realvco Docs

v2026.4.24 — Google Meet Joins + DeepSeek V4 + Talk Realtime Voice

Released: April 24, 2026 Theme: Channel / model / realtime voice expansion across three lines Breaking Changes: api.registerEmbeddedExtensionFactory(...) Pi-only compat path removed (plugins must use api.registerAgentToolResultMiddleware(...))

Core Highlights

1. Google Meet Joins OpenClaw

Google Meet is now a bundled participant plugin:

  • Personal Google auth
  • Explicit meeting URL joins
  • Chrome and Twilio realtime audio transports
  • Paired-node chrome-node support (Parallels-style Chrome/BlackHole/SoX hosts)
  • Full OpenClaw agent consult inside live voice sessions
  • Artifact and attendance export workflows (conference records, recordings, transcripts, smart notes, participant sessions)
  • Recovery for already-open Meet tabs: recover_current_tab, recover-tab — no duplicate tabs

2. DeepSeek V4 Joins

DeepSeek V4 Flash and V4 Pro enter the bundled catalog:

  • V4 Flash is the onboarding default
  • Thinking / replay behavior fixed for follow-up tool-call turns

3. Talk Mode Realtime Voice + Full-Agent Consult

Talk, Voice Call, and Google Meet can use realtime voice loops with full-agent consult for deeper tool-backed answers:

  • Browser WebRTC realtime voice sessions backed by OpenAI Realtime
  • Gateway-minted ephemeral client secrets
  • openclaw_agent_consult hands hard questions to the full agent

No longer limited to realtime voice model’s own tool capabilities.

4. Voice Call & Gemini Live

  • New Gemini Live realtime voice provider (for Voice Call and Google Meet audio bridges)
  • Bidirectional audio + function-call support
  • New voicecall setup and dry-run-default voicecall smoke commands check Twilio / provider readiness before placing a live test call

5. Browser Automation Upgrades

  • Viewport coordinate clicks (no longer just a11y refs): openclaw browser click-coords
  • Default action budget bumped to 60s (browser.actionTimeoutMs configurable)
  • Per-profile browser.profiles.<name>.headless overrides (one profile headless without forcing all)
  • Steadier tab reuse / recovery

6. Control UI Tool Access Redesigned

  • Compact live-tool chips
  • Collapsible tool groups
  • Direct per-tool toggles
  • Clearer runtime / source provenance
  • New Steer action on queued chat messages — inject follow-up into active run without retyping

7. Model / Plugin Startup Speedup

  • Bundled providers use safe static catalogs, models list faster
  • Manifest-sourced model rows, duplicate provider/model conflict reporting, shared src/model-catalog normalization
  • node-llama-cpp no longer installed by default (local embeddings load it only when operators install the optional runtime package)
  • Sharply reduced broad manifest scans on startup

8. Codex Harness Push

  • Codex app-server runs context-engine bootstrap, assembly, post-turn maintenance, engine-owned compaction
  • Legacy Codex auth-provider aliases accepted in app-server profile login / refresh
  • Codex-native tool hooks bridged into OpenClaw plugin hooks and approvals
  • Provider-owned transport / auth / follow-up seams so Codex-style runtimes participate in fallback policy

9. Gradium TTS Added

New Gradium bundled TTS provider with voice-note and telephony output support.

10. Memory Hybrid Search Transparency

memory-core hybrid search results expose raw vectorScore and textScore alongside the combined score:

  • Callers can inspect vector vs text retrieval contribution
  • Before temporal decay or MMR reordering

11. Other Improvements

  • Matrix self-device verification requires full cross-signing identity trust; new openclaw matrix verify self CLI
  • WebChat sessions keep runtime-only prompt context out of visible transcript history
  • agents.defaults.contextInjection: "never" disables workspace bootstrap injection for agents that own their prompt lifecycle
  • Plugin manifest gains modelCatalog contract (provider-owned model rows, aliases, suppression rules, discovery mode metadata)

Upgrade Recommendations

  • Customers wanting Google Meet automation: upgrade and use
  • DeepSeek users: V4 Flash / V4 Pro available
  • Talk Mode users: realtime voice + full-agent consult is a big upgrade
  • Customer-service voice-call automation: Voice Call upgrades worth trying
  • Customers running many plugins / providers: faster startup
  • Plugin developers rewriting tool-results: note the Breaking — switch to the new API

← v2026.4.23 · v2026.4.25 → · Back to list