FAQ & Glossary — Gao AI OS
Frequently Asked Questions#
For Business Owners
Q: What can an AI agent actually do for my business on Gao Internet?
AI agents on Gao Internet can automate routine operational workflows — processing orders, responding to customer inquiries, generating reports, managing inventory notifications, and initiating payments within your pre-authorized limits. Every action the agent takes is logged and auditable. The agent cannot do anything you haven’t explicitly authorized.
Q: Can an agent spend my money without my approval?
No. Agents can only initiate payment requests up to the ceiling you define in their CAP token. The authorization model is tiered: Tier 0–1 actions proceed automatically; Tier 2 actions require your in-band confirmation; Tier 3 actions — including any transfer above your ceiling — pause execution and require your out-of-band signed authorization before proceeding. Agents never directly sign blockchain transactions — only you can do that.
Q: What happens if an agent makes a mistake?
Every action is logged in an immutable receipt. You can see exactly what the agent did, in what order, under which policy. If an action caused an unintended outcome, you can trace it to the exact execution and receipt. You can also revoke an agent’s capability grant at any time to immediately prevent future executions.
Q: Do I need to be technical to use a Seal SMB agent?
No. Seal agents are configured through the Gao Workspace no-code interface. You choose a template, set your authorization limits, and activate. No code required. Developers can extend templates for custom workflows.
Q: What if I lose access to my Gao Domain?
Domain recovery is handled at Layer 5 (Gao Domain) through guardian-configurable recovery. Recovering your domain automatically restores your authority over all agents bound to it.
For Developers
Q: Can I use any AI model with GAR v1?
GAR v1 is model-agnostic. You can use any AI model or framework as the underlying agent logic. The runtime enforces policy, capabilities, and receipts regardless of what model powers the agent.
Q: How do I handle long-running tasks that span multiple executions?
Use the Memory Store for state persistence between executions. Design your agent as a state machine that reads its last known state at the start of each execution window and writes updated state before the window closes. See the Memory & State Model for patterns.
Q: Is there a limit on how many agents I can run per domain?
There is no hard protocol limit on agent count per domain. Practical limits are governed by your DePIN compute allocation (L7) and the settlement ceiling defined in your CAP tokens. Contact support for enterprise-scale deployments.
Q: Can I integrate Gao AI OS with an existing agent framework (e.g., LangChain, AutoGen)?
Yes. You can wrap existing agent frameworks as custom tools within the Gao SDK. The GAR v1 runtime handles policy enforcement and receipts; your agent logic runs inside the sandbox. See Integration Guide for the tool registration API.
Q: How do receipts work with my existing audit systems?
Receipts are queryable via the SDK and available as JSON. You can stream receipt events to your SIEM, data warehouse, or compliance system. Receipt format is stable and versioned — breaking changes require a MAJOR version bump with 90-day migration window.
For Institutions & Compliance Teams
Q: Is Gao AI OS a financial service?
No. Gao AI OS is infrastructure software that provides an execution environment for AI agents. All value transfers are user-signed blockchain transactions routed through Gao Payment (L4). The AI OS layer does not custody funds, pool assets, or provide yield.
Q: How does Gao AI OS comply with data privacy regulations?
Memory Store data is domain-scoped and user-controlled. Domain owners control retention policies (minimum 90 days for settlement-related namespaces). Cross-domain data access requires explicit bi-lateral authorization. Gao Labs does not have access to domain-scoped memory contents.
Q: Where are agent execution logs stored?
Canonical receipts are written to an on-chain append-only ledger. Extended execution logs are stored in domain-scoped infrastructure with retention periods configured by the domain owner. The on-chain receipt ledger is always accessible regardless of Gao Labs’ operational status.
Q: Can governance change the behavior of my deployed agents?
Governance can adjust protocol-wide parameters — such as standard risk tier definitions, fee structures, and interface specifications — and these changes apply prospectively. Governance explicitly cannot modify individual CAP token grants, agent-level policies, completed receipts, or domain ownership. These restrictions are enforced at the runtime level, not merely by convention. Your agents continue operating under the policy you configured; governance has no mechanism to alter that without your participation as a domain owner.
Glossary#
Term
Definition
GAR v1
Gao Agent Runtime version 1 — the execution engine for Gao AI OS
CAP Token
Capability token — a signed grant defining an agent’s permitted action scope, settlement ceiling, risk tier, and expiration
Receipt
A canonical, immutable, cryptographically signed record of a single agent execution
Policy Snapshot
A hashed copy of the domain’s active policy, locked at the start of each execution window
Risk Tier
A 4-level classification (0–3) of action consequence that determines whether human authorization is required
Execution Window
A bounded time period during which a single agent execution occurs (default max: 5 minutes)
Seal SMB Framework
A pre-certified collection of agent templates for small and medium business workflows
GAB v2.5
Gao Agent Builder — the developer tooling for creating custom agent definitions
Memory Store
A domain-scoped, capability-gated key-value store for agent state persistence between executions
Sandbox
The isolated execution environment within GAR v1 where agent logic runs with restricted system access
Settlement Ceiling
The maximum value an agent can initiate for a single payment request before requiring explicit human approval
Domain-Scoped
Restricted to the context of a specific Gao Domain — cross-domain access requires explicit authorization
Policy Enforcement Gate
The pre-execution validation layer that checks capability validity, policy compliance, and risk tier before entering the sandbox
Staged Write
A memory write operation that is committed only after receipt emission — ensures no partial state is persisted from failed executions
Cross-Domain CAP
A bi-laterally signed capability grant allowing an agent in one domain to access resources in another domain
INV-0x
Notation for a runtime-level invariant (e.g., INV-01 through INV-08) that cannot be overridden by any external actor
DePIN
Decentralized Physical Infrastructure Network — Layer 7 of Gao Internet, which provides compute and bandwidth for AI OS execution
L4 / L5 / L7 / L8
Shorthand references to Gao Internet stack layers: Payment, Domain, DePIN, and AI OS respectively
Have a question not answered here? Reach out via the official channels listed on the Official Links page.