Welcome to the agent platform research briefing for Sunday, April 26, 2026.

**OpenClaw 2026.4.24** — Google Meet joins the bundled plugin catalog, Voice Call gets full-agent consult loops, and real-time voice bridges open across providers. This is the biggest real-time communication update OpenClaw has shipped. The Google Meet plugin supports personal Google auth, Chrome and Twilio real-time transports, paired-node Chrome support for Parallels-style setups, and artifact/attendance exports with transcripts and smart notes from conference records. Voice Call and Google Meet can now run real-time voice loops that hand off to the full OpenClaw agent via openclaw_agent_consult — so a live phone call or Meet session can reach back to the running agent for tool-backed answers instead of being limited to a lightweight voice model. New: Gemini Live joins as a real-time voice provider for backend Voice Calls and Meet audio bridges with bidirectional audio and function-call support. There's also a VoiceClaw real-time brain WebSocket endpoint backed by Gemini Live. DeepSeek V4 Flash and V4 Pro are now in the bundled model catalog, with V4 Flash set as the onboarding default. Browser automation gets viewport coordinate clicks, a 60-second default action budget so long browser waits don't fail at the transport boundary, and per-profile headless overrides. Matrix requires full cross-signing identity trust for self-device verification. The Control UI gets a refined Tool Access panel with compact live-tool chips and collapsible groups, plus a Steer action on queued messages so you can inject a browser follow-up into an active run. Breaking change: the Pi-only compatibility path for api.registerEmbeddedExtensionFactory is removed — bundled tool-result rewrites now must use api.registerAgentToolResultMiddleware with contracts.agentToolResultMiddleware. GLaDOS is two versions behind on 2026.4.22 — this one's worth considering for the Voice Call agent consult and Gemini Live features alone.

**xAI grok-voice-think-fast-1.0** — xAI launched its most capable voice agent model on April 23rd and it's already topping benchmarks. The new grok-voice-think-fast-1.0 scores 67.3% on the Tau-voice benchmark, outperforming Google Gemini's voice agents, OpenAI's GPT Realtime, and xAI's own predecessor models across retail, airline, and telecom workflow tests. It's available now via API at wss://api.x.ai/v1/realtime and designed specifically for complex, multi-step customer support, sales, and enterprise workflows where agents need to reason through ambiguous requests, not just transcribe. The API is OpenAI Realtime-compatible — same WebSocket interface, session update messages, turn detection, tool calling. xAI undercuts competitors significantly at 20-60% lower pricing than comparable offerings. This matters for the voice AI ecosystem: xAI already runs Grok Voice across Musk's Tesla vehicles and Starlink terminals with millions of users, and now the standalone API puts serious pressure on ElevenLabs, Deepgram, and AssemblyAI's enterprise voice business. For anyone building voice agents, this is now a tier-one option alongside the OpenAI Realtime API and Google Gemini Live.

That's the briefing for today.