClient caching integration - Automatic cache-first lookup
  • CLI commands: cache stats, cache export, cache import
  • Human-readable export filenames with edge query detection
  • Zip archive support for export/import

✅ Previous Releases

Release v0.2.0 - Additional LLMs [January 2026]

Expanded LLM provider support from 2 to 7 providers.

Delivered:

  • OpenAI client for GPT-4o and GPT-4o-mini models
  • Anthropic client for Claude models
  • DeepSeek client for DeepSeek-V3 and R1 models
  • Mistral client for Mistral AI models
  • Ollama client for local LLM inference
  • OpenAI-compatible base client for API-compatible providers
  • Integration tests for all providers

Release v0.1.0 - Foundation LLM [January 2026]

Simple LLM queries to 1 or 2 LLMs about edge existence and orientation to support graph averaging.

Delivered:

  • Abstract KnowledgeProvider interface
  • EdgeKnowledge Pydantic model for structured responses
  • LLMKnowledge implementation using vendor-specific API clients
  • Direct API clients for Groq and Google Gemini
  • Single-model and multi-model consensus queries
  • Basic prompt templates for edge existence/orientation
  • CLI for testing queries
  • 100% test coverage
  • Comprehensive documentation

🛣️ Upcoming Releases

Release v0.4.0 - LLM Context

  • Variable descriptions and roles
  • Domain context specification
  • Literature retrieval (RAG) - evaluate lightweight alternatives to LangChain
  • Vector store integration for document search

Release v0.5.0 - Algorithm Integration

  • Integration hooks for structure learning algorithms
  • Knowledge-guided constraint generation
  • Integration with causaliq-analysis average() function
  • Entropy-based automatic query triggering

Release v0.6.0 - Legacy Reference

  • Support for deriving knowledge from reference networks
  • Migration of functionality from legacy discovery repo

📦 Dependencies Evolution

# v0.1.0 - v0.3.0 (current)
dependencies = [
    "click>=8.0.0",
    "httpx>=0.24.0",
    "pydantic>=2.0.0",
]

# v0.4.0 (evaluate - prefer lightweight solutions)
# Consider: llama-index, simple RAG, or custom implementation
# Avoid: langchain (heavy dependencies)