Docs/Custom Agents

Training Custom AI Agents

The true power of Nirmaanly AI lies in Custom Agents. Tailor the generation engine precisely to your company's exact branding, preferred component libraries, and corporate coding guidelines.

Persona Configuration

From the "Train Agent" dashboard, you can define specific system prompts. Tell the agent exactly which UI library to prefer (e.g., Shadcn UI vs pure Tailwind) and your exact hex color codes.

"Never use generic blue. Always use strict hex #7c3aed for primary CTA buttons. Components must be functional pure React. Always import Lucide icons for UI symbolism."

Contextual Memory

Agents accumulate "Memory" through continuous Builder feedback. Every time you ask for a tweak in the Editor via chat, the active Agent records the correction securely in your workspace profile.

Learned: Prefer Grid over Flex for layouts.
Learned: Always add 16px of top padding to Navbars.

Advanced Agent Architecture

Global Workspace Agents

Enterprise users can set an overarching "Global Agent" that enforces strict brand guidelines across every single project spawned within their workspace container. This overriding system prompt acts as a filter preventing off-brand UI decisions.

Project-Scoped Sub-Agents

Individual projects can possess sub-agents trained for specific micro-services. E.g., You can train an internal agent specifically on your legacy Postgres schema, allowing you to prompt: "Build an analytics dashboard pulling from the internal HR table structure."