OpenClaw v2026.4.11: ChatGPT Import, Structured Chat & Better Video

OpenClaw v2026.4.11 is here. The big one this time is ChatGPT conversation import for the Memory Wiki, so your agent can learn from conversations you have already had elsewhere. There are also structured chat bubbles in the Control UI, notable upgrades to video generation, and a batch of fixes across messaging platforms. Here is what changed.

ChatGPT import for Memory Wiki

You can now import your ChatGPT conversation history directly into the Memory Wiki. This means all those useful exchanges you had with ChatGPT over the past months can become part of your agent's persistent knowledge base.

The Dreaming UI got two new subtabs to go with this: "Imported Insights" shows compiled wiki pages built from your imported chats, and "Memory Palace diary" lets you browse full source pages and individual conversations. You can inspect what was imported, how it was compiled, and trace any wiki entry back to its source.

If you have been using ChatGPT for research, brainstorming, or project planning, this gives your OpenClaw agent access to that same context without manual copy-paste.

Structured chat bubbles in Control UI

The Control UI and Webchat now render assistant responses with proper structure. Media attachments, reply references, and voice directives display as distinct chat bubbles instead of plain text. This makes conversations in the web interface much easier to follow, especially when the agent is sending multiple types of content in one response.

There is also a new [embed ...] rich output tag for embedding content inline. External embed URLs are gated behind configuration, so you control what domains are allowed. Security by default.

Video generation upgrades

Video generation got several practical improvements in this release:

• Generated assets are now delivered via URL only, so large files no longer sit in memory during processing
• Typed providerOptions give you proper autocomplete and validation when configuring video providers
• Reference audio inputs let you guide video generation with an audio track
• Per-asset role hints let you specify how each reference image or video should influence the output
• Adaptive aspect-ratio support picks the best dimensions based on your input
• Higher image-input cap means you can pass more reference images to a single generation request

Plugin setup descriptors

Plugin manifests can now declare their own activation and setup descriptors. This covers auth flows, pairing steps, and configuration UI. Previously, each plugin type needed special handling in the OpenClaw core. Now the plugin itself describes what setup looks like, which makes it easier to build and distribute new plugins without core changes.

Platform improvements

Microsoft Teams: Reactions

Teams now supports reactions. You can send and list reactions, with proper Graph API pagination and delegated OAuth for the auth flow. If you are running a Teams bot, this brings it closer to feature parity with other channels.

Feishu integration

The Feishu integration now parses richer context from messages and documents. Comment reactions and typing feedback work in document comment sessions, making Feishu bots feel more responsive.

Ollama performance

Ollama model discovery now caches /api/show context-window and capability metadata. If you have several Ollama models configured, you will notice faster startup and fewer redundant API calls during model detection.

Model provider debug

Debug logs for embedded agents now show how OpenAI-compatible endpoints get classified. If you have ever struggled to figure out why a model is routed one way or another, these logs will save you time.

Bug fixes

A solid batch of fixes in this release across several platforms:

• OpenAI/Codex OAuth: fixed the invalid_scope error during sign-in
• Audio transcription: fixed for OpenAI, Groq, and Mistral providers
• macOS Talk Mode: mic permission no longer requires toggling twice
• WhatsApp: fixed default account handling, reactions route correctly now, image edit paths preserved
• Telegram: fixed topic-scoped session initialization
• ACP/agents: child run progress chatter no longer leaks into parent conversations
• Google/Veo: removed unsupported numberOfVideos field that caused errors
• Agent failover: scoped to current attempt instead of pulling stale history
• MiniMax/OAuth: config patch now writes the correct auth routing

How to update

npm install -g openclaw@latest

Or with Docker:

docker pull openclaw/openclaw:latest

Verify with:

openclaw version

What to try first

ChatGPT import if you have existing ChatGPT conversations with useful context. Export your history, import it into the Memory Wiki, and let your agent build on what you already know.

Structured chat bubbles if you use the Control UI or Webchat regularly. The new rendering makes multi-part responses much more readable.

Video generation upgrades if you do any video work with OpenClaw. The URL-based delivery alone makes a noticeable difference in reliability for longer clips.