feat: Add llama-agent template (#150)

---------
Co-authored-by: Marcus Schiesser <mail@marcusschiesser.de>

.changeset/shy-horses-think.md

@@ -0,0 +1,5 @@
+---
+"create-llama": patch
+---
+
+Add new template for a multi-agents app

helpers/env-variables.ts

@@ -5,6 +5,7 @@ import {
   ModelConfig,
   TemplateDataSource,
   TemplateFramework,
+  TemplateType,
   TemplateVectorDB,
 } from "./types";
 
@@ -378,6 +379,36 @@ const getSystemPromptEnv = (tools?: Tool[]): EnvVar => {
   };
 };
 
+const getTemplateEnvs = (template?: TemplateType): EnvVar[] => {
+  if (template === "multiagent") {
+    return [
+      {
+        name: "MESSAGE_QUEUE_PORT",
+      },
+      {
+        name: "CONTROL_PLANE_PORT",
+      },
+      {
+        name: "HUMAN_CONSUMER_PORT",
+      },
+      {
+        name: "AGENT_QUERY_ENGINE_PORT",
+        value: "8003",
+      },
+      {
+        name: "AGENT_QUERY_ENGINE_DESCRIPTION",
+        value: "Query information from the provided data",
+      },
+      {
+        name: "AGENT_DUMMY_PORT",
+        value: "8004",
+      },
+    ];
+  } else {
+    return [];
+  }
+};
+
 export const createBackendEnvFile = async (
   root: string,
   opts: {
@@ -386,6 +417,7 @@ export const createBackendEnvFile = async (
     modelConfig: ModelConfig;
     framework: TemplateFramework;
     dataSources?: TemplateDataSource[];
+    template?: TemplateType;
     port?: number;
     tools?: Tool[];
   },
@@ -406,6 +438,8 @@ export const createBackendEnvFile = async (
     ...getVectorDBEnvs(opts.vectorDb, opts.framework),
     ...getFrameworkEnvs(opts.framework, opts.port),
     ...getToolEnvs(opts.tools),
+    // Add template environment variables
+    ...getTemplateEnvs(opts.template),
     getSystemPromptEnv(opts.tools),
   ];
   // Render and write env file

helpers/index.ts

@@ -141,15 +141,18 @@ export const installTemplate = async (
     // This is a backend, so we need to copy the test data and create the env file.
 
     // Copy the environment file to the target directory.
-    await createBackendEnvFile(props.root, {
-      modelConfig: props.modelConfig,
-      llamaCloudKey: props.llamaCloudKey,
-      vectorDb: props.vectorDb,
-      framework: props.framework,
-      dataSources: props.dataSources,
-      port: props.externalPort,
-      tools: props.tools,
-    });
+    if (props.template === "streaming" || props.template === "multiagent") {
+      await createBackendEnvFile(props.root, {
+        modelConfig: props.modelConfig,
+        llamaCloudKey: props.llamaCloudKey,
+        vectorDb: props.vectorDb,
+        framework: props.framework,
+        dataSources: props.dataSources,
+        port: props.externalPort,
+        tools: props.tools,
+        template: props.template,
+      });
+    }
 
     if (props.dataSources.length > 0) {
       console.log("\nGenerating context data...\n");

helpers/python.ts

@@ -320,20 +320,27 @@ export const installPythonTemplate = async ({
     cwd: path.join(compPath, "loaders", "python"),
   });
 
-  // Select and copy engine code based on data sources and tools
-  let engine;
-  tools = tools ?? [];
-  if (dataSources.length > 0 && tools.length === 0) {
-    console.log("\nNo tools selected - use optimized context chat engine\n");
-    engine = "chat";
-  } else {
-    engine = "agent";
-  }
-  await copy("**", enginePath, {
-    parents: true,
-    cwd: path.join(compPath, "engines", "python", engine),
+  // Copy settings.py to app
+  await copy("**", path.join(root, "app"), {
+    cwd: path.join(compPath, "settings", "python"),
   });
 
+  if (template === "streaming") {
+    // For the streaming template only:
+    // Select and copy engine code based on data sources and tools
+    let engine;
+    if (dataSources.length > 0 && (!tools || tools.length === 0)) {
+      console.log("\nNo tools selected - use optimized context chat engine\n");
+      engine = "chat";
+    } else {
+      engine = "agent";
+    }
+    await copy("**", enginePath, {
+      parents: true,
+      cwd: path.join(compPath, "engines", "python", engine),
+    });
+  }
+
   console.log("Adding additional dependencies");
 
   const addOnDependencies = getAdditionalDependencies(

helpers/types.ts

@@ -16,7 +16,11 @@ export type ModelConfig = {
   dimensions: number;
   isConfigured(): boolean;
 };
-export type TemplateType = "streaming" | "community" | "llamapack";
+export type TemplateType =
+  | "streaming"
+  | "community"
+  | "llamapack"
+  | "multiagent";
 export type TemplateFramework = "nextjs" | "express" | "fastapi";
 export type TemplateUI = "html" | "shadcn";
 export type TemplateVectorDB =

questions.ts

@@ -9,6 +9,7 @@ import {
   TemplateDataSource,
   TemplateDataSourceType,
   TemplateFramework,
+  TemplateType,
 } from "./helpers";
 import { COMMUNITY_OWNER, COMMUNITY_REPO } from "./helpers/constant";
 import { EXAMPLE_FILE } from "./helpers/datasources";
@@ -122,6 +123,7 @@ const getVectorDbChoices = (framework: TemplateFramework) => {
 export const getDataSourceChoices = (
   framework: TemplateFramework,
   selectedDataSource: TemplateDataSource[],
+  template?: TemplateType,
 ) => {
   // If LlamaCloud is already selected, don't show any other options
   if (selectedDataSource.find((s) => s.type === "llamacloud")) {
@@ -137,10 +139,12 @@ export const getDataSourceChoices = (
     });
   }
   if (selectedDataSource === undefined || selectedDataSource.length === 0) {
-    choices.push({
-      title: "No data, just a simple chat or agent",
-      value: "none",
-    });
+    if (template !== "multiagent") {
+      choices.push({
+        title: "No data, just a simple chat or agent",
+        value: "none",
+      });
+    }
     choices.push({
       title:
         process.platform !== "linux"
@@ -281,25 +285,27 @@ export const askQuestions = async (
           },
         ];
 
-        const modelConfigured =
-          !program.llamapack && program.modelConfig.isConfigured();
-        // If using LlamaParse, require LlamaCloud API key
-        const llamaCloudKeyConfigured = program.useLlamaParse
-          ? program.llamaCloudKey || process.env["LLAMA_CLOUD_API_KEY"]
-          : true;
-        const hasVectorDb = program.vectorDb && program.vectorDb !== "none";
-        // Can run the app if all tools do not require configuration
-        if (
-          !hasVectorDb &&
-          modelConfigured &&
-          llamaCloudKeyConfigured &&
-          !toolsRequireConfig(program.tools)
-        ) {
-          actionChoices.push({
-            title:
-              "Generate code, install dependencies, and run the app (~2 min)",
-            value: "runApp",
-          });
+        if (program.template !== "multiagent") {
+          const modelConfigured =
+            !program.llamapack && program.modelConfig.isConfigured();
+          // If using LlamaParse, require LlamaCloud API key
+          const llamaCloudKeyConfigured = program.useLlamaParse
+            ? program.llamaCloudKey || process.env["LLAMA_CLOUD_API_KEY"]
+            : true;
+          const hasVectorDb = program.vectorDb && program.vectorDb !== "none";
+          // Can run the app if all tools do not require configuration
+          if (
+            !hasVectorDb &&
+            modelConfigured &&
+            llamaCloudKeyConfigured &&
+            !toolsRequireConfig(program.tools)
+          ) {
+            actionChoices.push({
+              title:
+                "Generate code, install dependencies, and run the app (~2 min)",
+              value: "runApp",
+            });
+          }
         }
 
         const { action } = await prompts(
@@ -331,7 +337,11 @@ export const askQuestions = async (
           name: "template",
           message: "Which template would you like to use?",
           choices: [
-            { title: "Chat", value: "streaming" },
+            { title: "Agentic RAG (single agent)", value: "streaming" },
+            {
+              title: "Multi-agent app (using llama-agents)",
+              value: "multiagent",
+            },
             {
               title: `Community template from ${styledRepo}`,
               value: "community",
@@ -395,6 +405,10 @@ export const askQuestions = async (
     return; // early return - no further questions needed for llamapack projects
   }
 
+  if (program.template === "multiagent") {
+    // TODO: multi-agents currently only supports FastAPI
+    program.framework = preferences.framework = "fastapi";
+  }
   if (!program.framework) {
     if (ciInfo.isCI) {
       program.framework = getPrefOrDefault("framework");
@@ -420,7 +434,10 @@ export const askQuestions = async (
     }
   }
 
-  if (program.framework === "express" || program.framework === "fastapi") {
+  if (
+    (program.framework === "express" || program.framework === "fastapi") &&
+    program.template === "streaming"
+  ) {
     // if a backend-only framework is selected, ask whether we should create a frontend
     if (program.frontend === undefined) {
       if (ciInfo.isCI) {
@@ -457,7 +474,7 @@ export const askQuestions = async (
     }
   }
 
-  if (!program.observability) {
+  if (!program.observability && program.template === "streaming") {
     if (ciInfo.isCI) {
       program.observability = getPrefOrDefault("observability");
     } else {
@@ -501,6 +518,7 @@ export const askQuestions = async (
         const choices = getDataSourceChoices(
           program.framework,
           program.dataSources,
+          program.template,
         );
         if (choices.length === 0) break;
         const { selectedSource } = await prompts(
@@ -695,7 +713,8 @@ export const askQuestions = async (
     }
   }
 
-  if (!program.tools) {
+  if (!program.tools && program.template === "streaming") {
+    // TODO: allow to select tools also for multi-agent framework
     if (ciInfo.isCI) {
       program.tools = getPrefOrDefault("tools");
     } else {

templates/components/settings/python/llmhub.py

@@ -0,0 +1,61 @@
+from llama_index.embeddings.openai import OpenAIEmbedding
+from llama_index.core.settings import Settings
+from typing import Dict
+import os
+
+DEFAULT_MODEL = "gpt-3.5-turbo"
+DEFAULT_EMBEDDING_MODEL = "text-embedding-3-large"
+
+class TSIEmbedding(OpenAIEmbedding):
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        self._query_engine = self._text_engine = self.model_name
+
+def llm_config_from_env() -> Dict:
+    from llama_index.core.constants import DEFAULT_TEMPERATURE
+
+    model = os.getenv("MODEL", DEFAULT_MODEL)
+    temperature = os.getenv("LLM_TEMPERATURE", DEFAULT_TEMPERATURE)
+    max_tokens = os.getenv("LLM_MAX_TOKENS")
+    api_key = os.getenv("T_SYSTEMS_LLMHUB_API_KEY")
+    api_base = os.getenv("T_SYSTEMS_LLMHUB_BASE_URL")
+
+    config = {
+        "model": model,
+        "api_key": api_key,
+        "api_base": api_base,
+        "temperature": float(temperature),
+        "max_tokens": int(max_tokens) if max_tokens is not None else None,
+    }
+    return config
+
+
+def embedding_config_from_env() -> Dict:
+    from llama_index.core.constants import DEFAULT_EMBEDDING_DIM
+
+    model = os.getenv("EMBEDDING_MODEL", DEFAULT_EMBEDDING_MODEL)
+    dimension = os.getenv("EMBEDDING_DIM", DEFAULT_EMBEDDING_DIM)
+    api_key = os.getenv("T_SYSTEMS_LLMHUB_API_KEY")
+    api_base = os.getenv("T_SYSTEMS_LLMHUB_BASE_URL")
+
+    config = {
+        "model_name": model,
+        "dimension": int(dimension) if dimension is not None else None,
+        "api_key": api_key,
+        "api_base": api_base,
+    }
+    return config
+
+def init_llmhub():
+    from llama_index.llms.openai_like import OpenAILike
+
+    llm_configs = llm_config_from_env()
+    embedding_configs = embedding_config_from_env()
+
+    Settings.embed_model = TSIEmbedding(**embedding_configs)
+    Settings.llm = OpenAILike(
+        **llm_configs,
+        is_chat_model=True,
+        is_function_calling_model=False,
+        context_window=4096,
+    )

templates/types/streaming/fastapi/app/settings.py → templates/components/settings/python/settings.py

@@ -2,7 +2,7 @@
 from typing import Dict
 
 from llama_index.core.settings import Settings
-from .llmhub import init_llmhub
+
 
 def init_settings():
     model_provider = os.getenv("MODEL_PROVIDER")
@@ -20,6 +20,8 @@ def init_settings():
         case "azure-openai":
             init_azure_openai()
         case "t-systems":
+            from .llmhub import init_llmhub
+
             init_llmhub()
         case _:
             raise ValueError(f"Invalid model provider: {model_provider}")
@@ -147,5 +149,3 @@ def init_gemini():
 
     Settings.llm = Gemini(model=model_name)
     Settings.embed_model = GeminiEmbedding(model_name=embed_model_name)
-
-