Context engineering

Feeding an LLM raw documents wastes tokens on duplication and noise, and gives no control over what actually reaches the model. A context pack is a curated, versioned bundle: pick whole documents or specific sections, add task instructions so the pack's purpose is explicit, and set a token budget.

Fitting the budget

When a pack exceeds its token budget, ContextBoard applies, in order:

1. Deduplication. Identical content repeated across documents is kept once.
2. Compression. If enabled, an LLM condenses the assembled content to fit, preserving facts and figures.
3. Truncation. If still over budget, the lowest-priority items are dropped until the pack fits.

Real token counts come from the model's own tokenizer, not an approximation -- a budget you can trust.

Export anywhere

Every pack exports in four formats: Markdown for human review, JSON for custom pipelines, OpenAI-style chat messages, or RAG chunks for a vector store.

Versioned like code

Editing a pack's configuration creates a new version rather than overwriting the last one -- previous exports stay reproducible.