The problem Zuul Gateway solves

On loan from the gates of Gozer, Zuul now guards yours — enforcing identity, permissions, and audit for AI agents. There is no leak. Only Zuul.

AI agents are proliferating. Keys scattered everywhere. Zero visibility into who's accessing what.

Prompts are not policies. You can't tell an agent "don't leak" and expect enforcement. Prompt injection, jailbreaks, bad actors. You can't prevent them all.

You can't secure every agent. But you CAN protect what they access. Protect the castle, not the horde.

No trail, no accountability. No accountability, no trust. If you can't prove what happened, you can't hold anyone responsible.

Today there's no unified layer for agent access governance:

• Identity: who is this agent?
• Permissions: what can it do?
• Audit: what did it do?

Zuul is that layer. An immutable record is the foundation of trust.

Use of AI tools and agents

AI tools used in development:

Claude: Ideation and product refinement. Architecture decisions, feature scoping, and iterative design feedback throughout the project.

Claude Code: PRD development, implementation planning, and code-level execution. Primary tool for translating product vision into working software.

Codex: Testing verification. Used to validate implementation against specs and catch edge cases.

Grok: Marketing feedback loop. Messaging refinement, positioning validation, and copy iteration alongside Claude.

Nano Banana: Image generation for marketing materials and visual assets.

Gamma: Pitch deck creation. AI-generated slide structure and layout from detailed prompts.
Workflow:

Ideation → Product refinement (Claude) → PRD → Implementation plans (Claude Code) → Code execution (Claude Code) → Testing verification (Codex) → Marketing feedback loop (Grok + Claude) → Marketing materials (Nano Banana + Gamma)

What Zuul governs:

Zuul itself is infrastructure for AI agents. It provides identity, permissions, and audit for autonomous agents accessing external services. Agents authenticate via wallet signature, permissions are enforced on-chain, and every action is logged immutably.