isA Creative

Multi-agent creative production system with 6 specialized AI agents for automated content creation, brand compliance, and digital asset management.

Overview

isA Creative is a coordinated multi-agent architecture where specialized agents collaborate to produce brand-consistent creative assets. A single orchestrator process runs all agents in-process via SwarmOrchestrator, following the same pattern as isA Trade.

Quick Start

# Install
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
 
# Start infrastructure
docker-compose up -d  # PostgreSQL, Redis, NATS, MinIO, Qdrant
 
# Start the orchestrator (all agents run in-process)
uvicorn src.orchestrator.main:app --host 0.0.0.0 --port 18791

The API is available at http://localhost:18791. All agents run within the single orchestrator process — no separate ports needed. Individual specialist agents can be run standalone for dev/testing only (not used in production).

Architecture

┌─────────────────────────────────────────────────────────┐
│                   Client Applications                    │
│              REST / WebSocket / A2A Protocol              │
└────────────────────────┬────────────────────────────────┘
                         ↓
┌─────────────────────────────────────────────────────────┐
│          Creative Director (Port 18791)                  │
│     Orchestrates workflows, coordinates all agents       │
│     Workflow Engine • Decision Engine • Event Handlers   │
└──┬──────┬──────────┬──────────┬──────────┬──────────┬───┘
   ↓      ↓          ↓          ↓          ↓          ↓
┌──────┐┌────────┐┌────────┐┌────────┐┌────────┐┌────────┐
│ DAM  ││Content ││Visual  ││ Copy   ││ Brand  ││Design  │
│Mngr  ││Creator ││Designer││ Editor ││Strtgst ││System  │
│      ││        ││        ││        ││        ││GUARDIAN │
│Assets││Drafts  ││Images  ││Polish  ││Strategy││ VETO   │
│Tags  ││Scripts ││Layouts ││Tone    ││Align   ││POWER   │
│Search││Copy    ││ComfyUI ││Grammar ││Guide   ││Validate│
└──────┘└────────┘└────────┘└────────┘└────────┘└────────┘
   │         │          │         │         │         │
   └─────────┴──────────┼─────────┴─────────┴─────────┘
                        ↓
┌─────────────────────────────────────────────────────────┐
│                   Infrastructure                         │
│  PostgreSQL • Redis • NATS • MinIO (S3) • Qdrant (Vec)  │
└─────────────────────────────────────────────────────────┘
                        ↓
┌─────────────────────────────────────────────────────────┐
│                 External Engines                         │
│  ComfyUI (Image Gen) • Remotion (Video) • TTS Service   │
│  GPU Cluster (A100/H100) • AWS Lambda (Render Farm)     │
└─────────────────────────────────────────────────────────┘

Key Features

Digital Asset Management (DAM)

The DAM Manager provides centralized asset storage and retrieval backed by MinIO (S3-compatible) and Qdrant vector search:

• Asset storage — Upload, version, and organize creative assets in S3-compatible object storage
• Auto-tagging — AI-powered metadata extraction and classification
• Semantic search — Vector-based asset search using Qdrant embeddings
• Asset lifecycle — Track asset versions, usage, and expiration

Brand Compliance

The Design System Agent acts as a guardian with veto power, enforcing brand standards across all outputs:

• Color validation — Ensures palette adherence to brand guidelines
• Typography checks — Verifies font usage, sizing, and hierarchy
• Layout validation — Confirms spacing, grid alignment, and composition rules
• Accessibility validation — WCAG compliance checks on generated content

Visual Design

The Visual Designer generates and composes visual content using AI:

• AI image generation — ComfyUI integration with SDXL, FLUX, and HunyuanVideo models
• Layout composition — Programmatic layout engine for combining visual elements
• ControlNet guidance — Structural control via pose, depth, and edge maps
• Style transfer — IP-Adapter and LoRA support for consistent brand aesthetics

Content Production

End-to-end content creation pipeline across multiple media types:

• Video composition — Remotion-based video rendering with React components and AWS Lambda render farm
• Text-to-speech — Multi-provider TTS (Chatterbox, ElevenLabs, Kokoro) with voice cloning
• Written content — AI-generated copy, articles, and scripts with editorial review
• Bulk operations — Batch processing for high-volume creative production

Content Safety

Multi-layer content moderation pipeline:

• Input filtering — Block unsafe prompts before generation via text classifier and blocklist
• Embedding guard — Prevent semantic prompt injection with PromptGuard vectors
• Output scanning — Vision API and classifier verification of generated content
• Human review queue — Escalation path for edge cases with moderate confidence scores

API Endpoints

Agent Capabilities

Content Creator

• Generates written content from creative briefs
• Produces multiple content variations with scoring
• Supports articles, ad copy, social posts, scripts, and more

Visual Designer

• AI image generation via ComfyUI workflows
• Layout composition with programmatic layout engine
• Supports ControlNet, LoRA, and IP-Adapter for style control
• Handles SDXL, FLUX, and HunyuanVideo model checkpoints

Copy Editor

• Grammar, tone, and style review
• Brand voice consistency checks
• Readability scoring and improvement suggestions

Brand Strategist

• Brand guideline alignment verification
• Campaign strategy recommendations
• Audience targeting and messaging consistency

DAM Manager

• S3-compatible storage via MinIO
• Vector-powered semantic asset search via Qdrant
• Automated tagging, categorization, and metadata management

Design System Agent (Guardian)

• Veto power over non-compliant outputs
• Color, typography, layout, and accessibility validation
• Quality gate enforcement via QualityGateRunner

Tech Stack

Infrastructure

All infrastructure runs in the local KIND cluster (isa-creative namespace):

# Start infrastructure
docker-compose up -d
 
# Verify services
curl http://localhost:18791/health