Deep Agents Code released versions 0.1.12 through 0.1.21 in June, adding new agent management features, authentication options, and diagnostic tools to improve developer workflows.
Agent Management and Workflow
- Dynamic subagents UI: Visualize and manage agent hierarchies directly in the interface for better control over complex agent structures.
- Agent filter dropdown: Filter and select specific agents in the threads picker, making it easier to manage multiple conversations.
- Session restart command: Use /restart to reset your agent session while preserving context without restarting the entire application.
- Rubric-backed goals: Define goals and rubrics via /goal, --goal, /rubric, and --rubric commands to enable more structured agent behavior.
Models and Authentication
- ChatGPT OAuth sign-in: Authenticate directly through your ChatGPT account for Codex models, streamlining the login process.
- Latest frontier models: Access cutting-edge language models added to the recommended list for your agent workflows.
Configuration and Debugging
- Config management: Use the config command and canonical config manifest to standardize and manage configuration settings more easily.
- Diagnostics command: Run dcode doctor to identify and troubleshoot issues with your setup.
- Vercel Sandbox execution: Execute code in safer, isolated runtime environments using Vercel Sandbox as a new code execution provider.