Karigar Studio is an AI-powered cross-platform desktop studio built around agentic workflows for creating animations and video content through natural-language instructions.

Instead of treating an LLM as a simple chatbot, Karigar Studio gives AI agents access to a structured set of tools that can generate code, interact with the project workspace, execute workflows, verify generated output, and render the final result.

The application is designed around the idea that AI should be able to participate in the actual creative engineering workflow rather than only generate text.

KEY CAPABILITIES

• Agentic creative workflows
• Natural-language driven animation generation
• AI-generated executable code
• Tool-driven agent execution
• Code generation and verification
• Sandboxed filesystem operations
• Video and animation rendering
• Speech generation and transcription
• Automatic caption synchronization
• Multi-provider LLM integration
• Local and cloud AI workflows
• Context management and compaction
• Agent memory
• Prompt caching
• Replay/cache mechanisms for repeated requests
• Cross-platform desktop distribution

AGENTIC ARCHITECTURE

A core part of Karigar Studio is its agentic execution architecture.

The AI agent can reason about a creative task, generate the required code, interact with available tools, execute the workflow, verify the generated result, and continue iterating when required.

This creates a workflow closer to:

User Request → Agent Planning → Tool Selection → Code Generation → Execution → Verification → Rendering

rather than a conventional:

Prompt → Text Response

architecture.

The system therefore combines LLM application development with software execution, sandboxing, media processing, and desktop application engineering.

DESKTOP APPLICATION

Karigar Studio is built as a cross-platform Electron application with a modern React and TypeScript interface.

The desktop architecture allows the application to interact with local project files, rendering tools, AI services, and other system capabilities while maintaining controlled boundaries between the application and executed/generated code.

The project uses Electron, React, TypeScript and a pnpm workspace-based architecture.

AI PROVIDER ARCHITECTURE

Karigar Studio is designed around provider-agnostic AI integrations rather than being tightly coupled to a single model provider.

The architecture supports compatible AI providers and allows the application to work with different LLM backends.

The project also incorporates mechanisms for prompt caching, context management, agent memory, and replaying previously completed work where appropriate.

CREATIVE AND MEDIA PIPELINE

The application combines generative AI with an actual media-processing workflow.

Its creative pipeline integrates technologies such as:

• FFmpeg
• faster-whisper
• Edge-TTS
• ElevenLabs
• GSAP
• PixiJS
• D3
• Matter.js
• Anime.js
• Rough.js

These technologies allow generated content to move beyond text and into executable animations, audio, captions, and rendered media.

AI-GENERATED ANIMATION

One of the central ideas behind Karigar Studio is allowing an AI agent to generate actual animation code rather than simply returning instructions.

The generated code can work with animation and visualization libraries such as GSAP, PixiJS, D3, Matter.js, Anime.js and Rough.js.

This makes the system capable of translating natural-language creative instructions into executable visual workflows.

SECURITY AND SANDBOXING

Because an agent can generate and execute code, security is an important part of the architecture.

Karigar Studio uses controlled/sandboxed filesystem and execution mechanisms to reduce the risk associated with autonomous code execution.

The application also uses Electron's secure storage mechanisms for protecting sensitive API credentials.

DATABASE AND APPLICATION ARCHITECTURE

The project uses TypeORM and sql.js for application data persistence and local data management.

The architecture is organized to separate the desktop UI, AI/agent functionality, tools, rendering workflows, persistence, and supporting services.

DISTRIBUTION

Karigar Studio is designed as a distributable desktop application rather than only a development prototype.

The project includes release/update infrastructure using GitHub Releases and electron-updater.

TECHNICAL STACK

Frontend:
React, TypeScript

Desktop:
Electron

Package Management:
pnpm workspaces

AI:
LLM APIs, agentic workflows, tool calling, context management, prompt caching

Animation / Visualization:
GSAP, PixiJS, D3, Matter.js, Anime.js, Rough.js

Media:
FFmpeg, faster-whisper, Edge-TTS, ElevenLabs

Persistence:
TypeORM, sql.js

Distribution:
GitHub Releases, electron-updater

Security:
Electron safeStorage, sandboxed/controlled execution

ENGINEERING FOCUS

Karigar Studio demonstrates my interest in building complete AI-powered software systems rather than simply integrating an LLM API.

The project combines:

• AI agent architecture
• software engineering
• desktop application development
• code generation and execution
• tool orchestration
• media processing
• local AI capabilities
• security and sandboxing
• application persistence
• provider abstraction
• distribution and update infrastructure

The project is an exploration of what an AI-native creative development environment can look like when the AI is given the ability to use tools and participate directly in the software creation pipeline.