taskfold

Long AI coding sessions drown in their own history. Every request re-sends hours of finished tool outputs, debug logs, and abandoned attempts; costs climb, the model gets distracted, and the context window eventually overflows. taskfold takes a notebook approach: name each unit of work (task_begin / task_end), and when the task closes, the entire back-and-forth is folded into one titled summary, keeping the conversation readable and every subsequent request cheaper.

Summaries deliberately preserve the decisions that matter, the feedback given, why things failed, what changed, and the final outcome — not a summary of a summary. Nothing is truly lost: every fold archives the exact original messages to a file, and fold_recall brings them back in full at any moment. With list_folds for an at-a-glance index, a gentle reminder that nags only when discipline slips, and a design that rewrites only a middle chunk of history to stay cache-friendly, taskfold cuts token cost without sacrificing traceability.

Built for developers and teams who run long, multi-step coding sessions with an AI agent — debugging, refactoring, and iterating on complex codebases — and who want the conversation to stay lean and the cost curve flat.