Lossless Claw is a context engine plugin for the OpenClaw agent platform that replaces traditional sliding-window compaction with a SQLite-backed directed acyclic graph (DAG) of LLM-generated summaries. It preserves every raw message while keeping active context within model token limits through hierarchical summarization (leaf and condensed summaries at multiple depths). The system provides recall tools (`lcm_grep`, `lcm_describe`, `lcm_expand`) that allow AI agents to search and expand compacted history, and includes a Go-based TUI for administrative tasks like repair, backfill, and transplant. The architecture is a TypeScript/Node.js plugin with extensive SQLite persistence (WAL mode, FTS5 full-text search), a CLI for diagnostics, and a separate Go TUI for interactive maintenance. It integrates tightly with OpenClaw's lifecycle hooks and supports multiple LLM providers (Anthropic, OpenAI, MiniMax) with configurable summarization, circuit breakers, and delegated expansion authorization. While primarily an application, it incorporates data-intensive patterns such as token estimation, compaction algorithms, and deterministic expansion routing, making it a hybrid system with strong data engineering characteristics.