Real-time Voice Chat Launched on Android! Open Source AI OpenClaw Releases New Version, Fully Unleashing GPT-5

According to Sentinel Beating monitoring, the open-source personal AI assistant service OpenClaw has released a new version v2026.5.18. This update switches the Android client's conversation mode to real-time voice chat based on gateway relays, and fully enables configuration validation and support for GPT-5 series models.

The new Android client implements streaming microphone input and real-time audio playback, and integrates tool-result bridging (which synchronizes tool invocation status with real-time audio stream) and real-time screen captions. This allows mobile users to directly activate and run a complex local toolchain through voice commands.

In terms of model support, OpenClaw has lifted the interception on GPT-5.1, GPT-5.2, GPT-5.3, and openai-codex models during the configuration validation phase. The new version discontinues the mandatory truncation logic for GPT-5 final responses to preserve the complete channel response and automatically log when in strict-agentic execution mode.

To simplify plugin extension, the new version introduces the defineToolPlugin minimal plugin interface and comes with command-line tools for openclaw plugins build, validate, and init. Developers can now declare simple tool plugins in a strongly typed manner, and the system will automatically generate the required manifest and context factory.

In terms of underlying performance, the Memory-core has introduced a startup incremental sync mechanism. The system now compares disk sessions and index files at startup, only performing incremental index updates for missing, changed, or resized files, significantly reducing cold start time.

Furthermore, the unmanaged signal configuration reloads, such as SIGUSR1, are now fully handled within the process, avoiding the issue of the parent daemon losing the process identifier (PID) due to spawning orphaned subprocesses.