AI-Native Platform for Cognitive Orchestration & Research Automation
Sigma Studio is a cognitive orchestration engine — an executable environment where AI agents create, verify, document, and organize knowledge, governed by Modelfile manifests that define their behavior.
Imagine a team of specialized AI agents (a mathematician, an architect, a tester, a reviewer, a viz designer) working 24/7 on your research, tracking every action, testing every theorem, and building a navigable knowledge graph of everything they produce.
This is Sigma Studio.
| Feature | Why It Matters |
|---|---|
| 🧠 Multi-Provider AI | Ollama (local), DeepSeek, OpenAI, Anthropic, Groq, OpenRouter — you choose the brain |
| 📜 Manifesto System | Every agent has a "code of conduct" written as an Ollama Modelfile. Not human instructions — executable contracts |
| 🔬 Automated Research | AI explores, proves, refutes, and documents — without you lifting a finger |
| 🏗️ Full-Stack AI | From academic theory to working software: theorems → tests → D3.js visualizations → whitepapers |
| 🔒 Sandbox Security | Every operation is confined to whitelisted paths. No agent touches system files |
| 🧩 Modular Architecture | Python backend + React 19 frontend + Multi-provider AI: fully composable |
| 🤖 Multi-Agent Orchestration | Parallel pipelines with specialized agents, context sharing, and automatic delegation |
| 📚 Research Sessions | Decompose complex objectives into micro-tasks with full traceability |
| 🧠 Training Lab | LLM fine-tuning & pre-training with a complete interface — datasets, methods, live monitoring |
| ⚡ Hardware & GPU Monitor | Real-time GPU monitoring, multi-GPU configuration, and Ollama parallelism tuning |
Topic Map with D3.js Relational Graph (left) and Automated Tests / Module Validation (right).
Research Lab with micro-task planning (left) and Multi-Agent Chat with Manifest integration (right).
Sigma Lab Editor for drafting and editing files (left) and Interactive D3.js visualizations generated by agents (right).
All the theory files, formula sheets, interactive D3.js graphs, and test scripts visible in the screenshots were generated from a single initial prompt entered in the Research Lab:
"Scriviamo tutti gli argomenti e i sottoargomenti trattati in un corso di Analisi 1 matematica ingegneria con dimostrazioni, formulari, esercizi in files separati e tutto il necessario a comprendere perfettamente la materia"
From this single input, the Sigma Architect coordinator and the agent pipeline:
- Analyzed the domain and split the roadmap into 7 sequential modules (from Sequences to Differential Equations).
- Generated the theory in Markdown files enriched with LaTeX formulas and rigorous definitions.
- Wrote and executed Python unit tests (
test-engineer) with mathematical validation and automatic self-healing on errors. - Designed interactive visualizations (
viz-designer) ready to navigate in Sigma Lab. - Authored whitepapers and formal validation reports to certify the work.
Sigma Studio's chat is not a simple chatbot — it's a flexible cognitive control panel featuring:
- 4 Operating Modes:
- Ask: Quick explanations and theoretical questions without modifying the workspace.
- Plan: Decomposition of complex goals into micro-tasks saved directly in the Roadmap.
- Execute: Real-time file creation and modification on disk with Sandbox control.
- Complete Task: Assisted or autonomous resolution of specific Roadmap tasks.
- Manifest Association: Switch agent behavior on the fly by associating configuration Manifests, with automatic restoration when navigating between different sessions.
- Real-Time Tracking: Every action (file creation, command execution, validation tests) leaves a structured notification for maximum operational transparency.
The Training Lab is a complete environment for LLM fine-tuning and pre-training, accessible from the sidebar menu 🧠 Training Lab and organized into 4 sections:
Complete guide to supported training methods, hardware requirements, and best practices.
Browse, search, and manage training datasets:
- ⭐ Featured: 15 curated open-source datasets for LLM training, organized by category:
- Instruction Tuning: Alpaca (52K), Dolly (15K), OpenHermes 2.5 (1M+), UltraChat 200K, OpenOrca (3.2M)
- Code Training: CodeAlpaca 20K, Python Code Instructions 18K, StarCoder Data (783GB)
- Math & Reasoning: MetaMathQA (395K), GSM8K (8.5K), MATH (12.5K)
- Pre-Training: TinyStories (2M+), OpenWebText, The Pile (825GB)
- Multilingual: Italian Dolly, OPUS-100
- 🔍 HuggingFace Search: Search 100K+ datasets on HuggingFace Hub with preview and metadata
- 📂 Local Import: Drag & drop JSONL, JSON, CSV, or TXT files — automatic parsing
- 🗂️ My Datasets: Manage imported datasets with selection for training
Complete training job configurator with:
- 4 Training Methods:
Method Description Min VRAM When to Use ⚡ LoRA (Unsloth) Efficient LoRA 4-bit fine-tuning 8 GB Recommended for most cases — 2x faster, 60% less VRAM 🔬 SFT (TRL) Supervised Fine-Tuning with PEFT 12 GB Stable and versatile, supports all HuggingFace models 🌐 Full Pre-Training Training from scratch on raw text 4-80 GB For training models from zero (e.g., TinyStories 4GB, The Pile 80GB) 🛠️ Custom Script Customizable Python template — Maximum flexibility, edit script before launching - Base Model Selection: Popular HuggingFace models (LLaMA 3.2, Mistral, Phi-3, Gemma) + local Ollama models + custom model
- Adjustable Hyperparameters: Epochs, Batch Size, Learning Rate, Max Sequence Length, LoRA Rank/Alpha, Gradient Accumulation
- Job Summary: Visual summary of all parameters before launch
Real-time training job monitoring:
- Hardware Strip: CUDA status, detected GPUs (name, VRAM, temperature), RAM, PyTorch version
- CUDA Diagnostics: Automatic CUDA/driver mismatch detection with recommended fix commands
- Job Selector: All job history with status (Ready/Running/Completed/Failed/Stopped)
- Loss Chart: Interactive SVG visualization of loss over time with automatic log parsing
- Live Terminal: Streaming output with color-coded SIGMA/Error/Warning/Success lines
- Controls: Start, Stop, Delete jobs — Auto-scroll, Copy logs, Clear output
- Export → Ollama: Export modal to transform trained models into Ollama Modelfiles
- Automatic Modelfile generation
- Customizable System Prompt
- Integrated
ollama create/ollama runcommands
| Method | Endpoint | Function |
|---|---|---|
GET |
/api/training/datasets |
List imported datasets |
GET |
/api/training/datasets/search?q=... |
HuggingFace search |
GET |
/api/training/datasets/featured |
Featured datasets |
GET |
/api/training/jobs |
List training jobs |
GET |
/api/training/job/status?job_id=... |
Specific job status |
GET |
/api/training/job/logs?job_id=...&offset=0 |
Live job logs |
GET |
/api/training/hardware |
Hardware status for training |
POST |
/api/training/dataset/import |
Import local dataset |
POST |
/api/training/dataset/register_hf |
Register HuggingFace dataset |
POST |
/api/training/dataset/delete |
Delete dataset |
POST |
/api/training/job/create |
Create training job |
POST |
/api/training/job/start |
Start job |
POST |
/api/training/job/stop |
Stop job |
POST |
/api/training/job/delete |
Delete job |
POST |
/api/training/export/ollama |
Export model → Ollama |
POST |
/api/training/dependencies |
Check method dependencies |
POST |
/api/training/job/clear_logs |
Clear job logs |
The Hardware & GPU Monitor is a complete control panel for GPU hardware monitoring and configuration, accessible from the sidebar menu ⚡ Hardware & GPU.
- Detailed GPU Cards: For each detected GPU shows:
- Model name, driver, PCIe bus (Gen/Width)
- VRAM: Usage bar with MB/GB values and percentage
- Compute Load: GPU utilization percentage
- Power Draw: Current vs limit watts, with proportional bar
- Temperature, free VRAM, Compute Capability
- Auto-Refresh: Configurable polling every 2 seconds with pause/resume
- Summary Badge: Active GPU count, auto-refresh status
Control panel for optimizing parallelism on multi-GPU systems:
| Parameter | Options | Description |
|---|---|---|
| CUDA_VISIBLE_DEVICES | 0,1 (both), 0 (GPU0 only), 1 (GPU1 only) |
Defines which GPUs are visible to models |
| OLLAMA_NUM_PARALLEL | 1/2/4/8 slots | Concurrent requests Ollama can process |
| OLLAMA_MAX_LOADED_MODELS | 1/2/3/4 models | Models kept in VRAM without reloading |
| Preferred Training Lab GPU | cuda:0 / cuda:1 / cuda:0,1 | Target GPU for PyTorch training jobs |
- Set your HF token to speed up model downloads (up to 10x)
- Persistent server-side storage
- Visual "Token configured" indicator
Table of processes running on the GPU (Ollama, PyTorch, system) with:
- Bus ID, PID, process name, executable path, used VRAM
| Method | Endpoint | Function |
|---|---|---|
GET |
/api/hardware/status |
Full hardware status + current configuration |
POST |
/api/hardware/config |
Save and apply multi-GPU configuration |
POST |
/api/config/hf_token |
Save HuggingFace Token |
Sigma Studio is an open-source project in continuous evolution and enthusiastically welcomes contributions from the community! You can contribute in many ways:
- 📜 New Manifests: Create and share new agent roles (
manifesti/*.md) specialized in scientific, engineering, or creative fields. - 🎨 UI/UX Improvements: Extend the glassmorphism design system in React 19.
- 🔧 Backend Extensions: Add new AI providers, optimize the test pipeline, or enrich the REST API.
- 🔬 Research Pipelines: Integrate new validation tools or multi-agent orchestration templates.
- 🧠 Training Lab: New training methods, quantization support (GGUF, AWQ), new dataset templates.
- ⚡ Hardware Monitor: AMD ROCm integration, additional metrics (fan speed, memory clock).
- Python 3.10+
- Node.js / npm
- Ollama (for local AI — download here)
# 1. Clone the repository
git clone https://github.com/Sigmanih/SigmaStudio.git
cd SigmaStudio
# 2. Install Python dependencies
pip install -r requirements-backend.txt
# 3. Install frontend dependencies
cd sigma_studio && npm install && cd ..
# 4. Start the backend (automatically builds frontend)
python sigma_server.py
# 5. (Optional) Start frontend in dev mode
cd sigma_studio && npm run devThe backend is now live at http://localhost:8000 and the frontend at http://localhost:5173.
# Verify core modules
python -c "from core.sandbox import is_path_allowed; from core.ai_providers import load_ai_config; print('✅ System OK')"
# Verify API
curl http://localhost:8000/api/tasks
# Create an Ollama model from a manifesto
curl -X POST http://localhost:8000/api/create_model \
-H "Content-Type: application/json" \
-d '{"name": "sigma_architect", "modelfile": "FROM llama3.2\nSYSTEM \"\"\"You are a software architect...\"\"\""}'
# Verify Training Lab
curl http://localhost:8000/api/training/datasets/featured
# Verify Hardware Monitor
curl http://localhost:8000/api/hardware/statusSigma Studio supports 6 AI providers dynamically configurable via config.json:
| Provider | Type | Setup |
|---|---|---|
| Ollama 🦙 | Local (free) | http://localhost:11434 |
| DeepSeek 🔍 | Cloud API | API Key |
| OpenAI 🤖 | Cloud API | API Key |
| Anthropic (Claude) 🟣 | Cloud API | API Key |
| Groq ⚡ | Cloud API | API Key |
| OpenRouter 🌐 | Cloud API (multi-model) | API Key |
Sigma Principle: "A notification not left is an action that never happened."
| Mode | Backend Params | What It Does | Notifications? |
|---|---|---|---|
| 💬 Ask | allow_actions=false |
AI responds without modifying anything | None (chat only) |
| 📋 Plan | planning_mode=true |
Analyzes a goal and creates tasks in the Roadmap | ✅ Each task gets a creation notification |
| ⚡ Execute | allow_actions=true |
AI creates, modifies, or deletes files | ✅ Automatic: every file action generates a notification |
| ✅ Complete Task | execute_task_id + allow_actions=true |
Executes a specific task from the Roadmap and marks it done | ✅ Notifications for every action + completion marking |
AI agents aren't black boxes. They're defined by Ollama Modelfiles that specify:
- Identity: "You are a mathematician specialized in number theory..."
- Rules: "Never modify files outside
data/..." - Protocol: "Before acting, analyze the context..."
- Parameters: Temperature, context window, conversation template
| Agent File | Base Model | Version | Role |
|---|---|---|---|
sigma_architect.md |
llama3.2 / sigma:latest | v7.0 | Sigma Architect — admin, main orchestrator, research coordinator |
agente0.md |
sigma:latest | v7.2 | Enterprise AI Architect — extended version with full execution workflow |
code_architect.md |
sigma:latest | v1.0 | Full-Stack Developer — modifies React/Python code with backup & build check |
math1.md |
llama3.2 | v6.0 | Math Research Assistant — generates formal theory, proofs, exercises |
math-collatz.md |
llama3.2 | v1.0 | Collatz Math Specialist — number theory, mod 6 analysis, formal proofs |
test-engineer.md |
llama3.2 | v1.0 | Test Engineer — writes & executes scientific Python tests |
viz-designer.md |
llama3.2 | v1.0 | Visualization Designer — creates interactive D3.js charts |
proof-reviewer.md |
llama3.2 | v1.0 | Critical Reviewer — validates proofs, refutes incorrect claims |
All agent manifests are located in manifesti/ and can be loaded via the API or the Manifesti Gallery in the UI.
# 1. Create a manifesto file
cat > manifesti/my_agent.md << 'EOF'
FROM llama3.2
SYSTEM """
You are an agent specialized in molecular biology...
Rules:
- Only modify files within data/biology/
- Use exclusively Ollama provider for research
- Every discovery must generate a notification in tasks.json
"""
PARAMETER temperature 0.3
PARAMETER num_ctx 32768
EOF
# 2. Load the model into Ollama
curl -X POST http://localhost:8000/api/create_model \
-H "Content-Type: application/json" \
-d "{\"name\": \"my_agent\", \"modelfile\": \"$(cat manifesti/my_agent.md)\"}"Sigma Studio supports advanced multi-agent collaboration through dedicated endpoints:
- Parallel Orchestration: Assign tasks to multiple agents simultaneously via
/api/chat/orchestrate - Context Broker (
core/context_broker.py): SQLite-based shared context between agents, enabling agents to reference each other's work - Agent Registry (
core/agent_registry.py): Metadata management, agent templates, and color-coding for UI - Research Sessions (
core/research_sessions.py): Long-running autonomous research decomposition with progress tracking - Pipeline Engine (
core/pipeline_engine.py): DAG-based pipeline execution with status monitoring and stop/resume
All AI agent operations are strictly confined:
- Path Whitelist:
data/,manifesti/,sigma_studio/src/,scratch/,core/ - Module Structure: Only 5 subdirectories allowed per module:
teoria/,test/,viz/,docs/,whitepapers/ - Sandbox API: Create, run scripts, install packages, and destroy isolated environments
- Backup Manager (
core/backup_manager.py): Automatic backups before critical file modifications - Rollback Support: Undo changes via
/api/rollback
Sigma_Studio/
│
├── sigma_server.py ← Python Backend — REST API + AI orchestration
├── config.json ← Multi-provider AI configuration
├── config.example.json ← Example AI configuration template
├── .gitignore ← Ignores node_modules, .env, dist, data/
├── LICENSE ← GPL v3 / Commercial License
├── README.md ← This file
├── README_IT.md ← Italian documentation
│
├── core/ ← Backend modules (separation of concerns)
│ ├── sandbox.py ← Path validation & whitelist
│ ├── sandbox_manager.py ← Isolated environment management
│ ├── ai_providers.py ← Config + provider resolver + API calls
│ ├── api_router.py ← HTTP route mapping (80+ endpoints)
│ ├── chat_handler.py ← Chat orchestration + planning modes
│ ├── config_handler.py ← Config CRUD
│ ├── data_handler.py ← Data operations
│ ├── file_handler.py ← File CRUD + sandbox
│ ├── loop_handler.py ← AI action loop
│ ├── pipeline_engine.py ← DAG pipeline execution engine
│ ├── context_broker.py ← SQLite shared context between agents
│ ├── agent_orchestrator.py ← Multi-agent parallel orchestration
│ ├── agent_registry.py ← Agent metadata & template management
│ ├── agent_templates.py ← Agent template definitions
│ ├── module_handler.py ← Module CRUD
│ ├── task_handler.py ← Task CRUD + notifications
│ ├── plan_handler.py ← Planning & decomposition
│ ├── output_validator.py ← Output format validation
│ ├── execute_loop.py ← Autonomous execution loop
│ ├── research_sessions.py ← Long-running research sessions
│ ├── tool_registry.py ← Tool registration & dispatch
│ ├── backup_manager.py ← Automatic backup & rollback
│ ├── store.py ← Persistent state store
│ ├── logger.py ← Structured logging
│ ├── training_handler.py ← Training Lab — full training lifecycle (datasets, jobs, export)
│ ├── chat/ ← Chat sub-modules
│ └── orchestration/ ← Orchestration sub-modules
│
├── sigma_studio/ ← React 19 Frontend (Vite)
│ ├── index.html ← Entry point
│ ├── vite.config.js ← Proxy /api/* → localhost:8000
│ └── src/
│ ├── main.jsx ← React mount (StrictMode)
│ ├── App.jsx ← Global state orchestrator
│ ├── components/
│ │ ├── Sidebar.jsx ← Module tree navigation + Research/Training/Hardware badges
│ │ ├── Workspace.jsx ← Central tab system
│ │ ├── Dashboard.jsx ← Roadmap task cards
│ │ ├── Chat/ ← AI Chat (4 operating modes)
│ │ │ ├── core/ ← Hooks: useChatCore, useResearchPipeline, usePipelineDesigner
│ │ │ └── layouts/ ← Floating panel & workspace tab
│ │ ├── TrainingLab/ ← Training Lab (4 sub-tab: Docs, Dataset, Config, Monitor)
│ │ │ ├── TrainingLab.jsx ← Main Training Lab orchestrator
│ │ │ ├── TrainingDocs.jsx ← Training methods documentation
│ │ │ ├── DatasetBrowser.jsx ← HuggingFace search + local import
│ │ │ ├── TrainingConfigurator.jsx ← Model, method & hyperparameter configurator
│ │ │ └── TrainingMonitor.jsx ← Live logs, loss chart, Ollama export
│ │ ├── HardwareLab/ ← Hardware & GPU Monitor
│ │ │ └── index.jsx ← GPU telemetry, multi-GPU config, HF token setup
│ │ ├── SigmaLab/ ← Multi-format editor (Markdown, D3 viz, code runner)
│ │ └── Workspace/ ← Sub-components (ManifestiGallery, ModuleView, etc.)
│ └── styles/ ← Modular CSS (glass-morphism dark theme)
│
├── manifesti/ ← AI Agent manifests (Modelfiles)
│ ├── sigma_architect.md ← Main orchestrator
│ ├── sigma_admin.md ← System administrator
│ ├── sigma_assistant.md ← General assistant
│ ├── code_architect.md ← Full-stack developer
│ ├── math_researcher.md ← Math researcher
│ ├── test_engineer.md ← Test engineer
│ ├── viz_designer.md ← D3.js visualization designer
│ └── proof_reviewer.md ← Critical proof reviewer
│
├── images/ ← UI screenshots & agent portraits
├── data/ ← Knowledge Base (sandbox for AI agents)
├── training/ ← Training data (datasets, jobs, scripts)
│ ├── datasets/ ← Imported datasets (meta.json + data files)
│ ├── jobs/ ← Training jobs (scripts, output, logs)
│ └── scripts/ ← Training script templates
└── scratch/ ← Temporary workspace for experiments
| Method | Endpoint | Function |
|---|---|---|
GET |
/api/modules |
List modules with files by category |
GET |
/api/topics |
Topics with parent/child hierarchy |
GET |
/api/tasks |
Roadmap + notifications |
GET |
/api/tasks/by_agent |
Tasks filtered by agent |
GET |
/api/get_file?path=... |
Read file (sandbox-safe) |
GET |
/api/list_manifesti |
List available AI agents |
GET |
/api/knowledge_db |
Knowledge graph for D3 visualization |
GET |
/api/config |
AI config (without API keys) |
GET |
/api/ollama_models |
Installed Ollama models |
GET |
/api/sandbox/list |
List sandbox environments |
GET |
/api/agents |
List registered agents |
GET |
/api/agents/get |
Get specific agent details |
GET |
/api/agents/for_topic |
Get agents assigned to a topic |
GET |
/api/agents/templates |
List agent templates |
GET |
/api/agents/colors |
Get agent color mappings |
GET |
/api/chat/pipeline/status |
Pipeline execution status |
GET |
/api/context/get |
Get context broker data |
GET |
/api/context/chat_log |
Get chat log from context |
GET |
/api/research/list |
List research sessions |
GET |
/api/research/status |
Research session status |
GET |
/api/research/chat_history |
Research session chat history |
GET |
/api/training/datasets |
List imported datasets (Training Lab) |
GET |
/api/training/datasets/search?q=... |
HuggingFace dataset search (Training Lab) |
GET |
/api/training/datasets/featured |
Featured datasets (Training Lab) |
GET |
/api/training/jobs |
List training jobs (Training Lab) |
GET |
/api/training/job/status?job_id=... |
Training job status (Training Lab) |
GET |
/api/training/job/logs?job_id=...&offset=0 |
Training job logs (Training Lab) |
GET |
/api/training/hardware |
Hardware info for training (Training Lab) |
GET |
/api/hardware/status |
Full hardware status + config (Hardware Lab) |
| Method | Endpoint | Function |
|---|---|---|
POST |
/api/tasks |
Save tasks.json |
POST |
/api/tasks/assign |
Assign task to agent |
POST |
/api/create_file |
Create/overwrite file |
POST |
/api/delete_file |
Delete file |
POST |
/api/rename_file |
Rename file |
POST |
/api/create_module |
New module with subdirectories |
POST |
/api/delete_module |
Delete module |
POST |
/api/update_module |
Rename module |
POST |
/api/create_topic |
New research topic |
POST |
/api/update_topic |
Update topic metadata |
POST |
/api/delete_topic |
Delete topic |
POST |
/api/upload_file |
Multipart file upload |
POST |
/api/chat |
AI Chat (4 modes, multi-provider) |
POST |
/api/chat/loop |
Chat with action loop |
POST |
/api/chat/execute |
Execute chat actions |
POST |
/api/chat/plan |
Plan mode chat |
POST |
/api/chat/execute_plan |
Execute a saved plan |
POST |
/api/chat/orchestrate |
Multi-agent orchestration |
POST |
/api/chat/pipeline/start |
Start DAG pipeline execution |
POST |
/api/chat/pipeline/stop |
Stop pipeline execution |
POST |
/api/create_model |
Create Ollama model from Modelfile |
POST |
/api/run_test |
Execute Python/Node script |
POST |
/api/config |
Update AI provider config |
POST |
/api/ollama_models |
Refresh Ollama models |
POST |
/api/sandbox/create |
Create sandbox environment |
POST |
/api/sandbox/run |
Run command in sandbox |
POST |
/api/sandbox/install |
Install packages in sandbox |
POST |
/api/sandbox/destroy |
Destroy sandbox environment |
POST |
/api/agents/register |
Register a new agent |
POST |
/api/agents/update |
Update agent metadata |
POST |
/api/agents/create |
Create agent from template |
POST |
/api/agents/upload_image |
Upload agent avatar |
POST |
/api/context/share |
Share context between agents |
POST |
/api/context/chat_message |
Save chat message to context |
POST |
/api/research/create |
Create research session |
POST |
/api/research/delete |
Delete research session |
POST |
/api/research/update_objective |
Update research objective |
POST |
/api/research/update_agents |
Update agents in session |
POST |
/api/research/decompose |
Deploy & decompose objective |
POST |
/api/research/next_steps |
Get next research steps |
POST |
/api/research/start |
Start research execution |
POST |
/api/manifesti/update_image |
Update manifesto agent image |
POST |
/api/ai/action |
Generic AI action dispatcher |
POST |
/api/rollback |
Rollback last file operations |
POST |
/api/hardware/config |
Save multi-GPU configuration |
POST |
/api/config/hf_token |
Save HuggingFace Token |
POST |
/api/training/dataset/import |
Import local dataset |
POST |
/api/training/dataset/register_hf |
Register HuggingFace dataset |
POST |
/api/training/dataset/delete |
Delete dataset |
POST |
/api/training/job/create |
Create training job |
POST |
/api/training/job/start |
Start training job |
POST |
/api/training/job/stop |
Stop training job |
POST |
/api/training/job/delete |
Delete training job |
POST |
/api/training/export/ollama |
Export model → Ollama |
POST |
/api/training/dependencies |
Check training dependencies |
POST |
/api/training/job/clear_logs |
Clear training job logs |
Sigma Studio is licensed under a Dual License (GPL v3 / Commercial). We welcome contributions of all kinds:
- 🐛 Report bugs — open a GitHub issue
- 💡 Propose features — discussions and PRs welcome
- 🧠 Create new AI agents — Modelfile manifests are easy to write
- 📚 Add research topics — the modular structure makes it immediate
- 🎨 Improve UI/UX — React 19 components are well-isolated
- 🔧 Extend the backend — new API endpoints, providers, sandbox features
# Clone and install as above, then:
cd sigma_studio && npm run dev # Hot-reload frontend
python sigma_server.py # Auto-builds + serves on :8000core/— Python modules, each < 600 lines, single responsibilitysigma_studio/src/components/— React components, one file per componentsigma_studio/src/hooks/— Custom React hooks for data fetchingsigma_studio/src/styles/— Modular CSS files per sectionmanifesti/— AI agent Modelfile manifests (add new agents here)
This project is dual-licensed under:
- GNU GPL v3: For community, open-source, research, and educational use.
- Commercial License: For companies, proprietary products, and closed-source software.
SigmaStudio
|
+------------+------------+
| |
GPL v3 Commercial License
| |
Community, research Companies, closed-source
free paid
For inquiries regarding commercial licensing, please contact Diego Saitta. See the LICENSE file for the full licensing terms.
"A system is well designed when an AI can understand it without external instructions." "A theorem is not proven until it has been refuted, corrected, and refuted again." "A notification not left is an action that never happened." "Separate responsibilities, compose modules, maintain the sandbox." — Sigma Principles
⭐ If Sigma Studio changed the way you do research, leave a star on GitHub!