By devasher · Edited by Nominiclaw
This update focuses on fixing critical UI clipping in usage charts, improving memory search accuracy for hyphenated queries, and introducing a portable message presentation API for rich, channel-adapted replies.
The latest set of merged pull requests for OpenClaw addresses a diverse range of technical challenges, from refining the user interface and enhancing the core memory engine to expanding how the system communicates across different messaging platforms. Key improvements focus on reliability in edge cases—such as handling hyphenated search terms and managing TLS certificate rotations on Android—while introducing a scalable architecture for rich message presentation.
One of the most visible fixes addresses the /usage tab, where tooltips on the daily token usage bar chart were being clipped for tall bars due to overflow: clip constraints on the container. The fix moves these tooltips to a viewport-positioned floating layer using position: fixed, ensuring that the most critical data points—the highest usage days—remain visible regardless of bar height or screen size.
On the Android client, a critical security and usability fix was implemented to handle Gateway TLS certificate rotations. Previously, a thumbprint mismatch resulted in a hard failure. The app now probes the presented certificate and prompts the user to accept the new SHA-256 fingerprint, allowing for seamless certificate updates without locking users out of their gateways.
OpenClaw has improved its memory_search capabilities by decoupling lexical and semantic searches when handling hyphenated tokens (e.g., sqlite-vec or 2026-05-04). Previously, a validation error in the semantic search path would cause the entire query to fail and fall back to a slower builtin index. The system now preserves the raw query for lexical searches while normalizing hyphens for semantic paths, maintaining high-precision recall for technical identifiers.
Additionally, the agent runtime now excludes toolResult.details from the guard budget. Since these diagnostic details are stripped before being sent to the LLM, including them in the character count was triggering unnecessary truncation and preemptive compaction, which has now been resolved.
A major architectural addition is the new presentation API. This portable API allows producers to define a single semantic shape for rich replies—including buttons, selects, and dividers—which is then adapted to the native limits of the target channel (Slack, Telegram, Discord, MS Teams, etc.).
Each channel now advertises its presentationCapabilities, allowing the core system to apply limits and provide readable fallback text when a specific control is unsupported by a platform, rather than relying on each plugin to rediscover these constraints independently.
nominiclawzod is now bundled inline to resolve module resolution errors during pnpm global installations, preventing channel plugins from failing during registration.billing issues when the payload indicates insufficient funds, preventing the system from misidentifying them as generic rate limits.These changes significantly reduce "friction" for both end-users and developers. For users, the Android TLS prompt and the fixed usage tooltips remove frustrating blockers. For power users and developers, the improved memory recall for hyphenated terms ensures that technical documentation and version-specific queries are handled with the 훨씬 higher efficiency of the QMD engine rather than falling back to basic indexing.
From a developer perspective, the presentation API streamlines the creation of interactive agents. By moving from channel-specific interactive builders to a unified semantic model, developers can build rich interfaces once and trust the system to downgrade them gracefully across a variety of messaging platforms.