Open WebUI v0.13.3 adds automation agents and improved terminal integration

Open WebUI has released version 0.13.3, introducing significant updates to its automation capabilities and terminal-based workflows. The release focuses on integrating coding agents directly into the automation editor, allowing users to define specific terminals and working directories for automated tasks.

Beyond automation, the update improves the user experience for local deployments by fixing clipboard and copy functionality over plain HTTP connections. It also addresses specific compatibility issues with DeepSeek thinking mode when using llama.cpp as a provider and refines how tool calls and knowledge base searches handle punctuation and labeling.

Technical improvements in Open WebUI v0.13.3

• Automation Engine: Users can now select specific terminals and working folders for automations, enabling the use of coding agents for scheduled runs that verify agent availability before execution.
• Terminal and File Management: The interface now allows switching terminals and working directories mid-chat, with the Files tab automatically updating to reflect the contents of the newly selected terminal.
• Model Provider Compatibility: Fixed a reasoning error for DeepSeek thinking mode when the connection provider is set to llama.cpp, ensuring compatibility with skills, knowledge, and memories.
• Local Network UX: Improved support for plain HTTP environments by enabling the copying of links, hashes, and spreadsheet cells, and allowing clipboard-based prompts to function even when browsers restrict access.
• Tool Call Logic: Enhanced tool call labeling ensures that groups of calls retain labels even if individual calls are unlabeled, and ensures models can reference previously written labels in subsequent replies.

Why it matters

These updates strengthen Open WebUI's position as a robust local AI orchestration layer, particularly for developers using terminal-based workflows. By improving reliability for llama.cpp users and enhancing local network functionality, it lowers the friction for self-hosting AI interfaces in private or air-gapped environments.