Skip to content

Feature: Prompt compression to reduce LLM token costs in long context sessions #12915

Description

@arjunkshah

Continue.dev accumulates significant context across coding sessions. Adding a prompt compression layer could reduce LLM token costs by ~65% by scoring each line of context against the current query and dropping irrelevant lines before sending to the LLM.

SuperCompress is an open-source (MIT) learned compression engine that runs in ~60ms on CPU: https://supercompress.vercel.app

This would be especially valuable for users with long Hub sessions where context accumulates across many turns.

Metadata

Metadata

Assignees

No one assigned

    Labels

    area:context-providersRelates to context providerskind:enhancementIndicates a new feature request, imrovement, or extension

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions