--- schema_version: "1" id: b635c213699e633eee98367f7d42b69bdb86af36e2eb8cfeab62777a85bb2a2d title: Multi-Agent Orchestration collection: admin version: current locale: en-us portal_url: https://docs.salesforce.com/admin/current/en-us/ai/264-0-0/agent_multi_orch.md canonical_url: https://help.salesforce.com/s/articleView?id=ai.agent_multi_orch.htm&release=264.0.0&type=5 source_site: help.salesforce.com content_type: documentation content_hash: 7eed3bae7b842b0545ee9dd27e26d8cac35a0a35820982c469d992be0ea9b1d8 description: Extend your agent’s capabilities with Multi-Agent Orchestration for Agentforce. Connect your Agentforce agents with other specialized Agentforce agents in your Salesforce org so that they can collaborate securely on complex tasks. With multi-agent orchestration, businesses can deploy a single agent across their channels with shared context and without customer friction. products: - Agentforce - Cross Cloud release: "264" --- # Multi-Agent Orchestration Extend your agent’s capabilities with Multi-Agent Orchestration for Agentforce. Connect your Agentforce agents with other specialized Agentforce agents in your Salesforce org so that they can collaborate securely on complex tasks. With multi-agent orchestration, businesses can deploy a single agent across their channels with shared context and without customer friction. ### Required Editions
Available in: Lightning Experience
Available in: Enterprise, Performance, Unlimited, and Developer Editions. Required add-on licenses vary by agent type.
![Note](https://sf-zdocs-cdn-prod.zoominsoftware.com/tdta-ai-generative_ai-264-0-0-production-enus/f8ef6498-d7b2-44e3-9b34-ffd5a850aae9/images/icon_note.png) Note Beginning in April 2026, agent topics are now called subagents. There are no changes to functionality. During this transition, you may see a mix of the new and previous terms in our documentation. ## Multi-Agent Solutions Suppose you've built a great customer service agent and now your team wants to add a field service use case. Or you're currently running three agents in different business units and the business requires that they talk to each other. Or you're staring at an Instructions block that's grown so long that it covers six different domains, and you're not sure why the routing keeps breaking. Whatever the trigger, you've arrived at the same problem. At some point, one agent is not enough. But expanding to multiple agents presents a challenge. If you have multiple agents and Messaging for In-App and Web (MIAW), end users are required to navigate a growing list of agents, figure out which one to talk to, and repeat themselves anytime they switch over from one to another. If a handoff goes wrong, engineers must untangle a complex web of routing decisions to get to the root cause. To get the best of both worlds, you want a unified agentic experience that can reliably tackle the full breadth of what customers can throw at it. Multi-Agent Orchestration for Agentforce provides the interoperability framework you can use to extend your agent’s capabilities. It also provides a single front door "superagent" that can interact with customers. Here are some supplemental blog posts: - _Salesforce Blog post_: [The Architect’s Guide to Agentforce Interoperability](https://www.salesforce.com/agentforce/resources/interoperability-guide/) - _Salesforce Blog post_: [Scaling Beyond One Agent? Here’s Where Most Teams Get Stuck](https://www.salesforce.com/blog/multi-agent-orchestration/) - _Salesforce Blog post_: [3 Signs You’ve Outgrown a Single Agent](https://www.salesforce.com/blog/3-signs-youve-outgrown-a-single-agent/) ## Orchestrator Agents and Connected Subagents Multi-agent solutions include two key roles: - An orchestrator agent handles the initial conversation with users, then routes and delegates tasks. - A connected subagent is an entirely separate agent in your Salesforce org that your orchestrator agent can use like a subagent. An orchestrator agent can route a conversation to any combination of subagents and connected subagents to get the job done. ![Note](https://sf-zdocs-cdn-prod.zoominsoftware.com/tdta-ai-generative_ai-264-0-0-production-enus/f8ef6498-d7b2-44e3-9b34-ffd5a850aae9/images/icon_note.png) Note In Multi-agent Orchestration for Agentforce solutions, subagent actions consume Flex credits in the orchestrator agent and in connected subagents. No additional costs are incurred when the orchestrator agent routes to a connected subagent or synthesizes responses. See [Flex Credits Billable Usage Types](https://help.salesforce.com/s/articleView?id=ai.usage_flex_credits.htm&release=264.0.0&type=5 "Using Agentforce and embedded AI features can impact credit consumption."). ## Connected Subagents and Standard Subagents Connected subagents differ from standard subagents added from the Asset Library. - Subagents are localized components that simply contain the instructions (which control logic and reasoning) and actions (the tools) available for the primary agent to use. Subagents belong to the agent definition. - Connected subagents represent a complete, independent agent with its own distinct expertise and identity. A connected subagent contains one or more subagents of its own. Connected subagents are modular, represent their own domain, and can be well suited for reuse across many multi-agent solutions. You can build connected subagents from the AEA or ASA template. Collectively, the mix of internal subagents and external connected subagents assigned to your orchestrator defines the full scale of capabilities your agent can handle. In the context of an orchestrator agent, a reference agent is the agent upon which your connected subagent is based. When you add a connected subagent, it contains settings from the reference agent (such as its description) that you can override in the Canvas to customize its behavior. ## Handoff and Supervisor Modes There are two main orchestrator routing patterns in Agentforce multi-agent solutions. Both handle single-intent and multi-intent queries across multiple connected sub agents. You decide which pattern to use in your multi-agent solution architecture. Handoff Mode (Default) The orchestrator agent hands over control to a connected subagent, which handles the request and takes over the conversation by responding directly to the user. ![In handoff mode, the connected subagent responds to the user directly.](https://sf-zdocs-cdn-prod.zoominsoftware.com/tdta-ai-generative_ai-264-0-0-production-enus/f8ef6498-d7b2-44e3-9b34-ffd5a850aae9/generative_ai/images/agent_multi_orch_handoff.png) This pattern, which is the default, provides a faster response to the user by bypassing the orchestrator agent’s response synthesis step. The orchestrator agent sees the response but never synthesizes it. In handoff mode, multi-intent queries occur between one connected subagent and multiple subagents. Supervisor Mode In this pattern, the orchestrator agent is solely responsible for managing conversations directly with users. ![In supervisor mode, the orchestrator agent delegates the request to the connected subagent to obtain the result, then synthesizes and delivers the response to the user.](https://sf-zdocs-cdn-prod.zoominsoftware.com/tdta-ai-generative_ai-264-0-0-production-enus/f8ef6498-d7b2-44e3-9b34-ffd5a850aae9/generative_ai/images/agent_multi_orch_supervisor.png) This pattern uses subagent classification to route the request (not the conversation) to a connected subagent. Routing is based on the connected subagent's description and any instructions in the Agent Router. - It retains context throughout the delegation. - It synthesizes the connected subagent's response and then conveys that information to the user. In this pattern, you can think of an orchestrator agent as the single “front door” that interacts with the user. It also coordinates efforts behind the scenes, delegating work to the various “back door” connected subagents, and communicating the results back to the user. Which is the Best Pattern To Use? That depends on your solution requirements. - Handoff mode is best for connected subagents with specialized domain expertise that require managing direct, end-to-end conversations with the user. For example, a billing or HR agent. - Supervisor mode is best for jobs in which the orchestrator agent requires synthesizing results from multiple agents before replying to the user. For AgentScript examples, see [Connected Subagent Block](https://developer.salesforce.com/docs/ai/agentforce/guide/ascript-blocks.html#connected-subagent-block) in the _Agentforce Developer Guide_. ## Streaming Responses to Users Streaming shows progress and live agent reasoning updates across all channels. Users experience continuous progress with immediate visibility and reduced wait times for answers. ![Streaming shows progress and live agent reasoning updates.](https://sf-zdocs-cdn-prod.zoominsoftware.com/tdta-ai-generative_ai-264-0-0-production-enus/f8ef6498-d7b2-44e3-9b34-ffd5a850aae9/generative_ai/images/agent_multi_orch_stream.png) - In handoff mode, the connected subagent streams the response directly to the user. Its answer appears in the main response area instead of the reasoning panel. - In supervisor mode, the orchestrator calls the first connected subagent with streaming enabled. The user sees a progress indicator, and then the connected subagent’s response streams as word chunks, which appear as thinking events in the reasoning pane of the orchestrator agent. For multi-intent queries, the orchestrator agent calls another connected subagent and streams the response similarly. When finished, the orchestrator agent synthesizes all the responses and streams the answer to the main response area. ## Tips for Designing Your Agent-to-Agent Solution Here are a few tips to get you started designing an agent for use in Multi-Agent Orchestration for Agentforce. - Start by deciding which design pattern (hand-off or orchestrator mode) best fits your solution requirements. - For determinism and consistent behavior, base connected subagents on agents that are unique and distinct. For example, authorizing a return, sending a shipping label, and refunding a purchase represent three discrete actions that different connected subagents can perform. - Think modularity and reuse. Specialized agents can represent different business domains that interact across organizational boundaries (for example, Sales, Human Resources, and Finance) to achieve a common goal. When designed for reuse, specialized agents can do the work across many multi-agent solutions. - Craft the descriptions (instructions) used for connected subagents. The orchestrator agent and Agent Router rely on the description to determine when to route to a connected subagent. Descriptions that are vague, or that overlap with other subagents or connected subagents, can result in inconsistent behavior. ## Ways to Minimize Latency in Multi-Agent Solutions Latency in multi-agent solutions is intrinsically higher than in single-agent solutions. Intrinsic factors include coordination overhead and a higher frequency of LLM calls (to synthesize responses, for example). Follow these approaches to mitigate latency. - An orchestrator agent can support multiple connected agents. The system enforces no limit on the number of connected subagents. Based on our internal performance testing, we found no impact to latency with a configuration of one orchestrator agent with 8 connected subagents (each of which has 8 subagents). - Use handoff mode (the default) wherever possible. Orchestration mode requires an additional LLM call to synthesize connected subagent results. - Use streaming (the default) wherever possible. Users see the answer sooner, before any post-processing overhead. - Use the Salesforce-owned HyperClassifier model for intent routing classification (see [Subagent Classification and Routing](https://help.salesforce.com/s/articleView?id=ai.agent_topics_routing.htm&type=5&language=en_US) ). The HyperClassifier model outperforms standard LLMs in speed and precision, particularly when working from complex instructions or adhering to negative instructions. To learn more, see this blog post: [Solving Real-Time AI Classification for Agentforce: How Single-Token Prediction Delivers 30x Faster Agent Responses](https://engineering.salesforce.com/solving-real-time-ai-classification-for-agentforce-how-single-token-prediction-delivers-30x-faster-agent-responses/?utm_source=linkedin&utm_medium=organic_social&utm_campaign=amer_sfjobs_sfteaw&utm_content=graphic_). - For ASA, minimize groundedness by using `additional_parameter__DISABLE_GROUNDEDNESS: True` in your scripts. When an orchestrator hands off to an ASA connected subagent, both the orchestrator and the connected subagent independently run their own full groundedness pass. Neither check blocks the first token if streaming is enabled. The two checks run sequentially, one after the other, on the path to the turn actually finishing. See [Hybrid Reasoning with New Agentforce Builder and Agent Script](https://architect.salesforce.com/docs/architect/fundamentals/guide/hybrid-reasoning-agentforce-builder-agent-script). ## Tips for Testing Your Multi-Agent Solution - Test your agent individually - in isolation - before connecting to it from an orchestrator agent. Verify its instructions and reasoning actions. Verify any agent actions associated with the connected subagent, including the underlying flows, prompt templates, or other reference actions. Unit testing helps you find and resolve issues in isolation so that you’re more confident when conducting end-to-end testing. - Conduct comprehensive, end-to-end integration testing. Testing your agent in various combinations with other solution components to make sure that they interact as expected. - Use Testing Center to verify whether the orchestrator agent selects the right connected subagent at run time. In Testing Center, an Agent Handoff Evaluation column shows you whether the handoff occurred as expected (Pass) or not (Fail). - **[Considerations and Limitations in Multi-Agent Solutions](https://help.salesforce.com/s/articleView?id=ai.agent_multi_orch_consider.htm&release=264.0.0&type=5)** Review these considerations and limitations for Multi-Agent Orchestration for Agentforce. - **[User-Level Authentication in Multi-Agent Solutions](https://help.salesforce.com/s/articleView?id=ai.agent_multi_orch_auth.htm&release=264.0.0&type=5)** Agentforce multi-agent solutions support user-level authentication. Authentication varies by agent type and by the combination of orchestrator agent and connected subagent. - **[Connect an Agent as a Subagent](https://help.salesforce.com/s/articleView?id=ai.agent_multi_orch_connect.htm&release=264.0.0&type=5)** Add an agent as a connected subagent to your orchestrator agent. Your orchestrator agent can coordinate efforts among connected subagents, such as delegating tasks and synthesizing results. - **[Remove a Connected Subagent from an Orchestrator Agent](https://help.salesforce.com/s/articleView?id=ai.agent_multi_orch_remove.htm&release=264.0.0&type=5)** Remove a connected subagent you no longer need from your orchestrator agent. Removing the connected subagent doesn’t affect the reference agent on which this connected agent is based. - **[Agent Script in Multi-Agent Solutions](https://help.salesforce.com/s/articleView?id=ai.agent_multi_orch_script.htm&release=264.0.0&type=5)** Implement determinism in your multi-agent solution by using Agent Script and these design patterns. - **[Troubleshoot Agentforce Multi-Agent Solutions](https://help.salesforce.com/s/articleView?id=ai.agent_multi_orch_troubleshoot.htm&release=264.0.0&type=5)** Solve common issues with AI solutions involving Multi-agent Orchestration for Agentforce.