Built in universal vector embedding and integrated video generation, OpenClaw releases v2026.5.27 According to Beating monitoring, the open-source AI assistant project OpenClaw has released version 2026.5.27, which incorporates OpenAI compatible vector embedding services that support local and hosted endpoints as a built-in core engine and abandons the original plugin compatibility registration mode. The new version also integrates Pixverse video generation service, supports API region selection and external plugin encapsulation, and supplements the AI video creation capability for the native intelligent agent ecosystem. At the same time, the new version has integrated vLLM deep thinking parameter docking and allowed direct routing of naked Anthropic model identifiers. In terms of security defense and channel delivery, the new version has implemented stricter security isolation. The system completely isolates the group prompt text outside the system prompt to prevent malicious prompt word injection, and intercepts command wrappers with side effects and non secure Node runtime environment variable overwriting. To prevent unauthorized Tailscale network exposure, node and device role approval permissions have been granted to administrators. On the instant messaging channel, the Telegram message sending mechanism has been changed to a persistent delivery queue, improving the success rate of sending across network fluctuations. [Original link]