OpenClaw has become one of the most watched open source AI projects because it solves a real problem. Most people obsess over the model. OpenClaw focuses on the agent harness around the model. That means memory, tools, sessions, channels, execution, and control. In simple words, the model is the brain, but the harness is what helps it do useful work in the real world.

That is a big reason people started liking it.

If you haven’t heard of it, this one’s for you. If you have heard of it but haven’t had time to catch up this one’s especially for you.

One Topic: 8 Major OpenClaw Updates Incl. Dreaming, Memory, Video, Music and Everything In Between

Let me give you some context first, because the backstory is almost as interesting as the product.

A Lobster 🦞 That Broke the Internet

OpenClaw was first published in November 2025 under the name Clawdbot by Austrian developer Peter Steinberger, who also founded the software company PSPDFKit. Within two months, it was renamed twice first to “Moltbot” following trademark complaints, and then three days later to “OpenClaw” because, as Steinberger put it, the name Moltbot “never quite rolled off the tongue.”

The real explosion came in early 2026. OpenClaw went from 9,000 to over 60,000 GitHub stars in just 72 hours. Developers were calling it a 24/7 JARVIS a personal AI that actually does things, not just talks about them.

Today, the OpenClaw repository sits at over 359K GitHub stars with more than 73K forks, a number that puts it ahead of Meta’s React JavaScript library, which took nearly a decade to reach a similar milestone. In February 2026, Steinberger announced he would be joining OpenAI, and that a non-profit OpenClaw foundation would be established to provide future stewardship of the project. The community didn’t slow down, if anything, it accelerated.

So What Actually IS OpenClaw?

Most people make the mistake of thinking OpenClaw is another AI chatbot. It’s not.

Think of it this way. Your car has an engine. But you can’t drive an engine by itself, you need the body, the steering wheel, the brakes, the fuel tank, and the dashboard before it becomes useful. AI models like Claude, GPT, or Gemini are the engine. They’re powerful, but raw.

OpenClaw is a free, open-source agent that runs locally and connects large language models to real software. You can give it simple chat commands, and it can read and write files, run shell commands, browse websites, send emails, control APIs, and automate tasks across different applications, it will actually carry out the steps, not just explain how to do them.

This structure is called an agent harness. It’s the scaffolding that turns a language model into something that can work on your actual life, your calendar, your emails, your code, your business.

Many describe it as “self-improving” because it can enhance its own capabilities by autonomously writing code to create new skills for tasks you want it to execute. It operates a local gateway that connects AI models with your favorite tools, integrating with familiar chat apps to facilitate convenient interactions.

Here’s the part that makes it different from everything else: OpenClaw bots run locally and are designed to integrate with an external large language model such as Claude, DeepSeek, or one of OpenAI’s GPT models. Configuration data and interaction history are stored locally, enabling persistent and adaptive behavior across sessions.

Your data. Your hardware. Your control.

The Last 10 Days Changed Everything

Here’s where it gets really interesting.

Between roughly April 5 and April 17, 2026, OpenClaw shipped a wave of releases 2026.4.5 through 2026.4.15 at a pace that averaged a new version every day or two. Most releases weren’t small patches. They were proper feature drops.

Here are the 8 that matter most to you.

1. Your Agent Now Sleeps And Gets Smarter While It Does

This is the headline feature. Dreaming is OpenClaw’s automatic three-phase background process that turns short-term memory signals into durable long-term knowledge. It runs in three stages: Light Sleep (ingest and stage), REM Sleep(reflect and extract patterns), and Deep Sleep (promote to MEMORY.md). Only entries that pass all three threshold gates get promoted.

Dreaming is disabled by default and requires an explicit opt-in. Nothing in your MEMORY.md will change after upgrading unless you enable it in your config file and a Deep phase runs.

In plain terms: your agent reads through everything you’ve done together, decides what actually matters long-term, and stores only that. Just like how you wake up from sleep having absorbed the important things and let go of the noise.

Six weighted signals score every candidate: Relevance (0.30), Frequency (0.24), Query diversity (0.15), Recency (0.15), Consolidation (0.10), and Conceptual richness (0.06). The result is an agent that gets genuinely smarter about you over time.

2. Active Memory: No More “Remember This”

The Active Memory plugin gives OpenClaw a dedicated memory sub-agent right before the main reply, so ongoing chats can automatically pull in relevant preferences, context, and past details without making users remember to manually say “remember this” or “search memory” first. It includes configurable message/recent/full context modes, live verbose inspection, and advanced prompt and thinking overrides for tuning.

Before this, your agent was a great worker who forgot everything between shifts. Now it’s a great worker who actually reads their notes before showing up.

3. Your Agent Can Now Generate Video

Recent release adds the built-in video generation tool, enabling AI agents to create videos through configured providers and return generated media directly in responses. This capability opens doors for automated video content creation workflows.

Supported providers include xAI Grok Imagine Video, Runway, Alibaba, and local ComfyUI. You describe the video you want in natural language, OpenClaw handles the rest, and the output lands in your chat.

4. Music Generation Is Now Native

Music generation in OpenClaw 4.5 follows the same unified interface pattern as video. Two backends ship at launch Google DeepMind’s Lyria model, which produces high-quality instrumental and vocal tracks from text descriptions, and MiniMax, which handles vocal generation well and costs less.

Lyria supports genre specification, mood parameters, tempo control, and instrumentation preferences. You can describe “upbeat electronic track, 120 BPM, synth-heavy, suitable for a product launch video” and get a usable result in under a minute.

5. Verbose Mode: Watch Your Agent Work

One of the things that made people nervous about autonomous agents was not knowing what they were doing in the background. Verbose mode fixes this directly.

Active Memory supports real-time verbose checks so users can see what the memory sub-agent is doing before the main response arrives. Type /verbose on and you get a live window into your agent’s reasoning process. This isn’t just a nice feature, it’s the trust layer that makes autonomous agents feel safe to use in the first place.

6. Token Optimisation: Cut Your API Costs

Prompt caching improvements in 4.5 keep prompt prefixes more reusable across transport fallback, enable deterministic MCP tool ordering, and include normalised system-prompt fingerprints and verbose cache diagnostics.

The practical outcome: your agent reuses more of what it already knows instead of re-sending the same context on every call. For anyone running an agent all day, that adds up quickly.

7. ChatGPT Memory Import

A new import function was added to Dreaming and Memory Wiki. Users can migrate ChatGPT conversation records to the OpenClaw memory system. New “Imported Insights” and “Memory Palace” tabs were added to the UI, AI interaction experience accumulated on other platforms can be seamlessly migrated.

If you’ve spent years building context in ChatGPT, you can now bring that with you instead of starting from zero.

8. Claude Opus 4.7 + GPT-5.4 Pro Now Supported

OpenClaw defaulted Anthropic selections, opus aliases, Claude CLI defaults, and bundled image understanding to Claude Opus 4.7. Forward-compatible support was also added for GPT-5.4 Pro, including Codex pricing and list visibility before the upstream catalog caught up.

This matters because OpenClaw’s whole value proposition is model flexibility. When better models drop, the harness supports them fast, often before official catalog updates.

Disclaimer on Security

This isn’t all sunshine. Because the software can access email accounts, calendars, messaging platforms, and other sensitive services, misconfigured or exposed instances present security and privacy risks. The agent is also susceptible to prompt injection attacks, where harmful instructions are embedded in data with the intent of getting the LLM to interpret them as legitimate user instructions.

One of OpenClaw’s own maintainers warned on Discord that “if you can’t understand how to run a command line, this is far too dangerous of a project for you to use safely.”

That’s not a reason to avoid it. It’s a reason to approach it carefully.

The team is actively patching. Each release cycle now includes a dedicated security hardening pass. But you should go in knowing: this is a powerful tool, and powerful tools require some responsibility on your end.

What This OpenClaw Updates Actually Means for You

OpenClaw is important not because it’s impressive software. It’s important because it represents a specific idea about where AI is going.

Most AI tools right now are reactive. You ask, they answer. OpenClaw is proactive. It watches, learns, acts, generates, and checks back in with you, across whatever messaging app you already use.

That shift from tool to teammate is the real story here. The 10 releases in 10 days aren’t about feature count. They’re about a community of over thousands of contributors who are building this direction as fast as humanly (and agentically) possible.

I shared a list of use cases or areas in the past to get started with OpenClaw or another AI agent called Hermes Agentby Nous Research.

What would you delegate to a 24/7 AI teammate if you had one?

 

---

Interested in travel or photography, read last week’s LensLetter newsletter about DaVinci Resolve Launched Photo Editing Feature in Release 21.

Read last week’s JustDraft about comparison between OpenClaw Vs Hermes and which one you should choose first?

---

Two Quotes to Inspire

Strategy is choosing the few moves that keep paying you back when the noise gets louder.

Automation used to mean replacing the repetitive. What’s coming next replaces the entire first draft of decision-making, and that changes what leadership actually means.

---

One Passage Summary From My Bookshelf

Ries opens the core of the book by arguing that traditional management thinking the kind built for predictable industries breaks down completely in conditions of extreme uncertainty. The plan that looks airtight in a boardroom becomes a liability the moment it meets real customers, real markets, and real feedback. What he proposes instead is a continuous loop he calls Build-Measure-Learn: you create the smallest possible version of something, measure how real people actually respond to it, and use those signals to decide whether to persevere or pivot. The idea sounds simple. The discipline required to follow it especially the willingness to throw away what you built is not.

From Book The Lean Startup by Eric Ries