OpenClaw's Lossless-Claw Plugin Redefines Agent Memory with Pluggable Context Engine
What Happened
OpenClaw's beta version 2026.3.7 introduced a pluggable context engine that fundamentally restructures how the platform handles conversation memory. The flagship implementation, called lossless-claw, stores conversations in SQLite databases and uses LLM-generated abstracts of old message blocks organized in a directed acyclic graph structure. Developers can search conversation history through dedicated tools — lcm_grep, lcm_describe, and lcm_expand — rather than relying on the context window to retain everything.
According to 36Kr's reporting, the new architecture replaces OpenClaw's previously hard-coded context management with a plugin-based system. The design philosophy, articulated by the PR author, is that "you don't actually need an Agent memory system — what you need is a context that won't be reset." In the OOLONG benchmark suite, lossless-claw scored 74.8, outperforming Claude Code's 70.3, with the performance gap widening as context length increases.
Why It Matters
Context loss has been one of OpenClaw's most persistent user complaints, especially for long-running agent sessions that handle complex multi-step tasks. The pluggable architecture is significant because it doesn't just fix the problem — it opens it up to the community. Third-party developers can now build alternative context engines optimized for different use cases, whether that's customer support conversations, coding sessions, or research workflows. The fact that lossless-claw already outperforms established commercial alternatives on standard benchmarks suggests the approach is sound.
What's Next
The pluggable context engine also ships with support for GPT-5.4 and Gemini Flash 3.1 as summarization backends, plus persistent agent-to-channel binding and topic-based agent routing for Telegram and Discord. Expect a wave of community-built context plugins as developers experiment with different storage backends, summarization strategies, and retrieval approaches. The iOS App Store release is also reportedly in preparation, which would bring these memory improvements to mobile users.