What's New in OpenClaw: QA Lab, Memory Wiki, Media Generation, and More

The last batch of public OpenClaw releases landed in early April — 2026.4.1 through 2026.4.8 — and they packed in a lot. Since our early April check-in, the feature density has only gone up.

Here's what landed.

QA Lab: A Full Interactive Testing Suite

The biggest single addition this cycle is QA Lab — a complete testing and evaluation framework for agent behavior:

Live suite runner with a Slack-like chat UI and light/dark mode
One-command Docker launcher for isolated test environments
Repo-backed scenario runner that executes real-world workflows against your agent
Frontier harness bakeoff loop for comparing model performance side by side
Attachment understanding scenarios for testing multimodal agent behavior

If you've ever wondered whether your agent actually handles edge cases the way you expect, QA Lab gives you a systematic way to find out — without spamming your real chat channels.

Memory Wiki: Belief Layers, Digests, and Health Reports

OpenClaw's memory system got a significant architectural upgrade:

Belief-layer digests let the agent compile and reconcile what it "knows" over time
Health reports surface memory quality and coverage at a glance
Compiled digest prompts improve retrieval relevance without bloating the context window
REM preview and safe promotion replay make it safer to review and commit memory changes

The practical effect: agents that keep getting more useful the longer you use them, instead of gradually losing track of what they've learned.

Media Generation Pipeline: Video, Music, and Comfy Workflows

Media capabilities expanded dramatically:

Runway video generation provider, with detached async task tracking
Music generation tooling for creative and audio workflows
Comfy workflow support for custom image generation pipelines
Vydra media provider for additional generation options
Explicit media provider capabilities so the agent knows which provider can handle which request

Video and music generation tasks now run as tracked background jobs — the agent fires them off and reports back when they're done, rather than blocking the conversation.

Six New Model Providers and Better Model Detection

The model catalog kept growing:

Arcee AI — new provider plugin with OpenRouter support and onboarding instructions
Gemma 4 — Google's latest, with cross-provider resolution so you don't need to specify which provider hosts it
Bedrock IAM credential auth via @aws/bedrock-token for AWS-native setups
Bedrock embedding provider for memory search over Amazon's infrastructure
Bedrock Guardrails for content safety policies
Ollama vision detection — automatically detects vision capability from /api/show, so local models with vision just work without manual config

Search, Inference, and Gateway Improvements

On the tooling side:

SearXNG joins the bundled web search providers, giving you a self-hostable, privacy-respecting search backend
Infer CLI — a first-class infer command for running inference workflows directly from the terminal
Compaction checkpoints with a pluggable provider registry, making long-running sessions more memory-efficient
Configurable chat history — set your own max chars limit in the gateway
Chat-native task board for tracking agent tasks inline
Prompt cache runtime context exposed to context engines for faster, cheaper repeated inference

Agent Controls and Desktop Improvements

Prompt override and heartbeat controls per agent — fine-tune behavior without editing config files
Global default provider params (agents.defaults.params) so all agents inherit consistent model settings
macOS Voice Wake can now trigger Talk Mode, making hands-free conversation more fluid
iOS gateway shows clearer connection error messages instead of silent failures
Cron --tools flag for per-job tool allow-lists — schedule a research job with web search but no shell access, for example
GitHub App helper (gh-read) for agent-assisted code review workflows

ClawSimple Platform Updates

On the managed side, the past month brought several improvements to how ClawSimple runs OpenClaw:

Automated release check rollback — if a new OpenClaw version fails validation, your deployment automatically rolls back to the last known-good version instead of staying broken
Deployment health monitoring — continuous SSH-based health probes catch issues before you notice them
Disk pressure reporting — the platform now warns and auto-cleans when disk space runs critically low
Diagram workflow skills — new built-in skills for flowcharts and tech diagrams, with SVG previews in the inspiration UI
Support inbox cron gate — the support AI only activates when there's actually new mail to process, keeping things efficient

As always, ClawSimple-managed agents get these updates automatically. If you want the latest right now, head to your Profile and hit Relaunch.