feat(ag2): add AG2 framework backend integration (#1156)

feat(ag2): add AG2 framework backend integration

Author: MervinPraison

examples/ag2/ag2_basic.yaml

@@ -0,0 +1,30 @@
+framework: ag2
+topic: "Research the latest developments in AI agents"
+
+# Install: pip install "praisonai[ag2]"
+# Run:     praisonai --framework ag2 examples/ag2/ag2_basic.yaml
+#       or praisonai run examples/ag2/ag2_basic.yaml --framework ag2
+
+roles:
+  research_agent:
+    role: "AI Research Specialist"
+    goal: "Research and summarise the latest developments in AI agent frameworks"
+    backstory: |
+      You are an experienced AI researcher with deep knowledge of multi-agent
+      systems, large language models, and the latest trends in AI tooling.
+      You excel at synthesising complex technical topics into clear summaries.
+    tasks:
+      research_task:
+        description: |
+          Research and summarise the latest developments in AI agent frameworks
+          for the topic: {topic}
+
+          Focus on:
+          1. Key frameworks and their unique capabilities
+          2. Recent innovations and improvements
+          3. Community adoption and ecosystem growth
+          4. Practical use cases and success stories
+        expected_output: |
+          A concise research summary covering the key developments,
+          major frameworks, and practical insights. Include 3-5 bullet
+          points of the most important findings.

examples/ag2/ag2_bedrock.yaml

@@ -0,0 +1,42 @@
+framework: ag2
+topic: "Cloud-native AI deployment strategies on AWS"
+
+# AG2 exclusive feature: native AWS Bedrock support via LLMConfig(api_type="bedrock")
+#
+# Prerequisites:
+#   pip install "praisonai[ag2]"
+#   aws configure  (or set AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_DEFAULT_REGION)
+#
+# Run:
+#   praisonai --framework ag2 examples/ag2/ag2_bedrock.yaml
+#
+# The AG2 adapter detects api_type="bedrock" from the llm config and uses
+# LLMConfig(api_type="bedrock", model=...) — no OPENAI_API_KEY required.
+# AWS credentials are sourced from boto3 (env vars, ~/.aws/credentials, IAM role).
+
+roles:
+  cloud_architect:
+    role: "AWS Cloud Architect"
+    goal: "Design and explain cloud-native AI deployment strategies on AWS"
+    backstory: |
+      You are an AWS Solutions Architect specialising in AI/ML workloads.
+      You have deep expertise in Amazon Bedrock, SageMaker, ECS, and Lambda,
+      and you help organisations deploy AI agents at scale securely and cost-effectively.
+    llm:
+      model: "bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0"
+      api_type: "bedrock"
+      aws_region: "us-east-1"
+    tasks:
+      architecture_task:
+        description: |
+          Design a cloud-native deployment strategy for AI agents on AWS for: {topic}
+
+          Cover:
+          1. Recommended AWS services (Bedrock, ECS, Lambda, etc.)
+          2. Scalability and cost optimisation patterns
+          3. Security and compliance considerations
+          4. A simple reference architecture overview
+        expected_output: |
+          A concise architecture guide with service recommendations,
+          a high-level deployment diagram description, and key
+          best practices for production AI agent deployments on AWS.

examples/ag2/ag2_multi_agent.yaml

@@ -0,0 +1,54 @@
+framework: ag2
+topic: "The impact of open-source AI on enterprise software development"
+
+# Install: pip install "praisonai[ag2]"
+# Run:     praisonai --framework ag2 examples/ag2/ag2_multi_agent.yaml
+#       or praisonai run examples/ag2/ag2_multi_agent.yaml --framework ag2
+#
+# This example demonstrates AG2's GroupChat multi-agent coordination.
+# Both agents participate in a collaborative conversation managed by
+# a GroupChatManager until the task is complete.
+
+roles:
+  researcher:
+    role: "Research Specialist"
+    goal: "Gather and analyse information on the given topic"
+    backstory: |
+      You are a meticulous researcher who excels at finding relevant
+      information, analysing trends, and presenting data-backed insights.
+      You always cite your reasoning and structure your findings clearly.
+    tasks:
+      research_task:
+        description: |
+          Research the topic: {topic}
+
+          Investigate:
+          1. Current state and adoption rates
+          2. Key players and projects driving the trend
+          3. Technical advantages and challenges
+          4. Business impact and cost implications
+        expected_output: |
+          A structured research briefing with findings on the topic,
+          including key data points, trends, and technical observations.
+
+  writer:
+    role: "Technical Content Writer"
+    goal: "Transform research findings into clear, engaging written content"
+    backstory: |
+      You are a skilled technical writer who turns complex research into
+      accessible, well-structured articles. You focus on clarity, logical
+      flow, and actionable takeaways for a professional audience.
+    tasks:
+      writing_task:
+        description: |
+          Using the research findings provided by the Research Specialist,
+          write a concise article on: {topic}
+
+          The article should:
+          1. Open with a compelling hook
+          2. Present key findings logically
+          3. Include practical implications for developers
+          4. Close with a forward-looking conclusion
+        expected_output: |
+          A 400-500 word article suitable for a technical blog,
+          with clear sections, professional tone, and concrete takeaways.

src/praisonai/praisonai/agents_generator.py

@@ -55,6 +55,16 @@
 except ImportError:
     pass
 
+AG2_AVAILABLE = False
+try:
+    import importlib.metadata as _importlib_metadata
+    _importlib_metadata.distribution('ag2')
+    from autogen import LLMConfig as _AG2LLMConfig  # noqa: F401 — AG2-exclusive class
+    AG2_AVAILABLE = True
+    del _AG2LLMConfig, _importlib_metadata
+except Exception:
+    pass
+
 try:
     import agentops
     AGENTOPS_AVAILABLE = True
@@ -65,7 +75,7 @@
     pass
 
 # Only try to import praisonai_tools if either CrewAI or AutoGen is available
-if CREWAI_AVAILABLE or AUTOGEN_AVAILABLE or PRAISONAI_AVAILABLE:
+if CREWAI_AVAILABLE or AUTOGEN_AVAILABLE or PRAISONAI_AVAILABLE or AG2_AVAILABLE:
     try:
         from praisonai_tools import (
             CodeDocsSearchTool, CSVSearchTool, DirectorySearchTool, DOCXSearchTool, DirectoryReadTool,
@@ -216,6 +226,8 @@ def __init__(self, agent_file, framework, config_list, log_level=None, agent_cal
             raise ImportError("AutoGen is not installed. Please install it with 'pip install praisonai[autogen]' for v0.2 or 'pip install praisonai[autogen-v4]' for v0.4")
         elif framework == "praisonai" and not PRAISONAI_AVAILABLE:
             raise ImportError("PraisonAI is not installed. Please install it with 'pip install praisonaiagents'")
+        elif framework == "ag2" and not AG2_AVAILABLE:
+            raise ImportError("AG2 is not installed. Please install it with 'pip install praisonai[ag2]'")
 
     def is_function_or_decorated(self, obj):
         """
@@ -391,7 +403,7 @@ def generate_crew_and_kickoff(self):
         tools_dict = {}
 
         # Only try to use praisonai_tools if it's available and needed
-        if PRAISONAI_TOOLS_AVAILABLE and (CREWAI_AVAILABLE or AUTOGEN_AVAILABLE or PRAISONAI_AVAILABLE):
+        if PRAISONAI_TOOLS_AVAILABLE and (CREWAI_AVAILABLE or AUTOGEN_AVAILABLE or PRAISONAI_AVAILABLE or AG2_AVAILABLE):
             tools_dict = {
                 'CodeDocsSearchTool': CodeDocsSearchTool(),
                 'CSVSearchTool': CSVSearchTool(),
@@ -462,6 +474,12 @@ def generate_crew_and_kickoff(self):
             else:
                 self.logger.info("Using AutoGen v0.2")
                 return self._run_autogen(config, topic, tools_dict)
+        elif framework == "ag2":
+            if not AG2_AVAILABLE:
+                raise ImportError("AG2 is not installed. Please install it with 'pip install praisonai[ag2]'")
+            if AGENTOPS_AVAILABLE:
+                agentops.init(os.environ.get("AGENTOPS_API_KEY"), default_tags=["ag2"])
+            return self._run_ag2(config, topic, tools_dict)
         elif framework == "praisonai":
             if not PRAISONAI_AVAILABLE:
                 raise ImportError("PraisonAI is not installed. Please install it with 'pip install praisonaiagents'")
@@ -711,6 +729,148 @@ async def run_autogen_v4_async():
             self.logger.error(f"Error running AutoGen v0.4: {str(e)}")
             return f"### AutoGen v0.4 Error ###\n{str(e)}"
 
+    def _run_ag2(self, config, topic, tools_dict):
+        """
+        Run agents using the AG2 framework (community fork of AutoGen, PyPI: ag2).
+
+        AG2 installs under the 'autogen' namespace — there is no 'import ag2'.
+        Uses LLMConfig context manager + AssistantAgent + GroupChat pattern.
+
+        Args:
+            config (dict): Configuration dictionary parsed from YAML
+            topic (str): The topic/task to process
+            tools_dict (dict): Dictionary of available tools
+
+        Returns:
+            str: Result prefixed with '### AG2 Output ###'
+        """
+        import re as _re
+        from autogen import (
+            AssistantAgent, UserProxyAgent, GroupChat, GroupChatManager, LLMConfig
+        )
+
+        model_config = self.config_list[0] if self.config_list else {}
+
+        # Allow YAML top-level llm block to override config_list values
+        yaml_llm = config.get("llm", {}) or {}
+        # Also check first role's llm block as a fallback
+        first_role_llm = {}
+        for role_details in config.get("roles", {}).values():
+            first_role_llm = role_details.get("llm", {}) or {}
+            break
+
+        # Priority: YAML top-level llm > first role llm > config_list > env vars
+        def _resolve(key, env_var=None, default=None):
+            return (yaml_llm.get(key) or first_role_llm.get(key)
+                    or model_config.get(key)
+                    or (os.environ.get(env_var) if env_var else None)
+                    or default)
+
+        api_type = _resolve("api_type", default="openai").lower()
+        model_name = _resolve("model", default="gpt-4o-mini")
+        api_key = _resolve("api_key", env_var="OPENAI_API_KEY")
+        base_url = (model_config.get("base_url")
+                    or yaml_llm.get("base_url")
+                    or os.environ.get("OPENAI_BASE_URL")
+                    or os.environ.get("OPENAI_API_BASE"))
+
+        # Build LLMConfig — Bedrock needs no api_key
+        if api_type == "bedrock":
+            llm_config_entry = {"api_type": "bedrock", "model": model_name}
+        else:
+            llm_config_entry = {"model": model_name}
+            if api_key:
+                llm_config_entry["api_key"] = api_key
+            if base_url and base_url not in ("https://api.openai.com/v1", "https://api.openai.com/v1/"):
+                llm_config_entry["base_url"] = base_url
+        llm_config = LLMConfig(llm_config_entry)
+
+        user_proxy = UserProxyAgent(
+            name="User",
+            human_input_mode="NEVER",
+            is_termination_msg=lambda x: "TERMINATE" in (x.get("content") or ""),
+            code_execution_config=False,
+        )
+
+        # Create one AssistantAgent per role
+        ag2_agent_entries = []
+        for role, details in config["roles"].items():
+            agent_name = details.get("role", role).replace("{topic}", topic)
+            backstory = details.get("backstory", "").replace("{topic}", topic)
+            agent_name_safe = _re.sub(r"[^a-zA-Z0-9_\-]", "_", agent_name)
+            assistant = AssistantAgent(
+                name=agent_name_safe,
+                system_message=backstory + "\nWhen the task is done, reply 'TERMINATE'.",
+                llm_config=llm_config,
+            )
+            ag2_agent_entries.append((role, details, assistant))
+
+        # Register tools via AG2 decorator pattern
+        for role, details, assistant in ag2_agent_entries:
+            for tool_name in details.get("tools", []):
+                tool = tools_dict.get(tool_name)
+                if tool is None:
+                    continue
+                func = tool if callable(tool) else getattr(tool, "run", None)
+                if func is None:
+                    continue
+
+                def make_tool_fn(f):
+                    def tool_fn(**kwargs):
+                        return f(**kwargs) if callable(f) else str(f)
+                    tool_fn.__name__ = tool_name
+                    return tool_fn
+
+                wrapped = make_tool_fn(func)
+                assistant.register_for_llm(description=f"Tool: {tool_name}")(wrapped)
+                user_proxy.register_for_execution()(wrapped)
+
+        all_assistants = [a for _, _, a in ag2_agent_entries]
+        if not all_assistants:
+            return "### AG2 Output ###\nNo agents created from configuration."
+
+        # Build initial message from all task descriptions
+        task_lines = []
+        for role, details, _ in ag2_agent_entries:
+            for task_name, task_details in details.get("tasks", {}).items():
+                desc = task_details.get("description", "").replace("{topic}", topic)
+                if desc:
+                    task_lines.append(desc)
+        initial_message = "\n".join(task_lines) if task_lines else topic
+
+        groupchat = GroupChat(
+            agents=[user_proxy] + all_assistants,
+            messages=[],
+            max_round=12,
+        )
+        manager = GroupChatManager(groupchat=groupchat, llm_config=llm_config)
+
+        try:
+            chat_result = user_proxy.initiate_chat(manager, message=initial_message)
+        except Exception as e:
+            return f"### AG2 Error ###\n{str(e)}"
+
+        # Prefer ChatResult.summary if available, otherwise scan messages
+        result_content = ""
+        summary = getattr(chat_result, "summary", None)
+        if summary and isinstance(summary, str) and summary.strip():
+            result_content = _re.sub(r'[\s\.\,]*TERMINATE[\s\.\,]*$', '', summary, flags=_re.IGNORECASE).strip().rstrip('.')
+
+        if not result_content:
+            for msg in reversed(groupchat.messages):
+                if msg.get("name") == "User":
+                    continue
+                content = (msg.get("content") or "").strip()
+                if content:
+                    result_content = _re.sub(r'[\s\.\,]*TERMINATE[\s\.\,]*$', '', content, flags=_re.IGNORECASE).strip().rstrip('.')
+                    if result_content:
+                        break
+
+        if not result_content:
+            result_content = "Task completed."
+
+        return f"### AG2 Output ###\n{result_content}"
+
     def _run_crewai(self, config, topic, tools_dict):
         """
         Run agents using the CrewAI framework.

src/praisonai/praisonai/auto.py

@@ -91,6 +91,23 @@ def _check_autogen_v4_available() -> bool:
     return _autogen_v4_available
 
 
+# --- AG2 lazy loading ---
+_ag2_available = None
+
+def _check_ag2_available() -> bool:
+    """Check if AG2 (community fork of AutoGen) is available (cached)."""
+    global _ag2_available
+    if _ag2_available is None:
+        try:
+            import importlib.metadata
+            importlib.metadata.distribution('ag2')
+            from autogen import LLMConfig  # noqa: F401 — AG2-exclusive class
+            _ag2_available = True
+        except Exception:
+            _ag2_available = False
+    return _ag2_available
+
+
 def _get_autogen():
     """Lazy load autogen module."""
     global _autogen_module
@@ -658,6 +675,11 @@ def __init__(self, topic="Movie Story writing about AI", agent_file="test.yaml",
 Praisonai is not installed. Please install with:
     pip install praisonaiagents
 """)
+        elif framework == "ag2" and not _check_ag2_available():
+            raise ImportError("""
+AG2 is not installed. Please install with:
+    pip install "praisonai[ag2]"
+""")
 
         # Only show tools message if using a framework and tools are needed
         if (framework in ["crewai", "autogen"]) and not _check_praisonai_tools_available():

src/praisonai/pyproject.toml

@@ -91,6 +91,7 @@ call = [
 train = []
 crewai = ["crewai>=0.157.0", "praisonai-tools>=0.1.0"]
 autogen = ["pyautogen==0.2.29", "praisonai-tools>=0.1.0", "crewai"]
+ag2 = ["ag2>=0.11.0", "praisonai-tools>=0.1.0"]
 autogen-v4 = [
     "autogen-agentchat>=0.4.0", 
     "autogen-ext[openai]>=0.4.0",