Frequently Asked Questions

By default, no. All write operations require your explicit approval through Telegram, Feishu, or other configured channels. You have full control over what gets changed, and can review, modify, or reject any proposed changes before they're applied.

Downcity supports all models compatible with ai-sdk v6, with OpenAI and DeepSeek as the standard examples in the docs. You can easily switch between models or use different models for different tasks.

The CLI currently focuses on local Agent development and execution. Remote deployment and managed hosting are on the roadmap for teams and enterprise use cases.

GitHub Copilot is for individual developers with code completion features. Downcity is designed for teams and enterprises, providing an executable AI team member with security controls, auditability, and human-in-the-loop approval workflows.

Yes. Downcity maintains project-level long-term memory. The Agent's context is your entire repository, and it remembers interactions, approvals, and actions across sessions.

Yes. The Agent SDK supports multiple independent Agents for different projects or purposes. Each Agent has its own configuration, plugins, and communication channels.

Absolutely. Downcity is designed to be extensible. You can add custom webhooks, integrate with your existing tools, and build custom skills. The webhook API makes it easy to connect with external systems.

Downcity is open source (Apache License 2.0) and free to use. You only pay for the LLM API costs, such as OpenAI or DeepSeek, that you choose to use. Enterprise editions with additional features and support will be available in the future.