34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237 | class SimpleAgent(BaseAgent):
"""A simple agent with env, tools, mcps, and context manager, wrapped on openai-agents."""
def __init__(
self,
*,
config: AgentConfig | str | None = None, # use config to pass agent configs
name: str | None = None,
instructions: str | Callable | None = None,
model: str | Model | None = None,
model_settings: ModelSettings | None = None,
tools: list[Tool] = None,
output_type: type[Any] | AgentOutputSchemaBase | None = None,
tool_use_behavior: Literal["run_llm_again", "stop_on_first_tool"] | StopAtTools = "run_llm_again",
):
self.config = self._get_config(config)
if name:
self.config.agent.name = name
if instructions:
self.config.agent.instructions = instructions
self.model = self._get_model(self.config, model)
self.model_settings = self._get_model_settings(self.config, model_settings)
self.tools: list[Tool] = tools or []
self.output_type: type[Any] | AgentOutputSchemaBase | None = output_type
self.tool_use_behavior: Literal["run_llm_again", "stop_on_first_tool"] | StopAtTools = tool_use_behavior
self.context_manager: BaseContextManager = None
self.env: BaseEnv = None
self.current_agent: Agent[TContext] = None # move to task recorder?
self.input_items: list[TResponseInputItem] = []
self._run_hooks: RunHooks = None
self._mcp_servers: list[MCPServer] = []
self._toolkits: list[AsyncBaseToolkit] = []
self._mcps_exit_stack = AsyncExitStack()
self._tools_exit_stack = AsyncExitStack()
def _get_config(self, config: AgentConfig | str | None) -> AgentConfig:
if isinstance(config, AgentConfig):
return config
return ConfigLoader.load_agent_config(config or "base")
def _get_model(self, config: AgentConfig, model: str | Model | None = None) -> Model:
if isinstance(model, Model):
return model
model_provider_config = config.model.model_provider.model_dump()
if isinstance(model, str):
model_provider_config["model"] = model
return AgentsUtils.get_agents_model(**model_provider_config)
def _get_model_settings(self, config: AgentConfig, model_settings: ModelSettings | None = None) -> ModelSettings:
if isinstance(model_settings, ModelSettings):
return model_settings
return config.model.model_settings
async def __aenter__(self):
await self.build()
return self
async def __aexit__(self, exc_type, exc_val, exc_tb):
await self.cleanup()
async def build(self):
"""Build the agent"""
self.env = await get_env(self.config, "None") # FIXME: trace_id
await self.env.build()
self.current_agent = Agent(
name=self.config.agent.name,
instructions=self.config.agent.instructions,
model=self.model,
model_settings=self.model_settings,
tools=await self.get_tools(),
output_type=self.output_type,
tool_use_behavior=self.tool_use_behavior,
mcp_servers=self._mcp_servers,
)
self.context_manager = build_context_manager(self.config)
async def cleanup(self):
"""Cleanup"""
logger.info("Cleaning up MCP servers...")
await self._mcps_exit_stack.aclose()
self._mcp_servers = []
logger.info("Cleaning up tools...")
await self._tools_exit_stack.aclose()
self._toolkits = []
logger.info("Cleaning up env...")
await self.env.cleanup()
async def get_tools(self) -> list[Tool]:
if self.tools:
return self.tools
tools_list: list[Tool] = []
tools_list += await self.env.get_tools() # add env tools
# TODO: handle duplicate tool names
for _, toolkit_config in self.config.toolkits.items():
if toolkit_config.mode == "mcp":
await self._load_mcp_server(toolkit_config)
elif toolkit_config.mode == "builtin":
toolkit = await self._load_toolkit(toolkit_config)
tools_list.extend(await toolkit.get_tools_in_agents())
else:
raise ValueError(f"Unknown toolkit mode: {toolkit_config.mode}")
tool_names = [tool.name for tool in tools_list]
logger.info(f"Loaded {len(tool_names)} tools: {tool_names}")
self.tools = tools_list
return tools_list
async def _load_toolkit(self, toolkit_config: ToolkitConfig) -> AsyncBaseToolkit:
logger.info(f"Loading builtin toolkit `{toolkit_config.name}` with config {toolkit_config}")
toolkit = await self._tools_exit_stack.enter_async_context(TOOLKIT_MAP[toolkit_config.name](toolkit_config))
self._toolkits.append(toolkit)
return toolkit
async def _load_mcp_server(self, toolkit_config: ToolkitConfig) -> MCPServer:
logger.info(f"Loading MCP server `{toolkit_config.name}` with params {toolkit_config.config}")
server = await self._mcps_exit_stack.enter_async_context(
MCPServerStdio( # FIXME: support other types of servers
name=toolkit_config.name,
params=toolkit_config.config,
client_session_timeout_seconds=20,
)
)
self._mcp_servers.append(server)
return server
def _get_run_config(self) -> RunConfig:
run_config = RunConfig(
model=self.current_agent.model,
model_settings=self.config.model.model_settings,
workflow_name=self.config.agent.name,
)
return run_config
def _get_context(self) -> dict:
return {
"context_manager": self.context_manager,
"env": self.env,
}
# wrap `Runner` apis in @openai-agents
async def run(self, input: str | list[TResponseInputItem], trace_id: str = None) -> TaskRecorder:
trace_id = trace_id or AgentsUtils.gen_trace_id()
logger.info(f"> trace_id: {trace_id}")
task_recorder = TaskRecorder(input, trace_id)
run_kwargs = {
"starting_agent": self.current_agent,
"input": input,
"context": self._get_context(),
"max_turns": self.config.max_turns,
"hooks": self._run_hooks,
"run_config": self._get_run_config(),
}
if AgentsUtils.get_current_trace():
run_result = await Runner.run(**run_kwargs)
else:
with trace(workflow_name="simple_agent", trace_id=trace_id):
run_result = await Runner.run(**run_kwargs)
task_recorder.add_run_result(run_result)
task_recorder.set_final_output(run_result.final_output)
return task_recorder
def run_streamed(self, input: str | list[TResponseInputItem], trace_id: str = None) -> RunResultStreaming:
trace_id = trace_id or AgentsUtils.gen_trace_id()
logger.info(f"> trace_id: {trace_id}")
run_kwargs = {
"starting_agent": self.current_agent,
"input": input,
"context": self._get_context(),
"max_turns": self.config.max_turns,
"hooks": self._run_hooks,
"run_config": self._get_run_config(),
}
if AgentsUtils.get_current_trace():
return Runner.run_streamed(**run_kwargs)
else:
with trace(workflow_name="simple_agent", trace_id=trace_id):
return Runner.run_streamed(**run_kwargs)
# util apis
async def chat(self, input: str) -> RunResult:
# TODO: set "session-level" tracing for multi-turn chat
self.input_items.append({"content": input, "role": "user"})
# print(f"< {self.input_items}")
recorder = await self.run(self.input_items)
run_result = recorder.get_run_result()
AgentsUtils.print_new_items(run_result.new_items)
self.input_items = run_result.to_input_list()
self.current_agent = run_result.last_agent
return run_result
async def chat_streamed(self, input: str) -> RunResultStreaming:
self.input_items.append({"content": input, "role": "user"})
run_result_streaming = self.run_streamed(self.input_items)
await AgentsUtils.print_stream_events(run_result_streaming.stream_events())
self.input_items = run_result_streaming.to_input_list()
self.current_agent = run_result_streaming.last_agent
return run_result_streaming
def set_run_hooks(self, run_hooks: RunHooks):
# WIP
self._run_hooks = run_hooks
|