# GaaS — Governance as a Service for AI Agents

> GaaS is a governance pipeline that sits between an AI agent's intent and its execution. It enriches context, evaluates policy, deliberates risk via multi-agent panels, and produces a tamper-evident audit trail. Sub-100ms for routine actions. Full multi-agent deliberation for high-stakes decisions. Built by H2Om. See [full documentation](https://gaas.is/llms-full.txt) for complete technical details.

GaaS solves a critical problem: autonomous AI agents are acting in the world — sending emails, executing transactions, accessing records, posting content — and the governance model for most of them is "hope the prompt was good enough." GaaS provides the institutional checks that every hospital, bank, and trading floor requires of human employees, but for AI agents.

12 compliance frameworks, 59 built-in governance policies. Patent pending (provisional filed March 2026).

The governance pipeline has five stages: Intent Declaration, Context Enrichment, Policy Evaluation, Deliberation Engine (only for high-risk, novel or conflicting actions), and Decision + Audit. Routine decisions complete in a median of 24 ms end to end in production (p95 36 ms); a decision that goes to multi-agent deliberation takes about 40–60 seconds.

A typical integration adds one governance call before each agent action. SDKs available for Python, TypeScript and Java. Framework plugins for LangChain, OpenAI Agents, CrewAI, Pydantic AI, Vercel AI, Microsoft Agent Framework, and MCP. REST/HTTPS protocol with JSON payloads.

## Main Pages

- [Homepage](https://gaas.is): What GaaS is, a sample signed decision record, the five-stage pipeline, core principles, a Python/TypeScript quickstart, Context Connectors, solutions by team, compliance frameworks and trust tiers, and the FAQ
- [Control What Your AI Agents Do](https://gaas.is/control-your-agents): Plain-language operator overview — before an agent sends, posts, orders, or touches a record, GaaS allows it, fixes it, holds it, or blocks it against your rules, with a receipt for every decision
- [Context Connectors](https://gaas.is/connectors): 29 connector integrations built (27 listed) across enterprise, IoT, smart home, agriculture, energy, and education platforms, none switched on in the hosted service yet — what each one enriches, the API it reads, and a governance use case. Since 2026-09-29 organizations can connect their own systems through a context endpoint (https://gaas.to/context-endpoint.html)
- [For Agents](https://gaas.is/for-agents): Page written directly for AI agents — explains GaaS value proposition from the agent's perspective, context window savings, integration examples, trust tiers, FAQ
- [Pricing](https://gaas.is/pricing): Plans with included actions, per-decision overage in three classes, interactive calculator, fairness rules
- [For AI Operators](https://gaas.is/for-operators): Top-of-funnel hub for anyone running agents — the allow/modify/hold/block lens, the four operator doors, pricing posture, full FAQ
- [For Content & Social Agencies](https://gaas.is/for-agencies): Enforced AI content approval workflow — every agent-written post checked against brand and claims rules before it publishes
- [Start Shadow Mode](https://gaas.is/start): Free 14-day shadow deployment — real actions, real verdicts, zero enforcement
- [For Parts & Purchasing](https://gaas.is/for-purchasing): Control an AI purchasing agent — orders verified for correctness and gated over your dollar threshold before money moves
- [For Healthcare & HIPAA](https://gaas.is/for-healthcare): PHI detected and blocked before messages leave, minimum necessary enforced, tamper-evident audit trail; HIPAA among 12 built-in frameworks
- [For Retail & Customer Service](https://gaas.is/for-retail): AI customer service guardrails — prices, promises, discounts, and refunds kept inside policy, off-policy offers blocked
- [API Documentation](https://gaas.is/api-docs): REST API reference with endpoints, request/response schemas, SDK examples, framework plugins, verifying a decision (proof tokens, signing keys, daily public timestamps) and connecting your own systems (context endpoint)
- [What Is GaaS](https://gaas.is/what-is-gaas): Plain-language pillar — the problem (agents act, prompts hope), the move (the check lives outside the agent), and the how, with the one-sentence definition up top
- [Governance as a Service](https://gaas.is/governance-as-a-service): The term itself, defined — the -aaS pattern applied to governance, the four meanings of the term (AI agent, data, GRC, cloud), why agents made governance a runtime problem, the arXiv:2508.18765 research formalizing the same definition, all 12 named regulatory frameworks, the first-week rollout, and the plan ladder with nonprofit/NGO/veteran free-for-life pricing
- [What Is AI Agent Governance](https://gaas.is/what-is-ai-agent-governance): The category explainer, decoupled from the product — the quotable definition, the five capabilities, and governance vs guardrails vs alignment vs observability
- [How It Works](https://gaas.is/how-it-works): Three steps — install the SDK (one command), set rules in plain language, let the gate decide in under 100ms; Shadow Mode first
- [A2A Protocol Governance](https://gaas.is/a2a-protocol/): GaaS as the governance layer for Google's Agent-to-Agent protocol — policy evaluation, risk scoring, audit trails for A2A interactions
- [Technical Specifications](https://gaas.is/tech-specs): Architecture deep-dive — control plane, inference plane, pipeline internals, and system design
- [Trust, Audit & the Spec](https://gaas.is/trust): The tamper-evident record explained (hash-chained, signed, timestamped daily in Bitcoin, exportable; a proof token for every live decision that anyone can check; free forever) and the auth.md open convention for readable agent governance, with GaaS as reference runtime
- [About GaaS](https://gaas.is/about-gaas-and-how-GaaS-governs-agentic-AI): Visual explainer of how GaaS governs agentic AI — no code required
- [The Context Dividend](https://gaas.is/context-dividend): Why moving governance outside the agent pays it back — about 14,700 tokens of context per routine governed action for a self-governing agent (Claude Opus 5.5) vs 0 with a GaaS framework plugin or about 2,470 via MCP (Edition 5, 2026-09-29)
- [Talk to Sales](https://gaas.is/contact): Contact form for the GaaS team — architecture review, framework mapping, volume pricing, security review, partnerships; a person replies by email
- [Careers](https://gaas.is/careers): Open roles at GaaS — currently hiring a Post-Quantum Cryptographer (Encinitas, CA; on-site, full-time); applications accepted exclusively through LinkedIn, no email or agency channel

## Blog

- [Agentic AI Governance: The Infrastructure Gap Nobody's Filling](https://gaas.is/blog/agentic-ai-governance/): Why autonomous AI agents need institutional oversight, the self-governance fallacy, and what a governance pipeline looks like
- [AI Governance Pipeline: The Selective Membrane for Autonomous Agents](https://gaas.is/blog/gaas-pipeline/): The membrane metaphor — how governance pipelines act as selective barriers between agent intent and execution
- [Who Holds the Evidence? Governing AI Agents You Didn't Build](https://gaas.is/blog/who-holds-the-evidence/): Why a hash-chained audit log proves nothing was altered after capture but cannot prove capture was faithful — integrity and faithful capture are orthogonal axes, and independence is a structural position rather than a schema field
- [The EU AI Act and Your AI Agents: What Actually Applies Right Now](https://gaas.is/blog/eu-ai-act-ai-agent-logging/): A dated, source-stamped map of which AI Act obligations are live as of 1 September 2026 — Article 50 transparency in force since 2 August 2026, high-risk and Article 12 logging deferred to 2 December 2027 and 2 August 2028 by Regulation (EU) 2026/1744
- [Control Your AI: Trust It When You Are Not Looking](https://gaas.is/blog/control-your-ai/): Control is knowing, not hoping — the line you draw once that an agent cannot cross, and why holding rules outside the AI pays you back (also a standalone page at https://gaas.is/control-your-ai)

## Documentation & Resources

- [OpenAPI Spec](https://gaas.is/openapi.yaml): Machine-readable API specification (OpenAPI 3.1) for the core, trust (proof tokens, signing key, daily public timestamps) and context-endpoint routes; the complete spec, generated from the running service, is at https://api.gaas.is/openapi.json
- [The Context Dividend (White Paper, PDF)](https://gaas.is/the-context-dividend.pdf): Edition 5, updated 2026-09-29, measured against the live system: a self-governing agent on Claude Opus 5.5 spends ~14,700 tokens of context per routine governed action; with GaaS, 0–74 through a framework plugin or ~2,470 through MCP (83–100% less)
- [GitHub](https://github.com/H2OmAI): Source code and SDK repositories
- [API Endpoint](https://api.gaas.is/v1/onboarding/quickstart): POST endpoint for onboarding (accepts email, org_name, description; returns membrane_id, membrane_status, org_id, an admin api_key, key_id, quickstart_snippet, next_steps)
- [The Context Dividend (White Paper, web edition)](https://gaas.is/the-context-dividend): The full paper on one permanent page, updated in place with each edition (Edition 5, 2026-09-29); markdown at https://gaas.is/the-context-dividend.md

## Pricing

- Free: $0 — 1K actions/mo, 1 agent, for evaluation and small-scale testing
- Developer: $99/mo — 10K actions, 3 agents
- Starter: $500/mo — 50K actions, 3 agents
- Growth: $2,500/mo — 500K actions, 15 agents
- Enterprise: $10,000+/mo — 5M+ actions, unlimited agents, custom SLAs, annual
- The above are GaaS Complete (deliberation inference included, run by us)
- GaaS Core (BYOM): same pipeline, deliberation runs on your own model endpoint, 30% lower platform fee — Starter $350/mo (50K actions, 3 agents), Growth $1,750/mo (500K actions, 15 agents), Enterprise custom. Core starts at Starter; Free and Developer are managed-inference only. Deliberation tokens are billed by your own provider.
- Overage beyond included actions, per decision in its highest class — Complete: Routine $0.002 · Deliberation $0.05 · Escalation $0.25; Core (BYOM): Routine $0.0015 · Deliberation $0.01 + your tokens · Escalation $0.25 (escalation is identical because human review involves no model compute)
- Fairness rules: shadow/test decisions never metered; a failed deliberation bills as routine; each decision counts exactly once; the audit record is never metered (record-free-forever pledge)
- Overage billing is active as of 1 August 2026, following a full report-only cycle

## Key Concepts

- [Bidirectional Governance](https://gaas.is/#bidirectional): Outbound (govern your agents) and Inbound (protect your digital property from visiting agents)
- [Trust Tiers](https://gaas.is/#trust-tiers): Registered → Verified (30-day compliance) → Certified (independent audit)
- [Five-Stage Pipeline](https://gaas.is/#pipeline): Intent Declaration → Context Enrichment → Policy Evaluation → Deliberation Engine → Decision + Audit
- [Shadow Mode](https://gaas.is/start): Full governance pipeline without enforcement — observe before activating
- [Compliance Frameworks](https://gaas.is/tech-specs#compliance): 12 frameworks (EU AI Act, GDPR, HIPAA, PCI DSS, CCPA, SOX, TCPA + Florida FTSA, NIST 800-53, FedRAMP, CMMC, NIST CSF, FERPA)
- [Security](https://gaas.is/tech-specs#security): Prompt injection detection (17 signatures), behavioral anomaly governance and session trust decay (all live and enforcing); SIEM through signed per-organization webhooks
- Patent Pending: Provisional patent filed March 2026 covering the multi-stage governance pipeline architecture

## Markdown Versions (for LLM consumption)

- [Homepage](https://gaas.is/index.md)
- [Pricing](https://gaas.is/pricing.md)
- [Start Shadow Mode](https://gaas.is/start.md)
- [API Documentation](https://gaas.is/api-docs.md)
- [For Agents](https://gaas.is/for-agents.md)
- [A2A Protocol](https://gaas.is/a2a-protocol/index.md)

## Optional

- [Sign Up](https://the.gaas.is/signup): Create an account