Issue Description
Summary
Implement an AI-powered interactive shell featuring a small, specialized language model trained specifically on apt commands. This would enable fast, offline natural language to apt command translation without requiring external API calls.
Motivation
Currently, Cortex relies on external LLM APIs (Anthropic/OpenAI) for natural language understanding. While powerful, this approach has limitations:
- Latency: API calls add network overhead to every command
- Cost: Each request incurs API usage costs
- Overkill: General-purpose LLMs are overqualified for apt command generation
Training Data
- Curate dataset of natural language queries → apt command pairs
- Include common installation patterns, search queries, package management tasks
- Cover edge cases: version pinning, repository management, dependency resolution