agent-pulse 1.2.0 → 1.4.0

package/README.md

@@ -99,19 +99,19 @@ import { Agent, openAI, google, grok } from 'agent-pulse';
 // OpenAI
 const bot1 = new Agent({
     name: 'gpt-bot',
-    provider: new openAI('gpt-4o')
+    provider: new openAI('gpt-5.2')
 });
 
 // Google Gemini
 const bot2 = new Agent({
     name: 'gemini-bot',
-    provider: new google('gemini-1.5-pro')
+    provider: new google('gemini-3-pro')
 });
 
 // xAI / Grok
 const bot3 = new Agent({
     name: 'grok-bot',
-    provider: new grok('grok-beta')
+    provider: new grok('grok-4.2')
 });
 ```
 
@@ -154,7 +154,8 @@ The `response` event and the `agent.run()` promise resolve to a standardized `Ag
 
 ```typescript
 {
-    content: string | object, // The Markdown text or parsed JSON
+    content: string | object, // The Markdown text, parsed JSON, or tool result (if iterations=1)
+    message?: string,          // The original LLM text response (useful when a tool is also called)
     usage: {
         input_tokens: number,
         output_tokens: number,
@@ -167,6 +168,9 @@ The `response` event and the `agent.run()` promise resolve to a standardized `Ag
 }
 ```
 
+> [!NOTE]
+> If an LLM responds with both text and a tool call (common in Gemini), `content` stays consistent with legacy behavior (holding the tool result), while the new `message` field preserves the original LLM text.
+
 You can access token usage stats from the `usage` property.
 
 ## Error Codes
@@ -213,7 +217,7 @@ const summaryTool = {
 
 const agent = new Agent({
   name: 'intake',
-  provider: new google('gemini-1.5-pro'),
+  provider: new google('gemini-3-pro'),
   tools: [summaryTool]
 });
 
@@ -226,7 +230,18 @@ if (result.content?.type === 'INTENT_COMPLETE') {
 }
 ```
 
-#### Option B: Events (Side Effects)
+#### Option B: Handling Text + Tool (Gemini Style)
+When using models like Gemini that often provide a text explanation *and* a tool call in one turn, use the `message` field to access the text.
+
+```typescript
+const result = await agent.run("Tell me a joke and then get the weather.");
+
+// If weatherTool was called:
+console.log(result.message); // "Sure! Here's a joke: ... Now, let me get the weather for you."
+console.log(result.content); // { temp: 20, unit: 'celsius' } (The tool result)
+```
+
+#### Option C: Events (Side Effects)
 Best for logging, UI updates, or real-time monitoring.
 
 ```typescript
@@ -248,7 +263,7 @@ By setting `max_tool_iterations`, the agent can autonomously call tools, receive
 ```typescript
 const agent = new Agent({
   name: 'researcher',
-  provider: new openAI('gpt-4o'),
+  provider: new openAI('gpt-5.2'),
   tools: [weatherTool, searchTool],
   max_tool_iterations: 5 // Allow up to 5 loop turns
 });
@@ -265,7 +280,7 @@ If your agent is running on a server but needs the **client** to perform an acti
 ```typescript
 const agent = new Agent({
   name: 'account-mgr',
-  provider: new openAI('gpt-4o'),
+  provider: new openAI('gpt-5.2'),
   tools: [requestConfirmationTool]
 });
 
@@ -297,7 +312,7 @@ The same pattern works for Gemini! While Google's API uses a different internal
 ```typescript
 const agent = new Agent({
   name: 'gemini-agent',
-  provider: new google('gemini-1.5-flash')
+  provider: new google('gemini-3-flash')
 });
 
 const final = await agent.run([
@@ -325,7 +340,7 @@ import { Agent, openAI } from 'agent-pulse';
 
 const agent = new Agent({
   name: 'analyst',
-  provider: new openAI('gpt-4o'),
+  provider: new openAI('gpt-5.2'),
   // You can pass file paths (if handled by environment) or load content yourself
   files: ['/path/to/data.txt'] 
 });
@@ -347,7 +362,7 @@ const recipeSchema = z.object({
 
 const agent = new Agent({
   name: 'chef',
-  provider: new google('gemini-1.5-pro'),
+  provider: new google('gemini-3-pro'),
   output_schema: recipeSchema
 });
 
@@ -374,7 +389,7 @@ app.get('/chat', async (req, res) => {
 
   const agent = new Agent({ 
       name: 'web-bot', 
-      provider: new openAI('gpt-4o') 
+      provider: new openAI('gpt-5.2') 
   });
   
   // Connect agent events to the response stream
@@ -394,7 +409,7 @@ import { Agent, google } from 'agent-pulse';
 
 const agent = new Agent({
   name: 'researcher',
-  provider: new google('gemini-2.5-flash-lite'),
+  provider: new google('gemini-3-flash'),
   config: {
     googleSearch: true
   }
@@ -410,6 +425,47 @@ agent.on('response', (result) => {
 await agent.run("Who won the Super Bowl in 2024?");
 ```
 
+### 7. Image Generation
+
+Generate images by setting the agent's model to an image generation model (e.g., `grok-imagine-image` for xAI or models like `gemini-1.5-pro` for Google).
+
+#### xAI (Grok Imagine)
+xAI uses `aspect_ratio` instead of `size`.
+
+```typescript
+import { Agent, grok } from 'agent-pulse';
+
+const agent = new Agent({
+  name: 'artist',
+  provider: new grok('grok-imagine-image'),
+  config: {
+    aspect_ratio: '16:9', // Supported: "1:1", "16:9", "9:16", "4:3", "3:2", etc.
+    response_format: 'b64_json' // Get base64 data instead of temporary URLs
+  }
+});
+
+const result = await agent.run("A futuristic city skyline in neon colors.");
+
+// result.content will contain markdown image strings:
+// "![Generated Image](data:image/png;base64,...)" or "![Generated Image](https://...)"
+```
+
+#### Google (Gemini)
+Gemini models can generate images as part of their response.
+
+```typescript
+import { Agent, google } from 'agent-pulse';
+
+const agent = new Agent({
+  name: 'painter',
+  provider: new google('gemini-3-pro')
+});
+
+const result = await agent.run("Generate an image of a serene mountain lake.");
+
+// result.content will contain the markdown image string.
+```
+
 ## Extensibility: Custom Providers
 
 To add a new provider (e.g. Anthropic, Mistral), create a class that implements the `LLMProvider` interface.
@@ -433,10 +489,6 @@ const agent = new Agent({
     provider: new MyProvider('my-model')
 });
 ```
-## To locally link the package
-
-1. Run `npm link` in the agent-pulse directory
-2. Run `npm link agent-pulse --legacy-peer-deps` in your project directory
 
 ## License
 

package/dist/agent.js

@@ -39,6 +39,10 @@ class Agent extends events_1.EventEmitter {
                 iterations++;
                 const response = await this.provider.generate(this.config.system, messages, this.config.files, this.config.tools, this.config.config, this.config.output_schema, (token) => this.emit('token', token));
                 lastResponse = response;
+                // Capture the original text as the "message" (LLM's primary text response)
+                if (typeof response.content === 'string') {
+                    lastResponse.message = response.content;
+                }
                 // Handle Tool Execution
                 if (response.tool_calls && this.config.tools) {
                     // Add Assistant's tool call message to history

package/dist/types.d.ts

@@ -29,6 +29,7 @@ export interface AgentTool {
 export interface AgentResponse {
     content: string | object;
     tool_calls?: any[];
+    message?: string;
     usage: {
         input_tokens: number;
         output_tokens: number;

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "agent-pulse",
-    "version": "1.2.0",
+    "version": "1.4.0",
     "description": "A lightweight, agentic AI framework for JavaScript/TypeScript",
     "main": "dist/index.js",
     "types": "dist/index.d.ts",
@@ -26,13 +26,12 @@
     "license": "MIT",
     "repository": {
         "type": "git",
-        "url": "git+https://github.com/mehere14/Agentic-Framework.git",
-        "directory": "agent-pulse"
+        "url": "git+https://github.com/mehere14/agent-pulse-public.git"
     },
     "bugs": {
-        "url": "https://github.com/mehere14/Agentic-Framework/issues"
+        "url": "https://github.com/mehere14/agent-pulse-public/issues"
     },
-    "homepage": "https://github.com/mehere14/Agentic-Framework/tree/main/agent-pulse#readme",
+    "homepage": "https://github.com/mehere14/agent-pulse-public#readme",
     "dependencies": {
         "@google/genai": "^1.33.0",
         "dotenv": "^16.4.5",