OpenClaw 4.15: What's New — Opus 4.7 Default, Gemini TTS, Cloud Memory & More

April 21, 2026 · ClawReady Team

OpenClaw 2026.4.15 dropped 4 days ago (April 17) as the latest stable npm release. It's a feature-dense update that touches models, audio, memory, and local setups. Here's what's actually worth your attention — and one big gotcha to be aware of before upgrading.

⚠️ Heads-up before you upgrade

4.15 has a confirmed Active Memory regression (#68825) — the plugin causes every blocking sub-agent call to hang at 120 seconds. If you rely on Active Memory for context injection, hold at 4.14 until a patch lands. See our Active Memory timeout fix guide for details and workarounds.

What's new in 4.15

1. Claude Opus 4.7 is now the Anthropic default

OpenClaw has updated its bundled Anthropic default selections to Claude Opus 4.7. This cascades to the opus alias, Claude CLI defaults, and bundled image understanding. If you've been using the opus shorthand or relying on the built-in Anthropic preset, you'll automatically get 4.7 on fresh installs and new agents.

Impact: Existing agents with model explicitly set in openclaw.json are unaffected. New agents and default-model fallbacks will use 4.7.

2. Gemini TTS in the bundled Google plugin

The bundled Google plugin now supports Gemini text-to-speech — your agent can synthesize and reply with audio over supported channels. Output formats: WAV (standard replies) and PCM (telephony output). Voice selection is configurable.

This is the first native TTS integration in a bundled OpenClaw plugin. If you've been running voice assistants via OpenClaw + custom TTS pipelines, this simplifies the stack significantly.

Setup note: Requires Google plugin to be enabled and a valid Google API key with TTS permissions. PR #67515.

3. Model Auth status card in the Control UI

A new Model Auth status card appears in the Overview panel showing OAuth token health and provider rate-limit pressure at a glance. Attention callouts fire when tokens are expiring or already expired. Credentials are stripped before display; results cache for 60 seconds.

If you've ever had a silent auth failure that looked like a model being slow — this finally surfaces it. PR #66211.

4. Cloud-backed LanceDB memory

The memory-lancedb plugin now supports remote object storage (S3-compatible) as a backend for durable memory indexes. Previously, LanceDB memory was local-disk only — which made it unavailable for cloud-hosted or Docker deployments without persistent volumes.

Who this helps: Anyone running OpenClaw in Docker, Fly.io, Railway, or any ephemeral compute environment. Your memory index persists across container restarts and redeploys. PR #63502.

5. GitHub Copilot embedding provider

OpenClaw now ships a GitHub Copilot embedding provider for memory search. If your team already has Copilot licenses, you can use them for vector embeddings instead of paying for a separate OpenAI embeddings bill. Includes a Copilot embedding host helper for plugins, token refresh, and safer payload validation. PR #61718.

6. localModelLean — lighter footprint for weaker models

New experimental config: agents.defaults.experimental.localModelLean: true. When enabled, this drops heavyweight default tools (browser, cron, message) from the system prompt, reducing prompt size for local model setups like Ollama.

Why it matters: Local models (Qwen, Llama, Mistral) struggle with large tool-list prompts. This flag gives them a fighting chance without requiring manual tool config per-agent. PR #66495.

Enable in openclaw.json:

"agents": {
  "defaults": {
    "experimental": {
      "localModelLean": true
    }
  }
}

7. Leaner plugin packaging

Bundled plugin runtime dependencies are now localized to their owning extensions, and the published docs payload has been trimmed. The result: smaller installs, faster npm downloads, and less core bloat carrying extension-owned runtime baggage. PR #67099.

Summary: should you upgrade?

Need help upgrading or staying on a stable version?

ClawReady handles safe upgrades, config migration, and Active Memory validation so you know exactly what you're running. See our setup options.