diff --git a/modules/ROOT/pages/af-agent-networks.adoc b/modules/ROOT/pages/af-agent-networks.adoc index 9cfed9ce8..2891238eb 100644 --- a/modules/ROOT/pages/af-agent-networks.adoc +++ b/modules/ROOT/pages/af-agent-networks.adoc @@ -60,6 +60,39 @@ These servers will be published to Exchange. The asset might also be published to Exchange. . An asset that has been added to this agent network project from Exchange. +[[llm-support]] +== Large Language Models + +Agent network brokers support these LLMs. + +* Azure OpenAI +* OpenAI API +* Gemini API + +The following table details the requirements and recommended models. + +|=== +|*Model Provider* |*Required Endpoint* |*Required Capabilities* |*Suggested Models* + +|OpenAI |`/responses` a| +* Reasoning +* Native structured output +* Function and Custom tool calling +a| +* For lower latency: GPT-5-mini +* For complex reasoning: Evaluate models + +|Gemini |`/generateContent` (Native API) a| +* Native Thinking (via thinkingBudget and thinkingLevel) +* Native structured output (responseSchema) +* Function and custom tool calling +a| +* For lower latency: Gemini 2.5 Flash, Gemini 2.5 Flash-Lite +* For complex reasoning: Gemini 3 Pro (Deep Think capabilities) +|=== + +Agent network supports text-based prompts and responses. Image and binary message types aren't supported. + == A2A Protocol The Agent2Agent (A2A) Protocol governs agent-to-agent communication. This protocol powers orchestration, observability, and governance features in agent networks. MuleSoft supports v0.3.0 of the A2A Protocol Specification. @@ -89,28 +122,8 @@ Consider a network with a client and two brokers (1 and 2). For more information, see https://a2a-protocol.org/v0.3.0/topics/life-of-a-task/#group-related-interactions[Life of a Task - Group Related Interactions]. -=== Non-Supported Features - -Agent network doesn't support streaming with Server-Sent Events (SSE). - -== Considerations - -Agent networks have these considerations. +NOTE: Agent network doesn't support streaming with Server-Sent Events (SSE). -* An agent network project supports only one domain (one YAML file per project). For multiple domains, create a separate project for each. -* These authentication types are supported. -** Basic -** Client Credentials -** Anypoint Client Credentials -* Text-based prompts and responses are supported. Image and binary message/artifact types aren't supported. -* All LLM models in your agent network project must support structured responses in JSON format. For example, a `gpt4.1` model is supported; a `gpt4` model isn't. -* These LLMs are supported. -** Azure OpenAI -** OpenAI Direct API -* When you add governance policies to your agent network project YAML file, those policies persist between deployments. However, if you add governance policies at run-time, those policies don't persist. The next time you deploy your agent network, the governance policies used are the ones in the `agent-network.yaml` file. -* These deployment targets are supported. -** CloudHub 2.0 -** Runtime Fabric (limited availability) == See Also diff --git a/modules/ROOT/pages/af-project-files.adoc b/modules/ROOT/pages/af-project-files.adoc index 89453bd82..e24b4ce64 100644 --- a/modules/ROOT/pages/af-project-files.adoc +++ b/modules/ROOT/pages/af-project-files.adoc @@ -231,14 +231,7 @@ The `spec` element has these properties. The value of this section is a reference to one of the LLMs defined in Anypoint Exchange or in the `llmProviders` section of `agent-network.yaml`. Because it's a reference, you can choose to share the same LLM across all the brokers in your agent network. Or, you can have different brokers use different LLMs to better suit their tasks. -Agent network brokers support OpenAI models. The model you use must support: - -* The `/responses` endpoint -* Reasoning -* Native structured output -* Function and Custom tool calling - -GPT-5-mini and GPT-5.1 meet these requirements. GPT-5-mini and GPT-5.1 meet these requirements. Evaluate the different models to find the right balance for your needs. For lower latency, consider smaller models like GPT-5-mini. +For more information on supported LLMs, see xref:af-agent-networks.adoc#llm-support[]. The `llm` element has these properties. @@ -264,7 +257,20 @@ a| |`configuration.openAI.topP` |Nucleus sampling parameter. Requires GPT-5.1 and `reasoningEffort` set to `NONE`. |Number |Any number |No |`configuration.openAI.topLogprobs` |Number of most likely tokens to return at each position. Requires GPT-5.1 and `reasoningEffort` set to `NONE`. |Integer |Any integer |No |`configuration.openAI.maxOutputTokens` |Maximum number of tokens to generate. |Integer |Any integer |No -|=== +|`configuration.gemini` |OpenAI specific configuration settings. |Object |Object with OpenAI settings |No +|`configuration.gemini.thinkingConfig.thinkingBudget` |Sets a token budget for the reasoning phase. (Applies only to Gemini 2.5 series.) |Integer a| +* `0` (Disabled) +* `-1` (Automatic) +* `1024` to `32768` |No +|`configuration.gemini.thinkingLevel` |Controls the depth of the reasoning process. (Applies only to Gemini 3 series.) |String a| +* `High` (Default) +* `Low` |No +|`configuration.gemini.temperature` |Controls randomness. |Number |For Gemini 3 and Gemini 2.5, Google recommends keeping this at `1.0` (default) to avoid breaking the reasoning chain. |No +|`configuration.gemini.topP` |Nucleus sampling parameter. |Number |Any number |No +|`configuration.gemini.responseLogprobs` |Whether to return log probabilities. |Boolean |true or false |No +|`configuration.gemini.maxOutputTokens` |Maximum number of tokens that can be generated in the response. |Integer | |No + +|=== [[instructions-section]] ==== Instructions Section