feat(agent): support single agent mode. (#67)

* upd: add support for main-agent only mode.

* add handle edge cases for sub_agents being null or {}

Author: JubSteven

config/agent_quickstart_single.yaml

@@ -0,0 +1,55 @@
+defaults:
+  - benchmark: gaia-validation
+  - override hydra/job_logging: none
+  - _self_  # Allow defining variables at the top of this file
+
+
+main_agent:
+  prompt_class: MainAgentPromptBoxedAnswer
+  llm: 
+    provider_class: "ClaudeOpenRouterClient"
+    model_name: "anthropic/claude-3.7-sonnet"
+    async_client: true
+    temperature: 0.3
+    top_p: 0.95
+    min_p: 0.0
+    top_k: -1
+    max_tokens: 32000
+    openrouter_api_key: "${oc.env:OPENROUTER_API_KEY,???}"
+    openrouter_base_url: "${oc.env:OPENROUTER_BASE_URL,https://openrouter.ai/api/v1}"
+    openrouter_provider: "anthropic"
+    disable_cache_control: false
+    keep_tool_result: -1
+    oai_tool_thinking: false
+  
+  tool_config: 
+    - tool-reasoning-os
+    - tool-searching
+    - tool-image-video-os
+    - tool-reading
+    - tool-code
+    - tool-audio-os
+
+  max_turns: -1  # Maximum number of turns for main agent execution
+  max_tool_calls_per_turn: 10  # Maximum number of tool calls per turn
+  
+  input_process:
+    hint_generation: false
+    hint_llm_base_url: "${oc.env:HINT_LLM_BASE_URL,https://api.openai.com/v1}"
+  output_process:
+    final_answer_extraction: false
+    final_answer_llm_base_url: "${oc.env:FINAL_ANSWER_LLM_BASE_URL,https://api.openai.com/v1}"
+
+  openai_api_key: "${oc.env:OPENAI_API_KEY,???}" # used for hint generation and final answer extraction
+  add_message_id: true
+  keep_tool_result: -1
+  chinese_context: "${oc.env:CHINESE_CONTEXT,false}"
+
+
+sub_agents: null
+
+
+# Can define some top-level or default parameters here
+output_dir: logs/
+data_dir: "${oc.env:DATA_DIR,data}"  # Points to where data is stored
+

src/core/orchestrator.py

@@ -40,10 +40,14 @@ async def wrapped():
         nonlocal cache
         if cache is None:
             # Only fetch tool definitions if not already cached
-            result = {
-                name: await tool_manager.get_all_tool_definitions()
-                for name, tool_manager in sub_agent_tool_managers.items()
-            }
+            # Handle empty sub_agent_tool_managers (single agent mode)
+            if not sub_agent_tool_managers:
+                result = {}
+            else:
+                result = {
+                    name: await tool_manager.get_all_tool_definitions()
+                    for name, tool_manager in sub_agent_tool_managers.items()
+                }
             cache = result
         return cache
 
@@ -409,6 +413,8 @@ async def run_sub_agent(
             )
 
         # Generate sub-agent system prompt
+        if not self.cfg.sub_agents or sub_agent_name not in self.cfg.sub_agents:
+            raise ValueError(f"Sub-agent {sub_agent_name} not found in configuration")
         sub_agent_prompt_instance = _load_agent_prompt_class(
             self.cfg.sub_agents[sub_agent_name].prompt_class
         )
@@ -761,7 +767,8 @@ async def run_main_agent(
 
         # 2. Get tool definitions
         tool_definitions = await self.main_agent_tool_manager.get_all_tool_definitions()
-        tool_definitions += expose_sub_agents_as_tools(self.cfg.sub_agents)
+        if self.cfg.sub_agents is not None and self.cfg.sub_agents:
+            tool_definitions += expose_sub_agents_as_tools(self.cfg.sub_agents)
         if not tool_definitions:
             logger.debug(
                 "Warning: No tool definitions found. LLM cannot use any tools."

src/core/pipeline.py

@@ -86,7 +86,7 @@ async def execute_task_pipeline(
 
         # Initialize sub agent LLM client
         # Require agent-specific LLM configuration for sub-agents
-        if cfg.sub_agents:
+        if cfg.sub_agents is not None and cfg.sub_agents:
             first_sub_agent = next(iter(cfg.sub_agents.values()))
             if hasattr(first_sub_agent, "llm") and first_sub_agent.llm is not None:
                 sub_agent_llm_client = LLMClient(
@@ -97,9 +97,8 @@ async def execute_task_pipeline(
                     "No LLM configuration found in sub-agent. Please ensure the agent configuration includes an LLM section."
                 )
         else:
-            raise ValueError(
-                "No sub agents defined. Please ensure the agent configuration includes sub-agent sections."
-            )
+            sub_agent_llm_client = None
+            logger.info("No sub agents defined, using main agent only for the task")
 
         # Initialize orchestrator
         orchestrator = Orchestrator(
@@ -181,15 +180,16 @@ def create_pipeline_components(cfg: DictConfig, logs_dir: str | None = None):
     )
 
     sub_agent_tool_managers = {}
-    for sub_agent in cfg.sub_agents:
-        sub_agent_mcp_server_configs, sub_agent_blacklist = (
-            create_mcp_server_parameters(cfg, cfg.sub_agents[sub_agent], logs_dir)
-        )
-        sub_agent_tool_manager = ToolManager(
-            sub_agent_mcp_server_configs,
-            tool_blacklist=sub_agent_blacklist,
-        )
-        sub_agent_tool_managers[sub_agent] = sub_agent_tool_manager
+    if cfg.sub_agents is not None and cfg.sub_agents:
+        for sub_agent in cfg.sub_agents:
+            sub_agent_mcp_server_configs, sub_agent_blacklist = (
+                create_mcp_server_parameters(cfg, cfg.sub_agents[sub_agent], logs_dir)
+            )
+            sub_agent_tool_manager = ToolManager(
+                sub_agent_mcp_server_configs,
+                tool_blacklist=sub_agent_blacklist,
+            )
+            sub_agent_tool_managers[sub_agent] = sub_agent_tool_manager
 
     # Create OutputFormatter
     output_formatter = OutputFormatter()