Deep Agents Code released significant updates throughout July, expanding model support and introducing automation features that streamline the agent workflow. The month saw the addition of multiple new models, interactive goal management, and approval automation modes.
Models and Providers
- Claude Sonnet 5: Now available in the model picker for the latest Claude model capabilities.
- Multiple new model providers: Added support for GPT-5.6, Grok 4.5, Meta, Kimi K3, GLM-5.2, and Fireworks router models. The tool now infers additional model providers automatically.
Agents and Automation
- Auto and YOLO approval modes: Skip manual approval cycles with Auto mode, or use YOLO mode for even faster iteration. Both modes are now generally available, with an experimental classifier-backed Auto approval option and manual exit capability.
- Interactive goal management: Define and adjust agent objectives directly within the workflow. The /goal command now generates criteria server-side and auto-accepts them in YOLO mode.
- Reasoning effort selector: Control the computational intensity of agent reasoning tasks to balance speed and quality.
Interface and Workflow
- Slash commands and thread management: New /tools command lists available tools, /threads -r [ID] resumes existing threads, and mode-switching commands are now available. An install subcommand has been added.
- Plugin marketplace: Extend functionality with plugins, now generally available in version 0.1.42.
- Task description expansion: Click or press Ctrl+O to expand task descriptions, and use Ctrl+X for editing goal reviews.