Speedwave 0.18

Automatic tokenization of sensitive data

Speedwave can now scan outbound requests sent to the model provider as well as data returned from integrations for sensitive information, such as personal identifiers, financial details, and API keys. Each detected value is replaced with a secure placeholder token before reaching the model. The architecture requires no central database or lookup table: tokens are encrypted with AES-128-SIV using a dedicated per-project key. As a result, identical values within a project map consistently to the same token, while modified tokens cannot be decrypted.

Three preconfigured policy templates are included: strict, gdpr-art32, and eu-ai-act-art5. You can enable them under Settings → Security once beta features are active. Organization policies enforced via MDM apply regardless of the beta toggle state. See Tokenization for detailed information on detection rules, key management, and audit logs.

Claude Opus 5

Claude Opus 5 is now available in the Latest category of the Anthropic model catalog with a 1M-token context window, while Opus 4.8 moves to Legacy. The full model overview is available in LLM providers.

Resizable chat composer

The message input field (composer) now includes a resize handle, allowing you to expand the input area vertically beyond the automatic height limit when drafting longer prompts. Learn more about the interface in Desktop app.

Bug fixes and improvements

This release restores container image building in installed applications, resolves an issue in execute_code that disconnected integrations when a service identifier contained dashes, and fixes default plugin installation in fresh projects. It also addresses a CSS injection vulnerability in PDF rendering and includes security updates for the h2 and vite dependencies.