Sign in Subscribe
3 min read

🦞 OpenClaw’s Claude Era Ended

Plus: GPT-5.4 Guideline, 4.5 Just Dropped, “Clawchief” Starter Kit

Good Morning, OpenClaw Owners!

Anthropic slammed the door on OpenClaw users overnight, cutting off the cheapest path to Claude Opus 4.6 and forcing thousands of agent workflows to find a new way forward.

OpenClaw’s Claude Era Ended

TL;DR: Anthropic has cutted off Claude subscription access for third-party tools like OpenClaw triggered a major backlash on X, where many users described it as a sudden shutdown of workflows they relied on daily. A large number of OpenClaw power users effectively lost access to Opus 4.6 overnight and were forced to switch to other models, exposing how fragile it is to build serious agent systems on a single provider’s subscription policy.

Locked Out of Opus 4.6? Switch to MyClaw

TL;DR: While Anthropic’s OpenClaw crackdown threw many power users off Claude Opus 4.6 overnight, MyClaw keeps the major frontier models available with stable access and discounted pricing. If you can’t keep getting near-free Opus 4.6 through OpenClaw anymore, you can still use Opus 4.6 on MyClaw at 10% below official pricing, and the same discount applies to other major models too.

Peter Tried Saving Claude

TL;DR: Peter Steinberger briefly restored Claude subscription access using a first-party harness workaround after a required update. It worked for a short time before Anthropic shut it down by directly targeting OpenClaw with system prompt string matching. Expect more attempts to route subscriptions through similar workarounds will likely appear, but each will only be temporary.

OpenClaw GPT-5.4 Guideline

TL;DR: A developer explained how to migrate OpenClaw after Anthropic stopped covering Claude subscriptions for third-party tools, pushing users toward GPT-5.4. The guide shows steps by steps that GPT needs prompt tweaks to act proactively—like calling tools first and skipping confirmations—while highlighting GPT’s stability for execution and Claude’s strength in creativity, requiring users to adjust their presets accordingly.

OpenClaw 4.5 Just Dropped

TL;DR: OpenClaw v2026.4.5 adds built-in multimodal generation (image, music, video), introduces the “Dreaming” memory system, and improves task execution with structured progress and better caching, alongside broader language support.

Highlights from this release:

  • 🎬 Built-in video + music generation
  • 🧠 /dreaming is now real — experimental memory consolidation (light / deep / REM-style phases)
  • 🔀 Structured task progress
  • ⚡ Better prompt-cache reuse
  • 🌍 Control UI + Docs now support 12 more languages

Agents are evolving with multimodal generation and persistent memory.

OpenClaw “Clawchief” Starter Kit

1

TL;DR: A developer released “clawchief,” an open source starter kit on GitHub that shows how to turn OpenClaw into a proactive executive assistant. The repo bundles structured workflows, markdown-based task systems, skills, memory, and cron jobs into a reusable setup, enabling OpenClaw to continuously manage emails, scheduling, and operations.

OpenClaw Docs Go Multilingual

TL;DR: OpenClaw rolled out documentation in 12 languages by moving translations into a dedicated repo with glossaries and translation memory, ensuring consistency and only updating pages when the English source changes. Peter added that GPT-5.4 now auto-translates docs via GitHub triggers, replacing Google Translate with higher-quality daily updates.

OpenClaw Tests Itself

TL;DR: Peter shared an upcoming QA system where OpenClaw agents test other OpenClaw agents through a synthetic message channel. An orchestrator assigns tasks, verifies outcomes using LLM judgment, and, if failures occur, spawns subagents to analyze and fix issues. The setup replaces manual testing loops with automated, self-correcting workflows built entirely on agents.

Run OpenClaw on MyClaw.ai - No setup, No tech skills, Always-on 👇