Opus Orchestrator AI

Full-flow AI book generation system using LangGraph, CrewAI, AutoGen, and PydanticAI

A comprehensive, production-ready system for generating publication-ready manuscripts from raw content.

---

⚡ Quick Start

# Install
pip install opus-orchestrator-ai

# Generate a manuscript (local mode)
opus generate --concept "A robot dreams of electric sheep" --words 5000

# Or use API server mode
opus serve --port 8000
opus --api-url http://localhost:8000 generate --concept "Your idea"

---

🏗️ Architecture

┌─────────────────────────────────────────────────────────────────────┐
│                      OPUS ORCHESTRATOR AI                          │
├─────────────────────────────────────────────────────────────────────┤
│                                                                     │
│  ┌─────────────┐    ┌─────────────┐    ┌─────────────────────┐  │
│  │   INGEST   │───►│   LANGGRAPH │───►│    OUTPUT           │  │
│  │   LAYER    │    │   WORKFLOW  │    │    (Manuscript)     │  │
│  └─────────────┘    └─────────────┘    └─────────────────────┘  │
│         │                  │                                       │
│    ┌────┴────┐       ┌────┴────┐                                 │
│    ▼         ▼       ▼         ▼                                  │
│ GitHub    S3/MinIO  CrewAI   AutoGen                              │
│ Ingestor  Ingestor   Agents   Critique                             │
│                                                                     │
│  ┌─────────────────────────────────────────────────────────────┐  │
│  │                   VALIDATION LAYER                          │  │
│  │                    PydanticAI                               │  │
│  └─────────────────────────────────────────────────────────────┘  │
│                                                                     │
└─────────────────────────────────────────────────────────────────────┘

---

📦 Features

Core Generation

Feature	Description	Status
Snowflake Method	7-stage fractal expansion from sentence to novel	✅
Story Frameworks	7 frameworks: Snowflake, 3-Act, Save the Cat, Hero's Journey, Story Circle, 7-Point, Fichtean	✅
LangGraph Workflow	State machine with streaming progress	✅
AutoGen Critique	Multi-agent debate (LiteraryCritic, GenreExpert, StoryEditor)	✅
PydanticAI Validation	Structured output validation with type-safe schemas	✅

Agent Systems

Agent	Role	Status
ArchitectAgent	Story structure & blueprint	✅
WorldsmithAgent	World-building & setting	✅
CharacterLeadAgent	Character development	✅
VoiceAgent	Narrative voice & tone	✅
EditorAgent	Editorial review	✅
ResearcherAgent (Nonfiction)	Fact-finding	✅
AnalystAgent (Nonfiction)	Argument analysis	✅

Ingestion Sources

Source	Description	Status
GitHub	Fetch from public/private repos	✅
S3/MinIO	S3-compatible object storage	✅
Local Files	Direct file/directory input	✅

Ingestion CLI Commands

# Ingest from GitHub
opus ingest --repo owner/repo

# Ingest from S3/MinIO
opus ingest-s3 --bucket my-bucket --prefix notes/

# Ingest from local files/directory
opus ingest-local ./my-notes/
opus ingest-local ./manuscript.txt
opus ingest-local ./folder --extensions md,txt --recursive
opus ingest-local ./notes --summarize --max-length 5000

Deployment

Mode	Description	Status
CLI	Standalone command-line tool	✅
API Server	FastAPI REST server	✅
API Client	Client mode for remote servers	✅
Python Module	Import as library	✅

Output Options

Destination	CLI Flag	Description
Local File	--output FILE	Save to local filesystem
S3/MinIO	--save-s3 BUCKET/PATH	Upload to S3-compatible storage
GitHub	--save-repo OWNER/REPO	Commit to GitHub repository

---

🚀 Usage

CLI Commands

# Generate manuscript from concept (local)
opus generate --concept "Your story idea" --framework snowflake --words 5000

# Generate from GitHub
opus generate --repo owner/repo --framework hero-journey --words 80000

# Generate from local files/directory
opus generate --local ./my-notes/ --framework snowflake --words 5000
opus generate --local ./manuscript.txt --words 5000

# Generate and save to S3/MinIO
opus generate --concept "..." --save-s3 my-bucket/manuscripts/
opus generate --concept "..." --save-s3 my-bucket/path/ --save-s3-endpoint https://nyc3.digitaloceanspaces.com

# Generate and save to GitHub repo
opus generate --concept "..." --save-repo owner/my-manuscripts
opus generate --concept "..." --save-repo owner/my-manuscripts --save-branch develop --save-commit-msg "New story draft"

# Generate from one source, save to another
opus generate --repo owner/notes --save-s3 output-bucket/manuscripts/
opus generate --local ./notes --save-repo owner/output-repo

# Ingest from various sources
opus ingest --repo owner/repo
opus ingest-s3 --bucket my-bucket --prefix notes/
opus ingest-local ./my-notes/
opus ingest-local ./folder --extensions md,txt,notes --recursive
opus ingest-local ./notes --summarize --max-length 5000

# Start API server
opus serve --port 8000

# Use as API client
opus --api-url http://localhost:8000 generate --concept "..."

# List frameworks
opus frameworks

# Show config
opus config

# Show docs
opus docs

Python API

from opus_orchestrator import run_opus

# Simple generation
result = await run_opus(
    seed_concept="A robot dreams of love",
    framework="snowflake",
    genre="science-fiction",
    target_word_count=5000,
)

manuscript = result["manuscript"]

Using CrewAI

from opus_orchestrator.crews import create_fiction_crew

crew = create_fiction_crew(
    genre="science-fiction",
    tone="literary",
    target_word_count=2000,
)

story = crew.write_full_story(
    story_outline="Your outline...",
    character_sheets="...",
    style_guide="Tone: literary",
    num_chapters=5,
)

Using PydanticAI Validation

from opus_orchestrator import create_style_guide_agent

agent = create_style_guide_agent()
result = agent.run_sync("Create a style guide for a literary novel")

# Result is a validated StyleGuide object
print(result.tone)        # "Contemplative and introspective"
print(result.pacing)      # "Deliberate with moments of acceleration"

API Server

from opus_orchestrator.server import app, run_server

# Run server
await run_server(host="0.0.0.0", port=8000)

# Or with uvicorn directly
# uvicorn opus_orchestrator.server:app --port 8000

API Client

from opus_orchestrator.cli import OpusAPIClient

client = OpusAPIClient("http://localhost:8000")

# Health check
health = client.health()

# Generate
result = client.generate(
    concept="A robot dreams",
    framework="snowflake",
    target_word_count=5000,
)

print(result["manuscript"])

---

⚙️ Configuration

Environment Variables

Variable	Description	Required
OPENAI_API_KEY	OpenAI API key	Yes (or MINIMAX_API_KEY)
MINIMAX_API_KEY	MiniMax API key	No
GITHUB_TOKEN	GitHub token for private repos	No
AWS_ACCESS_KEY_ID	AWS access key for S3	No
AWS_SECRET_ACCESS_KEY	AWS secret key for S3	No
S3_ENDPOINT_URL	Custom S3 endpoint (MinIO, DO Spaces)	No

Configuration File

# config.yaml
agent:
  provider: openai
  model: gpt-4o
  temperature: 0.7
  max_tokens: null

iteration:
  min_critic_rounds: 2
  max_critic_rounds: 5
  approval_threshold: 0.8

output:
  format: markdown
  include_toc: true
  output_dir: ./output

---

📚 Story Frameworks

Implemented Frameworks

Framework	Type	Stages/Beats
Snowflake Method	Fractal	7 stages
Three-Act Structure	Linear	7 beats
Save the Cat	Screenplay	15 beats
Hero's Journey	Mythic	12 stages
Story Circle	Circular	8 beats
7-Point Plot	Structural	7 beats
Fichtean Curve	Episodic	7 beats

---

🧩 Project Structure

opus_orchestrator/
├── __init__.py           # Main exports
├── __main__.py           # CLI entry point
├── cli.py                # CLI implementation
├── server.py             # FastAPI server
├── orchestrator.py       # Main orchestrator
├── langgraph_workflow.py # LangGraph pipeline
├── autogen_critique.py  # AutoGen critique crew
├── pydanticai_agent.py  # PydanticAI agents
├── config.py             # Configuration
├── frameworks.py          # Story frameworks
│
├── agents/               # Agent implementations
│   ├── fiction/          # Fiction agents
│   │   ├── architect.py
│   │   ├── worldsmith.py
│   │   ├── character_lead.py
│   │   ├── voice.py
│   │   └── editor.py
│   └── nonfiction/      # Nonfiction agents
│       ├── researcher.py
│       ├── analyst.py
│       ├── writer.py
│       ├── fact_checker.py
│       └── editor.py
│
├── crews/               # CrewAI crews
│   ├── base_crew.py     # Base crew class
│   ├── fiction_crew.py  # Fiction crew
│   └── nonfiction_crew.py
│
├── schemas/              # Pydantic schemas
│   └── book.py
│
└── utils/               # Utilities
    ├── github_ingest.py # GitHub ingestion
    ├── s3_ingest.py    # S3/MinIO ingestion
    ├── llm.py          # LLM client
    └── docs.py         # Documentation generator

---

🧪 Testing

# Run tests
pytest tests/

# Run with coverage
pytest --cov=opus_orchestrator tests/

# Lint
ruff check .

# Format
ruff format .

---

📄 License

MIT License

---

🤝 Built With

• LangGraph - Workflow orchestration
• CrewAI - Multi-agent systems
• AutoGen - Complex agent conversations
• PydanticAI - Structured output validation
• FastAPI - REST API server
• OpenAI - LLM provider

---

Built with the WE Architecture — witness and co-creation in code.