OpenClaw 2026.4.15: Gemini TTS, Model Auth Dashboard & Cloud Memory

Headline Features

Gemini Text-to-Speech

The bundled Google plugin now supports Gemini text-to-speech. This includes provider registration, voice selection, WAV reply output, and PCM telephony output. You can now use Google's high-quality voices for agent responses directly through OpenClaw.

// Example: Gemini TTS configuration
{
  "plugins": {
    "entries": {
      "google": {
        "config": {
          "tts": {
            "provider": "gemini",
            "voice": "en-US-Standard-C",
            "outputFormat": "wav"
          }
        }
      }
    }
  }
}

Model Auth Status Dashboard

The Control UI Overview now includes a Model Auth status card showing OAuth token health and provider rate-limit pressure at a glance. Get attention callouts when OAuth tokens are expiring or expired, backed by a new models.authStatus gateway method that strips credentials and caches for 60s.

GitHub Copilot Embeddings

Memory search now supports GitHub Copilot as an embedding provider. A dedicated Copilot embedding host helper is exposed so plugins can reuse the transport while honoring remote overrides, token refresh, and safer payload validation.

// Enable Copilot embeddings
{
  "memory": {
    "embeddingProvider": "github-copilot"
  }
}

LanceDB Cloud Storage

The memory-lancedb plugin now supports cloud storage, so durable memory indexes can run on remote object storage instead of local disk only. This enables scalable memory deployments without managing local storage capacity.

Anthropic Model Updates

Default Anthropic selections have been updated, including opus aliases, Claude CLI defaults, and bundled image understanding now defaults to Claude Opus 4.7.

Context & Memory Improvements

Smarter Context Trimming

Default startup/skills prompt budgets are now trimmed more aggressively, memory_get excerpts are capped by default with explicit continuation metadata, and QMD reads follow the same bounded excerpt contract. Long sessions now pull less context by default without losing deterministic follow-up reads.

Dreaming Storage Separation

The default dreaming.storage.mode has changed from inline to separate. Dreaming phase blocks (## Light Sleep, ## REM Sleep) now land in memory/dreaming/{phase}/YYYY-MM-DD.md instead of being injected into memory/YYYY-MM-DD.md. Daily memory files no longer get dominated by structured candidate output.

Dreaming Transcript Cleanup

Dreaming narrative transcripts are now skipped from session-store metadata before bootstrap records land, so dream diary prompt/prose lines do not pollute session ingestion.

Security Fixes

Tool Name Collision Protection

Client tool definitions whose names normalize-collide with a built-in or with another client tool in the same request are now rejected with a 400 invalid_request_error. A client-supplied tool named like a built-in can no longer inherit its local-media trust.

WebChat Audio Security

LocalRoots containment is now enforced on webchat audio embedding path, preventing potential path traversal issues.

Matrix DM Pairing Protection

DM pairing-store entries are now blocked from authorizing room control commands, keeping the room path narrower without changing room auth behavior.

Dangerous Config Protection

The gateway tool now rejects config.patch and config.apply calls when they would newly enable flags like dangerouslyDisableDeviceAuth, allowInsecureAuth, or tools.exec.applyPatch.workspaceOnly: false. Already-enabled flags pass through unchanged.

Channel & Integration Fixes

Discord Tool Call Display

Standalone Gemma-style ... tool-call payloads are now stripped from visible assistant text without truncating prose examples or trailing replies.

WhatsApp Web Session Reliability

The pending per-auth creds save queue is now drained before reopening sockets, preventing race conditions between in-flight creds.json writes and reconnect-time auth bootstrap.

BlueBubbles Catchup Retry

A per-message retry ceiling (catchup.maxFailureRetries, default 10) prevents persistently-failing messages from blocking the entire catchup process.

Cron Announce Delivery

Mixed-content isolated cron announce replies that end with NO_REPLY no longer leak summary text to the target channel.

Developer Experience

Local Models Lean Mode

Experimental agents.defaults.experimental.localModelLean: true drops heavyweight default tools like browser, cron, and message, reducing prompt size for weaker local-model setups without changing the normal path.

Packaging Improvements

Bundled plugin runtime dependencies are now localized to their owning extensions, the published docs payload is trimmed, and install/package-manager guardrails are tightened so published builds stay leaner.

CLI Transcript Persistence

Successful CLI-backed turns now persist into the OpenClaw session transcript, so google-gemini-cli replies appear in session history and the Control UI.

Upgrade Guide

Immediate Upgrade Recommended If:

• You use Google Gemini TTS or want to try it
• You use GitHub Copilot for embeddings
• You run LanceDB with cloud storage
• You use dreaming/background processing
• Long-running session context management has been an issue

Upgrade Commands

# Via CLI
openclaw update

# Via agent
"Update OpenClaw to version 2026.4.15"

# Verify installation
openclaw doctor

Post-Upgrade Checklist

1. Check the new Model Auth status card in Control UI Overview
2. Test Gemini TTS if you use Google voice features
3. Verify dreaming files are now stored separately (check memory/dreaming/)
4. If using Copilot, test memory search functionality

Release date: April 15, 2026 · 25+ fixes
Full changelog: github.com/openclaw/openclaw/releases/tag/v2026.4.15