Agent Mode (Default)
Agent mode is the default when you send a message. Aurora will interpret your request and immediately start making changes to your project. It can:- Explore your codebase for the necessary context
- Read and edit files on demand
- Fix and refactor code as needed
- Search the web for documentation and examples
- Fetch and read a reference URL you attach to your message
Built-in debugging
When a build error or runtime issue appears, Aurora can inspect your project’s logs to find the cause and fix it. Tell Aurora what you see, or paste the error, and it will investigate.Pricing
Aurora uses usage-based pricing, which you can read more about here. A small tweak costs about 0.3 credits, a typical request about 1.5, and a multi-file feature about 3. Every request is itemized in your usage history.Chat Mode
Chat mode is an opt-in conversational mode for when you want to think through a problem before making changes. Chat messages use the same usage-based pricing as Agent mode, and they cost little because they do not edit files.Activate Chat mode
To activate Chat mode, click “Chat” next to where you input your messages in Aurora. Unlike Agent mode, Chat mode won’t make changes to your code directly. Instead, it will create a step-by-step plan for what needs to be done. Once you’re happy with the plan, press the “Implement the plan” button and Aurora will switch to Agent mode to carry out the changes. Chat mode is still agentic — it can reason across multiple steps and decide when to search files, inspect logs, query the database and more. It just won’t touch your code until you approve the plan.Best use cases for Chat mode
- Debugging complex issues by investigating in depth before requesting a fix
- Planning database structures and relationships
- Scoping out new features by identifying the minimal changes required
- Understanding how potential changes will impact your application
- Getting improvement tips and architectural advice
Choosing a Model
You pick the model for every conversation, and you can switch in the middle of a project. The picker offers:- Claude Opus 5: the default on paid plans. Use it for hard problems: architecture, refactors, and stubborn bugs.
- Claude Sonnet 4.6: the default on the free plan. Strong quality at a lower credit cost.
- GPT-5.6 Terra: a second opinion with a different set of strengths.
- GPT-5.6 Luna: the fastest option for quick iterations.
- Kimi K3: fast and cheap for everyday iteration.