@ -8,6 +8,20 @@ At the center is **ABP Agent**, our AI coding assistant. Around it are long-stan
---
## Meet ABP Agent
ABP Agent is the AI coding assistant built into ABP Studio. It operates in three modes, each tuned for a different stage of work:
- **Agent mode**: the implementation mode. The agent reads the solution, writes and edits files, builds the affected projects, runs your apps, watches the runtime, and iterates until the change works end-to-end.
- **Plan mode**: read-only planning. The agent investigates the codebase, researches the official ABP documentation, and produces a structured implementation plan (Problem, Solution, Workflow Diagram, Files Affected, Expected Result). When you are happy with the plan, a single click promotes it into Agent mode and the implementation starts from the plan.
- **Ask mode**: read-only Q&A. The agent explains how something works, draws diagrams, and answers questions about your code without touching any files.
The agent is **ABP-aware by default**. It is instructed to prefer ABP base classes over plain POCOs, repositories over direct `DbContext` injection, `ApplicationService` over plain services, the ABP permission system over `[Authorize(Roles=…)]`, localized strings over hardcoded text, `BusinessException`/`UserFriendlyException` over plain `Exception`, the distributed cache abstraction over raw memory cache, and background job abstractions over hand-rolled hosted services. When it is unsure about an ABP feature, it consults the official ABP documentation as a primary source of truth, not random blog posts on the web.
---
## Why It Was Needed
General-purpose AI coding tools (Cursor, Claude Code, Windsurf, opencode and similar) are excellent for horizontal, file-shaped work. They read source files, edit them, and run shell commands. That works well for small scripts or front-end apps.
@ -25,20 +39,6 @@ ABP Agent and the surrounding ABP Studio features were built to close exactly th
---
## Meet ABP Agent
ABP Agent is the AI coding assistant built into ABP Studio. It operates in three modes, each tuned for a different stage of work:
- **Agent mode**: the implementation mode. The agent reads the solution, writes and edits files, builds the affected projects, runs your apps, watches the runtime, and iterates until the change works end-to-end.
- **Plan mode**: read-only planning. The agent investigates the codebase, researches the official ABP documentation, and produces a structured implementation plan (Problem, Solution, Workflow Diagram, Files Affected, Expected Result). When you are happy with the plan, a single click promotes it into Agent mode and the implementation starts from the plan.
- **Ask mode**: read-only Q&A. The agent explains how something works, draws diagrams, and answers questions about your code without touching any files.
The agent is **ABP-aware by default**. It is instructed to prefer ABP base classes over plain POCOs, repositories over direct `DbContext` injection, `ApplicationService` over plain services, the ABP permission system over `[Authorize(Roles=…)]`, localized strings over hardcoded text, `BusinessException`/`UserFriendlyException` over plain `Exception`, the distributed cache abstraction over raw memory cache, and background job abstractions over hand-rolled hosted services. When it is unsure about an ABP feature, it consults the official ABP documentation as a primary source of truth, not random blog posts on the web.
---
## How ABP Agent Sees Your Application: The Analyze Engine
Before ABP Agent answers anything, it needs to *understand* your solution. This is where the **Analyze** feature does the heavy lifting.
İsmail ÇAĞDAŞ
4 months ago