SIGMA Foundation

Agent Governance Platform (AGP)
The governance layer for intelligent agent ecosystems.

🚀 What is SIGMA?

SIGMA is an Agent Governance Platform (AGP) — the missing layer between foundation models and agent execution frameworks.

Foundation models provide intelligence. Agent frameworks provide execution. SIGMA provides governance.

Modern frameworks (CrewAI, LangGraph, OpenAI Agents, Google ADK) have made it trivial to build agents. But as organizations scale from 5 agents to 5,000, a new problem emerges:

How do you govern identity, lifecycle, audit, and runtime migration across thousands of agents?

SIGMA answers that.

🧬 Design Philosophy

SIGMA's architecture rests on a fundamental distinction: an agent is not just a model, nor just a harness. It is a triplet that combines three ontologically distinct layers:

Agent := ⟨ Model, SIGMA, Paradigm ⟩

Model: The reasoning engine (LLM, embeddings, etc.). It is interchangeable and does not define the agent's identity.
SIGMA: The governance platform that wraps the model. It is not a simple harness; it is the sum of five essential capabilities.
Paradigm: The ontological core that defines who the agent is. It answers "What am I?" and not "What can I do?".

This separation makes SIGMA agnostic to runtime and model, allowing the agent to retain its identity and governance even if the underlying execution engine or LLM is swapped.

The SIGMA Equation

SIGMA = Harness + Governance + Identity + Observability + Evolution

Component	What it governs
Harness	Execution, orchestration, and tool environment (MCP, A2A, sandboxing).
Governance	Policies, validation, and control: what the agent can do and when.
Identity	The agent's formal identity: its Paradigm, hierarchy level, trust level.
Observability	Complete traceability of every decision, from intent to outcome.
Evolution	Full agent lifecycle: birth, promotion, degradation, quarantine, migration, retirement.

⚡ Try the Pipeline Demo

Upload a CSV with a text column and watch SIGMA govern a complete sentiment analysis pipeline — with full traceability, human approval, and automated testing.

# Install the harness
pip install sigma-agp

# Run a governed pipeline
from sigma import Pipeline, GovernanceKernel

pipeline = Pipeline.from_yaml("sentiment-analysis")
kernel = GovernanceKernel()
result = kernel.run(pipeline, data="./tweets.csv")

# Every step is traced. Every decision is auditable.
print(result.trace_url)  # View in Langfuse

This demo runs on the Hito 1 Harness — the validated infrastructure that powers SIGMA's agent governance.

🏗️ Architecture

                    SIGMA FOUNDATION

        ┌──────────────────────────────────────┐
        │          SIGMA Studio (ARI)          │
        │      Visual Control & Dashboard      │
        └──────────────────────────────────────┘
                         │
                         ▼
        ┌──────────────────────────────────────┐
        │        SIGMA Core (ARP)              │
        │     Agent Runtime Platform           │
        │  Identity · Lifecycle · Governance   │
        │  Audit (AgDR) · Sandbox · Policies   │
        └──────────────────────────────────────┘
                         │
                         ▼
        ┌──────────────────────────────────────┐
        │        Runtime Adapters              │
        │   OpenAI · Claude · Gemini · Ollama  │
        │   · Azure · MCP · Future Adapters    │
        └──────────────────────────────────────┘

ARP = The governance kernel. Owns identity, lifecycle, AgDR, sandbox.
ARI = The visual interface. Inspect, configure, and govern without writing code.
Adapters = Connect any LLM or framework without changing your governance model.

"The model is the engine; the harness is the car, the road, and the traffic rules."

🔄 What SIGMA Enables

Runtime Migration at Scale

# Move 200 agents from GPT to Claude. One command.
# Identities, Agent Cards, AgDR history — all preserved.
# Only the Runtime Adapter changes.

Dynamic Task Redistribution

# One Engineer degraded after audit failures.
# SIGMA recalculates the Organizational Graph automatically.
# Tasks redistributed. Continuity preserved. Every action audited.

Governed Lifecycle

# Every agent evolves through explicit, auditable states.
# Creation → Deployment → Promotion → Degradation → Quarantine → Migration → Retirement

📊 Current Status

Milestone	Status
Hito 1: Governance Harness	✅ Complete — 6 skills, 65/65 tests, real data, full traceability
Hito 2 Rollout 1: Director + Engineer Data	✅ Complete
Hito 2 Rollout 2: Engineer Models	🔄 In Progress
Hito 2 Rollout 3: Engineer Auditor + Sandbox	📋 Planned
Hito 2 Final: Full Multi-Agent Governance	🎯 Target — Runtime independence, persistent identity, AgDR

The Harness is proven. The Agents are coming.

🗺️ Roadmap

Phase	When	What
Hito 1	✅ Done	Governance Harness — pipeline execution, tests, HITL, traceability
Hito 2	Q4 2026	Multi-agent governance — 4 agent types, runtime independence, AgDR
Hito 3	2027	Real-time streaming, Hamilton Selector, continuous governance
Enterprise	2027+	Federated governance, RBAC, SOC 2, on-premise

📚 Learn More

White Paper — The 5 pillars, architecture, and validation
GitHub — Code, ADRs, and the validated Hito 1 Harness
ADR Index — 23 Architecture Decision Records

Intelligence without governance does not scale.
Execution without identity cannot evolve.
Knowledge without traceability cannot be trusted.

SIGMA Foundation exists to govern the future of intelligent agent ecosystems.

SIGMA does not compete with CrewAI, LangGraph, or the OpenAI SDKs. It governs them. And that is its true innovation.

Built with 🧠 and disciplined governance by Prof. Marx Agustín García Delgado

Edit this README.md markdown file to author your organization card.