OpenClaw v2026.4.15 + Operator Mode

Two things shipping today. First: OpenClaw v2026.4.15 — the biggest patch release yet, with dreaming fixes, security hardening, and 60+ contributor patches. Second: Operator Mode— a guided onboarding layer that makes it easy for new users to deploy their first agent in under two minutes, without touching any of the advanced controls.

OpenClaw v2026.4.15

This release has the highest contributor count we've seen. Highlights:

• Dreaming storage mode change — default dreaming storage moved from inline to separate, so dream phase blocks land in memory/dreaming/ instead of bloating daily memory files. This fixes OOM crashes in constrained containers
• Dreaming session ingestion fix — narrative transcripts from dreaming are now skipped from session-store metadata before bootstrap records land, preventing dream prose from polluting session ingestion
• Anthropic Claude Opus 4.7 — default model selections, opus aliases, CLI defaults, and bundled image understanding
• Gemini TTS — text-to-speech support added to the bundled Google plugin, including voice selection and WAV/PCM output
• Model Auth status card — new Control UI card showing OAuth token health and provider rate-limit pressure at a glance
• LanceDB cloud storage — durable memory indexes can now run on remote object storage instead of local disk only
• GitHub Copilot embeddings — dedicated embedding provider for memory search with token refresh and safer payload validation
• Local model lean mode — agents.defaults.experimental.localModelLean: true drops heavyweight tools for weaker local setups

Security Fixes

• Gateway tool trust anchor — trusted local MEDIA tool-result passthrough now anchored on exact raw names of registered built-in tools. Client tool definitions that normalize-collide with a built-in are rejected with 400
• Secrets in exec approvals — inline approval review no longer leaks credential material in rendered prompt content
• MCP loopback auth — bearer comparison switched to constant-time safeEqualSecret, browser-origin requests rejected before auth gate
• Feishu webhook hardening — fail-closed on missing encryptKey and blank callback tokens
• Workspace file safety — agents.files.get/set routed through shared fs-safe helpers, symlink aliases rejected

Platform Stability

• WhatsApp reconnect fix — pending creds save queue drained before reopening sockets
• BlueBubbles catchup — persistent per-account cursor replays missed webhook messages after gateway restart
• Ollama model IDs — provider prefix stripped from chat request model IDs so configured refs stop 404ing
• Skills snapshot invalidation — cached snapshot version bumped on config writes to skills.*, fixing "Tool not found" loops
• Unknown tool guard — enabled by default, prevents hallucinated tools from looping until timeout
• TUI streaming watchdog — resets to idle after 30s of delta silence, guards against lost state events

Full changelog: v2026.4.15 on GitHub. All existing agent containers will pick up the new version on next deploy.

Operator Mode

Agentbot has always been built for power users. 50+ routes, 6 navigation sections, deep runtime controls, workflow editors, wallet management, skill configuration — that's the core product and it's not going anywhere.

But we kept hearing the same thing from new users: "I just want to get an agent running. Where do I start?"

Operator Modeis our answer. It's a thin, feature-flagged entry layer that sits on top of the existing system. It doesn't replace anything — it simplifies the first experience.

What's New

• /app/start — Guided onboarding: pick a template, name your agent, launch in 3 clicks
• /app/templates — Starter templates: Music Promoter, Community Manager, Content Creator, Crypto Analyst, DJ Radio Host, Event Scout. Each creates a normal Workflow + Agent that's fully editable from Advanced Mode
• /app/activity — Activity feed: see everything happening with your agents in one place
• /app/tutorials — Step-by-step tutorials: deploy first agent, connect channels, explore skills, set up wallet, build workflows
• /app/advanced — One-click switch to the full Advanced Mode dashboard

What Didn't Change

Everything. The entire advanced experience is untouched:

• All 50+ existing routes work exactly as before
• No navigation items removed or reordered
• No runtime or OpenClaw execution behaviour changed
• No existing API response shapes modified
• No existing users forced into Operator Mode

How Routing Works

New users (no agents, no workflows) see the guided experience by default. Existing users stay on Advanced Mode. Anyone can switch between modes at any time — the preference is saved and persists.

Feature Flags

Operator Mode is behind three Vercel environment flags:

• OPERATOR_MODE_ENABLED — master switch for guided routes
• NEW_USER_OPERATOR_DEFAULT — route first-time users to /app/start
• SHOW_OPENCLAW_BADGE_IN_OPERATOR — subtle "Powered by OpenClaw" in beginner mode

Currently enabled in preview only. Production rollout will happen after internal testing.

Templates Are Real Workflows

Every template launched from Operator Mode creates a real Workflow and Agent in the database — the same models used by Advanced Mode. There's no separate execution system. You can launch a Music Promoter from /app/templates, then open it in /dashboard/workflows and edit every node. Templates are presets, not cages.

What's Next

• More templates based on user feedback
• Tutorial system connected to achievement badges
• Operator Mode analytics to understand where new users get stuck
• Production rollout of Operator Mode after preview validation

Questions? Feedback? Join us on Telegram or open an issue on GitHub.