examples: add voice agent example

# Conflicts:
#	examples/cerina_bindu/cbt/README.md
#	examples/multilingual-collab-agent/agent_config.json
#	examples/multilingual-collab-agent/main.py
#	examples/multilingual-collab-agent/skills/collaborate/skill.yaml
#	examples/multilingual-collab-agent/skills/research/skill.yaml
#	examples/multilingual-collab-agent/skills/translate/skill.yaml
#	examples/pdf_research_agent/README.md
#	examples/pdf_research_agent/pdf_research_agent.py
#	examples/pdf_research_agent/skills/pdf-research-skill/skill.yaml

Author: Co-vengers

examples/README.md

@@ -63,6 +63,7 @@ For full URL override, use `BINDU_DEPLOYMENT_URL` (e.g. `http://127.0.0.1:5001`)
 - `speech-to-text/` - Audio transcription using Gemini 2.0 Flash (MP3, WAV, OGG, M4A)
 - `ai-data-analysis-agent/` - Autonomous data analyst with CSV profiling and visualization
 - `cybersecurity-newsletter/` - Security news aggregator with CVE tracking
+- `voice-agent/` - Voice-enabled agent with WebSocket voice session endpoints
 
 ### TypeScript (Language-Agnostic via gRPC)
 - `typescript-openai-agent/` - OpenAI SDK agent bindufied with TypeScript SDK

examples/cerina_bindu/cbt/README.md

@@ -57,7 +57,32 @@ cp .env.example .env
 uv run python supervisor_cbt.py
 ```
 
-Agent starts at `http://localhost:3773`. See [`cbt/README.md`](cbt/README.md) for full usage instructions including example curl commands.
+Agent starts at `http://localhost:3773`.
+
+Minimal smoke test:
+
+```bash
+curl -X POST http://localhost:3773/ \
+     -H "Content-Type: application/json" \
+     -d '{
+          "jsonrpc": "2.0",
+          "method": "message/send",
+          "params": {
+               "message": {
+                    "role": "user",
+                    "parts": [{"kind": "text", "text": "I feel overwhelmed and procrastinate constantly. Give me a short CBT exercise."}],
+                    "kind": "message",
+                    "messageId": "msg-cbt-1",
+                    "contextId": "ctx-cbt-1",
+                    "taskId": "task-cbt-1"
+               },
+               "configuration": {"acceptedOutputModes": ["application/json"]}
+          },
+          "id": "1"
+     }'
+```
+
+Expected result: HTTP 200 with a JSON-RPC response containing a task/result payload with structured CBT guidance.
 
 ---
 

examples/document-analyzer/document_analyzer.py

@@ -16,15 +16,19 @@
 import os
 import io
 import base64
+from typing import Any
 
 from pypdf import PdfReader
 from docx import Document
+from bindu.utils.logging import get_logger
+
+logger = get_logger("examples.document_analyzer")
 
 load_dotenv()
 
 # Define LLM agent
 agent = Agent(
-    instructions = """
+    instructions="""
 You are an advanced document analysis assistant.
 
 Your job is to analyze uploaded documents and answer the user's prompt
@@ -42,12 +46,13 @@
 - If the prompt asks for summary, provide concise bullet points
 - Do not hallucinate information outside the document
 """,
-    model = OpenRouter(
-        id = "arcee-ai/trinity-large-preview:free",
+    model=OpenRouter(
+        id="arcee-ai/trinity-large-preview:free",
         api_key=os.getenv("OPENROUTER_API_KEY"),
     ),
 )
 
+
 # Document Parsing
 def extract_text_from_pdf(file_bytes):
     """Extract text from pdf bytes"""
@@ -67,10 +72,16 @@ def extract_text_from_pdf(file_bytes):
 
     return "\n".join(text)
 
+
 def extract_text_from_docx(file_bytes):
     """Extract text from docx bytes"""
-    doc = Document(io.BytesIO(file_bytes))
-    return "\n".join([p.text for p in doc.paragraphs])
+    try:
+        doc = Document(io.BytesIO(file_bytes))
+        return "\n".join([p.text for p in doc.paragraphs])
+    except Exception as e:
+        logger.error(f"Error extracting DOCX text: {e}")
+        return ""
+
 
 def extract_document_text(file_bytes, mime_type):
     """Parse document according to their mime type"""
@@ -84,6 +95,7 @@ def extract_document_text(file_bytes, mime_type):
 
     raise ValueError(f"Unsupported file type: {mime_type}")
 
+
 # FilePart processing
 def get_file_bytes(part):
     """Extract file bytes from FilePart"""
@@ -98,39 +110,38 @@ def get_file_bytes(part):
 
     if isinstance(data, str):
         import base64
+
         return base64.b64decode(data)
 
     return data
 
-# Handler
-def handler(messages: list[dict]):
-    """
-    Receives task.history — a list of A2A Message objects.
-    Each message has: role, parts[], kind, messageId, contextId, taskId
-    Each part has: kind="text"|"file", and either text or file.bytes+mimeType
-    """
-    if not messages:
-        return "No messages received."
-    import json
-    print("DEBUG messages:", json.dumps(messages, indent=2, default=str))
 
-    prompt = ""
-    extracted_docs = []
+# Handler
+def _collect_prompt_and_documents(
+    messages: list[dict[str, Any]],
+) -> tuple[str, list[str], list[str]]:
+    """Support both raw A2A messages and runtime chat-format messages."""
+    prompt_parts: list[str] = []
+    extracted_docs: list[str] = []
+    errors: list[str] = []
 
     for msg in messages:
-        # if a role is provided, only process user messages; treat missing
-        # roles as coming from the user so that tests/clients without a role
-        # field still work.
         role = msg.get("role")
         if role is not None and role != "user":
             continue
 
-        # be defensive: parts could be None or omitted
+        # Runtime path: manifest worker passes chat-format messages.
+        content = msg.get("content")
+        if isinstance(content, str) and content.strip():
+            prompt_parts.append(content)
+
+        # Compatibility path: raw A2A messages with parts.
         parts = msg.get("parts") or []
         for part in parts:
             if part.get("kind") == "text":
-                prompt = part.get("text", "")
-
+                text = part.get("text", "")
+                if text:
+                    prompt_parts.append(text)
             elif part.get("kind") == "file":
                 try:
                     file_info = part.get("file", {})
@@ -147,30 +158,46 @@ def handler(messages: list[dict]):
                     )
                     doc_text = extract_document_text(file_bytes, mime_type)
                     extracted_docs.append(doc_text)
-
                 except Exception as e:
-                    extracted_docs.append(f"Error processing file: {str(e)}")
+                    errors.append(str(e))
+
+    return "\n".join(prompt_parts).strip(), extracted_docs, errors
+
+
+def handler(messages: list[dict]):
+    """
+    Receives task.history — a list of A2A Message objects.
+    Each message has: role, parts[], kind, messageId, contextId, taskId
+    Each part has: kind="text"|"file", and either text or file.bytes+mimeType
+    """
+    if not messages:
+        return "No messages received."
+    prompt, extracted_docs, errors = _collect_prompt_and_documents(messages)
 
     if not extracted_docs:
+        if errors:
+            return "Failed to process documents:\n" + "\n".join(errors)
         return "No valid document found in the messages."
 
     combined_document = "\n\n".join(extracted_docs)
-    result = agent.run(input=f"""
+    result = agent.run(
+        input=f"""
 User Prompt:
 {prompt}
 
 Document Content:
 {combined_document}
 
 Provide analysis based on the prompt.
-""")
+"""
+    )
     return result
 
 
 # Bindu config
 config = {
-    "author" : "vyomrohila@gmail.com",
-    "name" : "document_analyzer_agent",
+    "author": "vyomrohila@gmail.com",
+    "name": "document_analyzer_agent",
     "description": "AI agent that analyzes uploaded PDF or DOCX documents based on a user prompt.",
     "deployment": {
         "url": "http://localhost:3773",

examples/multilingual-collab-agent/main.py

@@ -1,12 +1,3 @@
-# |---------------------------------------------------------|
-# |                                                         |
-# |                 Give Feedback / Get Help                |
-# | https://github.com/getbindu/Bindu/issues/new/choose    |
-# |                                                         |
-# |---------------------------------------------------------|
-#
-#  Thank you users! We ❤️ you! - 🌻
-
 """Multilingual Collaborative Agent v2 — A Bindu Agent.
 
 An identity-aware agent that detects user language (English, Hindi, Bengali)
@@ -35,6 +26,8 @@
 agent: Agent | None = None
 _initialized = False
 _init_lock = asyncio.Lock()
+DEFAULT_MODEL = "openai/gpt-oss-120b"
+_mem0_enabled = False
 
 
 def load_config() -> dict:
@@ -46,9 +39,11 @@ def load_config() -> dict:
 
 def build_agent() -> Agent:
     """Build and return the multilingual agent instance."""
+    global _mem0_enabled
+
     openrouter_api_key = os.getenv("OPENROUTER_API_KEY")
     mem0_api_key = os.getenv("MEM0_API_KEY")
-    model_name = os.getenv("MODEL_NAME", "openai/gpt-oss-120b")
+    model_name = os.getenv("MODEL_NAME", DEFAULT_MODEL)
 
     if not openrouter_api_key:
         raise ValueError("OPENROUTER_API_KEY environment variable is required.")
@@ -160,6 +155,8 @@ async def handler(messages: list[dict[str, str]]) -> Any:
             print("🔧 Initializing multilingual agent...")
             agent = build_agent()
             _initialized = True
+            if _mem0_enabled:
+                print("   Memory: Mem0 persistent memory enabled")
             print("✅ Agent initialized")
 
     response = await agent.arun(messages)
@@ -170,9 +167,8 @@ def main() -> None:
     """Start the Bindu agent server."""
     config = load_config()
     print("🌍 Starting Multilingual Collaborative Agent...")
-    print(f"   Supported languages: English, Hindi (हिन्दी), Bengali (বাংলা)")
-    print(f"   Model: {os.getenv('MODEL_NAME', 'openai/gpt-4o-mini')}")
-    print(f"   Memory: Mem0 persistent memory enabled")
+    print("   Supported languages: English, Hindi (हिन्दी), Bengali (বাংলা)")
+    print(f"   Model: {os.getenv('MODEL_NAME', DEFAULT_MODEL)}")
     bindufy(config, handler)
 
 

examples/multilingual-collab-agent/skills/collaborate/skill.yaml

@@ -64,7 +64,7 @@ requirements:
     - "agno>=2.5.2"
     - "mem0ai>=0.1.0"
   system: []
-  min_memory_mb: 256
+  min_memory_mb: 1280
 
 performance:
   avg_processing_time_ms: 2500

examples/multilingual-collab-agent/skills/research/skill.yaml

@@ -54,7 +54,7 @@ requirements:
     - "duckduckgo-search>=8.1.1"
     - "agno>=2.5.2"
   system: []
-  min_memory_mb: 256
+  min_memory_mb: 640
 
 performance:
   avg_processing_time_ms: 3000
@@ -63,6 +63,7 @@ performance:
   scalability: horizontal
 
 allowed_tools:
+  # "Read" covers web access/search in this runtime.
   - Read
 
 documentation:

examples/multilingual-collab-agent/skills/translate/skill.yaml

@@ -76,6 +76,9 @@ documentation:
     Particularly useful for technical content about Bindu and AI agents where
     standard translators produce poor results for domain-specific terms.
 
+    Configuration: requires `OPENROUTER_API_KEY` in the environment (and optionally
+    `OPENROUTER_API_BASE` / `OPENROUTER_API_URL` if using a non-default endpoint).
+
   use_cases:
     when_to_use:
       - "User provides text and asks to translate to a specific language"
@@ -89,19 +92,33 @@ documentation:
       - "Real-time speech translation (text only)"
 
   input_structure: |
-    Text with explicit translation instruction.
+    Text with explicit translation instruction (text/plain) OR JSON (application/json).
 
     Examples:
       "Translate to Hindi: [text]"
       "हिंदी में अनुवाद करें: [text]"
       "বাংলায় অনুবাদ করুন: [text]"
       "Translate this to English: [text in any language]"
 
+    JSON schema (application/json):
+      {
+        "text": "string",
+        "target_language": "string",
+        "source_language": "string (optional)"
+      }
+
   output_format: |
-    Translated text in the target language.
+    Translated text in the target language (text/plain) OR JSON (application/json).
     For technical terms without direct translation:
       Original term kept + brief explanation in target language.
 
+    JSON schema (application/json):
+      {
+        "translated_text": "string",
+        "source_language": "string",
+        "target_language": "string"
+      }
+
   error_handling:
     - "Unsupported language requested: informs user of supported languages (EN/HI/BN)"
     - "Ambiguous source language: asks for clarification"

examples/pdf_research_agent/README.md

@@ -36,10 +36,10 @@ pip install bindu agno pypdf python-dotenv
 ### 3. Set Up Environment Variables
 ```bash
 # Create .env file
-echo "OPENROUTER_API_KEY=your_api_key_here" > .env
+echo "OPENROUTER_API_KEY=your_api_key_here" > .env # pragma: allowlist secret
 
 # Or export directly
-export OPENROUTER_API_KEY="your_api_key_here"  # pragma: allowlist secret
+export OPENROUTER_API_KEY="your_api_key_here" # pragma: allowlist secret
 ```
 
 ## 🎯 Usage
@@ -268,7 +268,7 @@ uv add pypdf
 
 #### "OPENROUTER_API_KEY not set"
 ```bash
-export OPENROUTER_API_KEY="your_api_key_here"  # pragma: allowlist secret
+export OPENROUTER_API_KEY="your_api_key_here" # pragma: allowlist secret
 ```
 
 #### "Port 3773 already in use"