Feature Request: Add cod-like learning and memory capabilities

[Clausinho]

Feature Request

I'm looking at migrating from cod to carapace for shell completions, and I'd love to see cod's learning capabilities integrated into carapace.

What cod does well

1. Learns from usage patterns - cod learns which commands and arguments you use most frequently
2. Remembers sessions - maintains context across sessions for smarter completions
3. Personalized suggestions - prioritizes suggestions based on your usage history
4. Auto-learns new commands - discovers and adapts to new tools you install

What carapace does well

1. Excellent descriptions - provides detailed help text for completions
2. Wide support - works with many CLIs out of the box
3. Clean output format - integrates beautifully with fzf

Suggestion

Could carapace add an optional learning layer that:

• Tracks command usage patterns (with opt-in privacy)
• Prioritizes frequently-used options in completion lists
• Remembers project-specific command patterns
• Provides a persistent cache of learned completions

This would give us the best of both worlds:

• Carapace's excellent descriptions and wide support
• Cod's intelligent, personalized suggestions

Use case

As a developer who works on multiple projects, I often:

• Use different commands in different contexts (e.g., kubectl commands vary by cluster)
• Have preferred flags I always use (e.g., --format json, --watch)
• Want completions to learn from my actual usage, not just generic help text

Having carapace learn from my usage would significantly speed up my workflow while maintaining the excellent descriptions carapace provides.

Thanks for considering this!

[rsteube]

Thanks for the input, I've got something in the works with ideas like this in mind.

related carapace-sh/carapace-bin#3079