| Internet-Draft | Agent Gateway Scenario Analysis and Func | June 2026 |
| Miao, et al. | Expires 28 December 2026 | [Page] |
This document analyzes the communication gaps in three representative multi-agent scenarios: single-domain single-agent, single-domain multi-agent, and cross-domain multi-agent. For each scenario, it examines whether an Agent Gateway is needed, what capabilities it should provide, and where it should be deployed. Based on the consolidated gap analysis, this document derives a set of universal functional requirements for the Agent Gateway as a standardized cross-domain trust boundary and protocol termination layer. The findings aim to guide the design of interoperable agent communication infrastructure.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 28 December 2026.¶
Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
As agent technology evolves from single-device intelligence to multi-device collaboration and cross-domain interconnection, there is a growing demand for multi-agent communication across home, vehicle, and enterprise campus scenarios. Typical use cases include:¶
Multiple sub-agents (lighting, security, HVAC) within a home collaborating to execute an "arrive home" scene;¶
A vehicle agent communicating with a user home agent across domains to enable "auto-arm when leaving, pre-condition when approaching";¶
Agents from different vendors (e.g., User Home Agent and 3rd-party Vehicle Agent) needing to cooperate across trust domains.¶
However, current agent communication architectures lack a standardized cross-domain trust boundary and protocol termination layer. This document aims to analyze three representative scenarios (single-domain single-agent, single-domain multi-agent, and cross-domain multi-agent), identify gaps in existing solutions, and argue for the necessity, capability requirements, and deployment options of an Agent Gateway as a universal infrastructure component.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
A home has a single Agent managing multiple smart devices (physical nodes). For example:¶
A central hub (Node A) runs the Supervisor Agent;¶
A smart speaker (Node B) serves as the voice interaction interface;¶
A ZigBee gateway (Node C) connects ZigBee lights and sensors.¶
The Supervisor Agent communicates with each node via a soft bus (e.g., DSoftBus, local MQTT) and provides a unified service externally. Users issue commands via voice or app; the Supervisor Agent parses the intent and dispatches tasks to the appropriate node. Naturally, if all household devices are assigned IP addresses, a mesh network can be established over the home network. Furthermore, the aforementioned devices can communicate directly via peer-to-peer (P2P) protocols.¶
No. In this scenario, all communication occurs within the same trust domain (home LAN). The Supervisor Agent can interact directly with each node. There is no cross-domain security boundary. External access (e.g., mobile app) typically goes through the vendor cloud, which is outside the scope of Agent Gateway.¶
*-------------------------------------------------------------------------* | Issue | Description | *-------------------------*-----------------------------------------------* | Lack of standardized |The Supervisor Agent's capabilities (Skills) | | external interface |have no standard way to be exposed externally; | | |apps must use vendor-specific private APIs | *-------------------------*-----------------------------------------------* | Node discovery and |Nodes register via soft bus, but registration | | registration rely on |format and heartbeat mechanisms are | | proprietary protocols |non-standard | *-------------------------*-----------------------------------------------* | Insufficient offline |Some solutions rely on cloud-based intent | | autonomy |parsing, degrading local orchestration | | |when disconnected | *-------------------------*-----------------------------------------------*¶
Note: Gaps in this scenario are more about "external exposure" and "standardization," but introducing an Agent Gateway is not strictly necessary; these gaps could be addressed by enhancing the Supervisor Agent itself.¶
A home contains multiple independent agents, each managing a different subsystem, requiring inter-operation. For example:¶
Lighting Agent: manages all lighting devices;¶
Security Agent: manages door locks, cameras, alarms;¶
Environment Agent: manages HVAC, curtains, air purifiers.¶
These agents may run on different hardware nodes (central hub, smart speaker, dedicated gateway). They need to discover each other's capabilities, subscribe to events, and collaboratively execute scenes (e.g., "away mode": Security Agent arms, Lighting Agent turns off lights, Environment Agent switches to energy-saving mode).¶
Yes, but with a different role than in the cross-domain scenario. Here, the Agent Gateway should act as an intra-domain registry and task routing layer, rather than a cross-domain security boundary.¶
Intra-domain Agent Card Registration and Discovery: Each agent registers its Skills and capability description (Agent Card) with the Gateway; the Gateway maintains a domain-wide capability index.¶
Intra-domain Task Routing: When Agent A needs to invoke a Skill of Agent B, the request goes through the Gateway, which handles addressing and optional load balancing.¶
Event Bus: Support publish/subscribe model so agents can subscribe to events from other agents (e.g., security alarm triggered �� lighting turns on).¶
Local-first: All communication should complete within the LAN, without depending on public internet connectivity.¶
*-------------------------*-----------------------------------------------* | Issue | Description | *-------------------------*-----------------------------------------------* | No standard inter-agent |Agents discover each other via proprietary | | discovery |soft bus or hardcoded addresses; no unified | | |registry | *-------------------------*-----------------------------------------------* | Task routing depends on |Agent A must know Agent B's address and | | point-to-point |protocol,creating tight coupling | | connections | | *-------------------------*-----------------------------------------------* | No standardized event |Event formats and subscription mechanisms | | subscription |vary across vendors, hindering cross-brand | | |collaboration | *-------------------------*-----------------------------------------------* | Unclear intra-domain |No unified authentication between agents; a | | security boundary |malicious agent could impersonate another | *-------------------------*-----------------------------------------------*¶
Two or more agents from different trust domains need to communicate across domain boundaries. Typical use cases:¶
Home Domain <-> Vehicle Domain: As the owner drives home, the vehicle agent sends an "estimated arrival in 10 minutes" event to the home agent, which pre-conditions the HVAC, turns on lights, and disarms the security system.¶
Home Domain <-> Cloud LLM Agent: A user queries home status or issues commands via a cloud-based LLM agent, which invokes Skills exposed by the home agent.¶
Home Domain A <-> Home Domain B (neighbor assistance): Neighbor A's security agent detects an anomaly and notifies neighbor B's security agent to increase vigilance.¶
Strongly needed. Cross-domain scenarios face fundamental challenges including NAT traversal, trust establishment, protocol translation, and information hiding. An Agent Gateway is an essential cross-domain trust boundary and protocol termination layer.¶
External A2A Endpoint: Expose standardized endpoints, such as for /.well-known/agent.json and POST /a2a/tasks, supporting HTTP/HTTPS.¶
Authentication and Authorization: for example, Support OAuth2 Device Flow, JWT issuance and verification, and Scope whitelisting.¶
Aggregated Agent Card: Merge all intra-domain Skills into a single Card for external exposure, hiding internal topology.¶
Reverse Channel Management: Accept reverse WebSocket/MQTT connections from the intra-domain Supervisor Agent for task dispatch and artifact return.¶
Cross-domain Federation: Establish mutual trust with Gateways in other domains, exchange aggregated Cards, and enable cross-domain discovery and Task Proxy.¶
HITL (Human-in-the-Loop): For high-risk Skills (e.g., unlock door), support suspending the Task and pushing a confirmation request to the user's app.¶
Audit Logging: Record the full chain of cross-domain calls (issuer, skill_id, status, timestamp).¶
Home network edge: Deployed on the home gateway (router/ONT) or a standalone edge computing device, serving as the sole external entry point for the home domain.¶
Cloud: If the home lacks public reachability, deploy on the vendor cloud as a Relay Gateway; the home Supervisor Agent connects via a reverse channel.¶
Hybrid mode: Cloud Gateway acts as global Registry and Auth Broker; Edge Gateway performs local A2A termination (when publicly reachable).¶
*-------------------------*-----------------------------------------------* | Issue | Description | *-------------------------*-----------------------------------------------* | Cross-domain discoverye |mDNS/BLE limited to LAN; cloud-to-cloud | | unavailable |integration costly; no standardized Agent | | |Card format | *-------------------------*-----------------------------------------------* | No standard NAT |Direct connection requires public IP/DDNS; | | traversal solution |reverse channel protocols are proprietary | *-------------------------*-----------------------------------------------* | No standardized event |Event formats and subscription mechanisms | | subscription |vary across vendors, hindering cross-brand | | |collaboration | *-------------------------*-----------------------------------------------* | Cross-domain trust |No unified Federation trust bootstrap process; | | establishment is |Scope models inconsistent | | difficult | | *-------------------------*-----------------------------------------------* | No standard HITL |Actions like unlocking doors lack a | | mechanism for high-risk |standardized human-in-the-loop confirmation | | operations |workflow | *-------------------------*-----------------------------------------------* | No cross-domain |External agents require multiple protocol | | semantics for task |translations to reach the target sub-agent | | routing | | *-------------------------*-----------------------------------------------* | Audit logs scattered |Cross-domain call chains cannot be traced | | |end-to-end | *-------------------------*-----------------------------------------------*¶
Based on the analysis of the three scenarios above, common gaps across current agent communication architectures can be summarized as follows:¶
*-------------------------*-----------------------------------------------*-------------* | Function | Description | Applicable | | | | Scenarios | *-------------------------*-----------------------------------------------*-------------* | Agent Card Registration |Receive capability registrations from | | | & Aggregation |intra-domain agents; merge into an aggregated |2, 3 | | |Card; expose via standardized endpoint | | *-------------------------*-----------------------------------------------*-------------* | A2A Protocol Termination|Support standard A2A Task interface (POST | | | |/a2a/tasks); translate external requests into |3 | | |internal commands | | *-------------------------*-----------------------------------------------*-------------* | Authentication & |Support JWT/OAuth2; verify caller identity and |3 (optional | | Authorization |Scope whitelist |for 2) | *-------------------------*-----------------------------------------------*-------------* | Reverse Channel |Accept reverse WebSocket/MQTT connections from | | | Management |intra-domain Supervisor Agent for task dispatch|3 | *-------------------------*-----------------------------------------------*-------------* | Task Routing |Route Tasks to the correct intra-domain agent | | | |or cross-domain Gateway based on Skill ID |2, 3 | *-------------------------*-----------------------------------------------*-------------* | Information Hiding |Do not expose sub-agent count, IP addresses, | | | |protocols, or other internal details externally|3 | *-------------------------*-----------------------------------------------*-------------* | HITL Trigger |Mark high-risk Skills with | | | |require_human_approval; suspend Task and wait |3 | | |for user confirmation | | *-------------------------*-----------------------------------------------*-------------* | Audit logging |Record full chain of cross-domain calls |3 | *-------------------------*-----------------------------------------------*-------------*¶
Core conclusion: As agents evolve from single-node to multi-node and from intra-domain to cross-domain operation, the lack of a standardized trust boundary and protocol termination layer has become a primary bottleneck. The Agent Gateway is the key infrastructure component to fill this gap.¶
*-------------------------*-----------------------------------------------*-------------* | Function | Description | Applicable | | | | Scenarios | *-------------------------*-----------------------------------------------*-------------* | Intra-domain Event Bus |Support publish/subscribe between intra-domain | | | |agents |2 | *-------------------------*-----------------------------------------------*-------------* | Federation |Exchange aggregated Cards with other domain | | | Synchronization |Gateways for cross-domain discovery |3 | *-------------------------*-----------------------------------------------*-------------* | Local Caching & Offline |Cache aggregated Cards and authorization | | | Autonomy |policies; handle authorized local Tasks when |2, 3 | | |disconnected | | *-------------------------*-----------------------------------------------*-------------*¶
Single-domain multi-agent scenario: The Agent Gateway can be deployed on an intra-domain central hub device, collocated with or separate from the Supervisor Agent.¶
Cross-domain multi-agent scenario: The Agent Gateway should be deployed at the home network edge or in the vendor cloud, serving as the sole external entry point.¶
Hybrid deployment: Cloud Gateway acts as global Registry and Auth Broker; Edge Gateway performs local A2A termination, balancing reachability and low latency.¶
The Agent Gateway, as the sole entry point for cross-domain communication, must enforce strict security measures:¶
All external interfaces MUST use TLS 1.3 encryption.¶
Authentication tokens SHOULD have a limited validity period (recommended no more than 24 hours) and support refresh.¶
Cross-domain requests MUST pass mTLS or Federation-level cascading trust verification.¶
Highly sensitive operations (e.g., door lock control) MUST enable the HITL mechanism, and HITL confirmation requests SHOULD be delivered to the user through an independent channel (e.g., mobile push notification).¶
All gateway components SHOULD record complete audit logs, including request source, target, operation, result, and timestamp. Logs MUST be tamper-proof.¶
TBA.¶
TBA.¶