GEIANT — Geospatial AI Governance Runtime

Every AI agent action, cryptographically traceable to the human who authorized it.

GEIANT is an open-source runtime that gives AI agents verifiable identity, jurisdictional boundaries, and an immutable audit trail — so enterprises can deploy autonomous agents and prove compliance with the EU AI Act, GDPR, and FINMA regulations.

The Problem

AI agents are operating without identity. When an agent processes medical data in Frankfurt, trades securities in Zurich, or classifies satellite imagery over Rome — there is no cryptographic proof of:

• Who authorized it (which human principal?)
• Where it operated (which jurisdiction?)
• What it did (which tools, which data?)
• Whether it was allowed to (delegation scope?)

Existing solutions (OAuth tokens, CloudTrail logs, service accounts) fail three critical tests:

1. Delegation chains are not auditable — you can't trace a service account back to the human who approved the agent's creation
2. No pre-flight jurisdiction enforcement — logs record violations after the fact, not prevent them
3. Mutable logs — CloudTrail entries can be deleted; breadcrumb chains cannot

The Solution

GEIANT provides three primitives that existing infrastructure lacks:

1. Agent Identity (Ed25519)

Every agent gets a cryptographic keypair. The public key is the identity — no usernames, no passwords, no OAuth tokens. Same key signs breadcrumbs, delegation certificates, and Stellar transactions.

import { generateAgentIdentity } from '@gns-aip/sdk';

const agent = generateAgentIdentity();
// agent.publicKey  → "c14094ea7efb6122..." (64 hex chars)
// agent.stellarAddress → "GCAU..." (derived from same key)

2. Delegation Certificates

A human principal signs a certificate that says: "This agent may operate within these H3 cells, using these capabilities, until this date." Every agent action is checked against this certificate before execution.

import { createDelegationCert } from '@gns-aip/sdk';

const cert = await createDelegationCert({
  agentIdentity: agent.publicKey,
  principalIdentity: human.publicKey,
  territoryCells: ['851e8053fffffff'], // Rome
  facetPermissions: ['energy'],
  validityHours: 720, // 30 days
}, human.secretKey);

3. Virtual Breadcrumbs

Every tool call produces a signed, hash-chained breadcrumb — an immutable record of what the agent did, where, when, and under whose authority. Breadcrumbs are SHA-256 chained: tamper with one and every subsequent block is invalid.

import { createVirtualBreadcrumb } from '@gns-aip/sdk';

const crumb = await createVirtualBreadcrumb({
  agentIdentity: agent.publicKey,
  operationCell: '851e8053fffffff',
  meta: {
    operationType: 'perception_weather',
    delegationCertHash: cert.certHash,
    facet: 'energy',
    withinTerritory: true,
  },
}, agent.secretKey, previousBlock);

Who Is This For?

Audience	Use Case
Enterprise AI teams	Deploy agents with auditable delegation chains for EU AI Act compliance
AI framework authors	Add identity + jurisdiction to LangChain, CrewAI, AutoGen agents
GRC/compliance teams	Runtime evidence generation (not just documentation) for Art. 12/14
Infrastructure providers	Embed agent governance into cloud platforms and edge networks

Architecture

GEIANT is built on four layers:

Layer	Name	Purpose
0	H3 Grid	Uber's hexagonal geospatial indexing — 569 trillion addressable points
1	GEP	Geographic addressing fabric — open infrastructure
2	Agent Identity Protocol	Delegation, jurisdiction, breadcrumbs, trust tiers
3	Perception Runtime	Satellite imagery, weather, embeddings — MCP-native

→ Architecture deep dive

Get Started

The fastest path: install the SDK and have a working agent identity in under 10 minutes.

→ Quick Start Guide