Skip to content

Token estimates for curated context #404

Description

@jamesrochabrun

Parent: #398

Depends on: Context Builder core selection and assembly model.

Goal

Show approximate token cost for selected context before launch, with per-file and total estimates plus clear over-budget warnings.

Implementation Notes

Add a model-agnostic estimator first; exact provider tokenizers are not required for this issue.

Suggested behavior:

  • estimate tokens as roughly UTF-8 bytes / 4 with a small safety multiplier
  • label all values as approximate
  • calculate per-file and total selected context tokens
  • keep token work off the main actor and cached by path/content fingerprint where practical
  • warn when a selection exceeds a configured/default model window
  • do not silently drop selected files

Wire estimates into the Context Builder UI. Smart-mode plan row estimates can be a follow-up unless this issue already has the necessary context payload in place.

Acceptance Criteria

  • Selecting/deselecting files updates approximate per-file and total token counts.
  • Over-budget context shows a visible warning.
  • The app never silently removes files to fit a budget.
  • Empty context and missing model/window information degrade gracefully.

Tests

Add pure tests for token estimation and truncation/warning thresholds if implemented. Add view-model tests for total-token updates.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions