Cabin Kit

Assemble, compress, and copy multi-file AI context with a live token budget. No API keys. Runs entirely in VS Code.

TL;DR: pick files, see your token count live, compress if needed, copy a clean multi-file prompt. Everything stays on your machine.

Features at a glance

• Context Builder -- file picker with live token count, four compression modes, named presets, and one-click prompt assembly; add files from the sidebar or via Explorer right-click
• Prompt Templates -- reusable scaffolds with placeholder support; fill in values and copy without storing anything permanently
• Line Filter -- regex log filter with chain support, per-step match counts, and configurable context lines around each match
• SQLite Viewer -- browse any .sqlite or .db file as a searchable, sortable, paginated table directly inside VS Code
• Team Tracker -- lightweight board for links, tasks, and notes, stored in VS Code global state

Getting started

1. Open the Cabin Kit panel in the Activity Bar.
2. Add files using the toolbar icons or by right-clicking a file or folder in the Explorer.
3. Choose a model and compression mode in the panel, then click Copy Prompt.

To count tokens for a file or folder without adding it to context, right-click in the Explorer and choose Count Tokens in Selection.

Context Builder

Tracks a token budget across every file you add, shows the live count in the status bar, and assembles the final prompt in one formatted block.

Status bar: displays as 12,345 / 25K (49%) where the denominator is the practical token limit for the selected model. Shifts to warning when you approach the limit and error when you exceed the hard context window.

Presets: save the current file list by name and reload it in one click. Presets are workspace-specific and stored locally.

Prompt format: each file is wrapped in a fenced code block with its path as a label. Copied to clipboard; never written to disk.

Compression modes

Compression is applied per file at copy time. The token count in the sidebar reflects the compressed size.

Filter large log files by regex pattern

Narrows any text file down to the lines you need without modifying the source. Open the Line Filter panel and click "Keep matching lines..." or "Remove matching lines...", or right-click any open file in the editor. Select text first and click "From selection" to use it as the pattern directly.

Results open as read-only tabs. Filter a result again to chain steps -- the history panel shows the full chain with per-step counts, e.g. ERROR (42) > auth (5).

Input format: plain text matches case-insensitively as a literal. Wrap in /slashes/ for regex with optional flags: /\bERROR\b/i.

Context lines: click "Context: N lines" to include N lines above and below each match, like grep -C.

Supported models

Token counting uses gpt-tokenizer locally. The status bar tracks both the hard context window and a practical limit (12K to 30K) above which output quality tends to degrade. Claude and Gemini counts are approximations.

Browse SQLite files in VS Code

Right-click any .sqlite or .db file in the Explorer and choose Open in SQLite Viewer. Search across all non-binary columns, sort, apply per-column filters, and copy cell values. BLOB columns show byte size only. The viewer is read-only.

No API keys. No telemetry.

Token counting runs locally with no network calls. File content is read only to count tokens and build the clipboard prompt -- never stored, logged, or transmitted. No analytics of any kind. Safe for proprietary codebases and air-gapped environments.