dejadb-context

Budget-aware context rendering (SML/TOON/Markdown/JSON) for DejaDB.

dejadb-context turns recall results into model-ready context. It takes the search hits produced by the store and renders them into budget-aware, provider-optimal strings — SML, Markdown, TOON, or JSON — while respecting a token budget through progressive disclosure (full, summary, or omit per grain) and grain-type sectioning. This is the last hop in the recall path: it decides what fits in the context window and how it is formatted for the target model, so an agent receives compact, relevant memory rather than raw rows.

Part of DejaDB — an embedded memory engine for AI agents. See the architecture overview.

Licensed under MIT OR Apache-2.0.