GIZ AI AGENTS ARCHITECTURE

Governed by enterprise transactional invariants, not mere prompt text.

Enterprise agent fabric architecture featuring Tri-Guard 3-stage safety boundaries, official Anthropic MCP Hub, and idempotent outbox transactions.

Architecture illustration
Giz Systems Architecture illustration

Architecture illustration

PRODUCT IN USE

Inspect the product in use.

Check the capture date and sample-data label, then select an image to inspect it. Operating performance, security requirements and customer outcomes are separate acceptance checks.

Giz Agent consultation example

Enlarge view ↗
Giz Agent consultation example

Giz Agent consultation example

An example of entering a business goal and reviewing a consultation response on the website. 2026-09-13 · Sample data.

Explore this capability →

Natural language prompts alone cannot safeguard enterprise data integrity

“Check the inventory shortage in Factory 2 and expedite 500 units from the nearest warehouse.” Executing this single sentence requires item unique identifiers, physical available-to-promise validation, transfer cost calculations, journal reversal paths, and authorized cryptographic signatures. Granting raw database write privileges based on prompt heuristics invites severe data corruption.

The Giz AI Agents fabric operates strictly within ontological transaction invariants and the Tri-Guard 3-stage safety boundary, orchestrating 109 enterprise tools through the global Anthropic Model Context Protocol (MCP) Hub.

4 Architectural Layers of the Enterprise Agent Fabric

Architecture Layer Core Component & SSOT Engineering Specification & Operational Role
1. Dialogue-Canvas Surface apps/giz/web/Agent.vue Real-time bidirectional context synchronization between chat composer and live workspace canvas
2. Tri-Guard Safety Governance apps/giz/model/services/agentTool.server.ts Enforces Read(Inspection) $\rightarrow$ Simulate(Scenario) $\rightarrow$ Approve(Signature) boundaries
3. Official Anthropic MCP Hub apps/giz/server/external/external.controller.ts Interoperates with Claude Code, ChatGPT, and Codex across 109 internal tools via /api/mcp
4. Idempotent Outbox Mesh apps/ontology/model/services/outbox.server.ts Guarantees exact-once semantics with Idempotency-Key headers and ERP receipts

Tri-Guard 3-Stage Safety Pipeline

  1. Inspection (Read & Reason): Under read-only permissions (risk: 'read'), explore ERP/SCM records and KM standards to analyze anomalies.
  2. Simulation (Simulate & Evaluate): Simulate supply shifts or alternative routings on virtual branches, calculating cost and schedule impacts.
  3. Approval & Execution (Action Execution): Actions exceeding delegation limits require human sign-off before committing to the outbox_events transactional ledger.

Giz AI Agents Solution · Sovereign AI · Giz Computer

YOUR NEXT STEP

Bring us the workflow that needs to work.

Start with the files and systems you use today, and the work you want to change.

Talk to our team ↗