File size: 26,945 Bytes
82bf89e |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 |
from __future__ import annotations
import asyncio
import time
from typing import Any, Callable, Dict, List, Optional
from agents import custom_span, gen_trace_id, trace
from openai.types.responses import ResponseTextDeltaEvent
from pydantic import BaseModel, Field, ValidationError
from util import formate_message
from tools.search_tool import SimpleArticle
from utils.baseclass import ResearchRunner
from tools.detail_plan_agent import CoreSection
from tools.knowledge_gap_agent import (
KnowledgeGapOutput,
knowledge_gap_agent,
)
from tools.long_writer_agent import (
LongWriterOutput,
clean_json_response,
extract_from_failed_json,
)
from tools.thinking_agent import thinking_agent
from tools.tool_selector_agent import (
AgentSelectionPlan,
AgentTask,
tool_selector_agent,
)
from tools.writer_agent import writer_section_agent
from utils.schemas import TOOL_AGENTS, ToolAgentOutput, InputCallbackTool
from utils.parse_output import create_type_parser
from config_logger import logger
# logger = logging.getLogger(__name__)
class IterationData(BaseModel):
"""Data for a single iteration of the research loop."""
gap: str = Field(
description="The gap addressed in the iteration", default_factory=list
)
tool_calls: List[str] = Field(
description="The tool calls made", default_factory=list
)
findings: List[str] = Field(
description="The findings collected from tool calls", default_factory=list
)
thought: List[str] = Field(
description="The thinking done to reflect on the success of the iteration and next steps",
default_factory=list,
)
class Conversation(BaseModel):
"""A conversation between the user and the iterative researcher."""
history: List[IterationData] = Field(
description="The data for each iteration of the research loop",
default_factory=list,
)
def add_iteration(self, iteration_data: Optional[IterationData] = None):
if iteration_data is None:
iteration_data = IterationData()
self.history.append(iteration_data)
def set_latest_gap(self, gap: str):
self.history[-1].gap = gap
def set_latest_tool_calls(self, tool_calls: List[str]):
self.history[-1].tool_calls = tool_calls
def set_latest_findings(self, findings: List[str]):
self.history[-1].findings = findings
def set_latest_thought(self, thought: str):
self.history[-1].thought = thought
def get_latest_gap(self) -> str:
return self.history[-1].gap
def get_latest_tool_calls(self) -> List[str]:
return self.history[-1].tool_calls
def get_latest_findings(self) -> List[str]:
return self.history[-1].findings
def get_latest_thought(self) -> str:
return self.history[-1].thought
def get_all_findings(self) -> List[str]:
return [
finding
for iteration_data in self.history
for finding in iteration_data.findings
]
def compile_conversation_history(self) -> str:
"""Compile the conversation history into a string."""
conversation = ""
for iteration_num, iteration_data in enumerate(self.history):
conversation += f"[ITERATION {iteration_num + 1}]\n\n"
if iteration_data.thought:
conversation += f"{self.get_thought_string(iteration_num)}\n\n"
if iteration_data.gap:
conversation += f"{self.get_task_string(iteration_num)}\n\n"
if iteration_data.tool_calls:
conversation += f"{self.get_action_string(iteration_num)}\n\n"
if iteration_data.findings:
conversation += f"{self.get_findings_string(iteration_num)}\n\n"
return conversation
def get_task_string(self, iteration_num: int) -> str:
"""Get the task for the current iteration."""
if self.history[iteration_num].gap:
return f"<task>\nAddress this knowledge gap: {self.history[iteration_num].gap}\n</task>"
return ""
def get_action_string(self, iteration_num: int) -> str:
"""Get the action for the current iteration."""
if self.history[iteration_num].tool_calls:
joined_calls = "\n".join(self.history[iteration_num].tool_calls)
return (
"<action>\nCalling the following tools to address the knowledge gap:\n"
f"{joined_calls}\n</action>"
)
return ""
def get_findings_string(self, iteration_num: int) -> str:
"""Get the findings for the current iteration."""
if self.history[iteration_num].findings:
joined_findings = "\n\n".join(self.history[iteration_num].findings)
return f"<findings>\n{joined_findings}\n</findings>"
return ""
def get_thought_string(self, iteration_num: int) -> str:
"""Get the thought for the current iteration."""
if self.history[iteration_num].thought:
return f"<thought>\n{self.history[iteration_num].thought}\n</thought>"
return ""
def latest_task_string(self) -> str:
"""Get the latest task."""
return self.get_task_string(len(self.history) - 1)
def latest_action_string(self) -> str:
"""Get the latest action."""
return self.get_action_string(len(self.history) - 1)
def latest_findings_string(self) -> str:
"""Get the latest findings."""
return self.get_findings_string(len(self.history) - 1)
def latest_thought_string(self) -> str:
"""Get the latest thought."""
return self.get_thought_string(len(self.history) - 1)
class IterativeResearcher:
"""Manager for the iterative research workflow that conducts research on a topic or subtopic by running a continuous research loop."""
def __init__(
self,
max_iterations: int = 5,
max_time_minutes: int = 10,
verbose: bool = True,
tracing: bool = False,
thoughts_callback: Optional[Callable[[str], Any]] = None,
hooks=None,
u_id: str = "",
):
self.max_iterations: int = max_iterations
self.max_time_minutes: int = max_time_minutes
self.start_time: float = None
self.iteration: int = 0
self.conversation: Conversation = Conversation()
self.should_continue: bool = True
self.verbose: bool = verbose
self.tracing: bool = tracing
self.thoughts_callback = thoughts_callback
self.hooks = hooks
self.u_id = u_id
if thoughts_callback is None:
async def noop(x):
pass
self.thoughts_callback = noop
self.references = []
async def run(
self,
query: str,
output_length: str = "", # A text description of the desired output length, can be left blank
output_instructions: CoreSection = None,
# Instructions for the final report (e.g. don't include any headings, just a couple of paragraphs of text)
background_context: str = "",
) -> tuple[Any, List[str]]:
"""Run the deep research workflow for a given query."""
self.start_time = time.time()
if self.tracing:
trace_id = gen_trace_id()
workflow_trace = trace("iterative_researcher", trace_id=trace_id)
print(
f"View trace: https://platform.openai.com/traces/trace?trace_id={trace_id}"
)
workflow_trace.start(mark_as_current=True)
# await self._log_message("=== Starting Iterative Research Workflow ===")
# Iterative research loop
while self.should_continue and self._check_constraints():
is_constraints = self._check_constraints()
# print(f"max_iteration:{self.max_iterations},now iteration is {self.iteration}")
self.iteration += 1
# await self._log_message(f"\n=== Starting Iteration {self.iteration} ===")
# Set up blank IterationData for this iteration
self.conversation.add_iteration()
# await self._log_message(f"Query is {query}")
# 1. Generate observations
observations: str = await self._generate_observations(
query, background_context=background_context
)
# await self._log_message(f"Observations is {observations}")
# 2. Evaluate current gaps in the research
evaluation: KnowledgeGapOutput = await self._evaluate_gaps(
query, background_context=background_context
)
# await self._log_message(f"Observations gaps is {evaluation.outstanding_gaps}")
# await self._log_message(f"Observations research_complete is {evaluation.research_complete}")
# Check if we should continue or break the loop
if not evaluation.research_complete:
next_gap = evaluation.outstanding_gaps[0]
# 3. Select agents to address knowledge gap
selection_plan: AgentSelectionPlan = await self._select_agents(
next_gap, query, background_context=background_context
)
# await self._log_message(
# f"Selection_plan.tasks:{selection_plan.tasks}\n"
# )
# 4. Run the selected agents to gather information
results: Dict[str, ToolAgentOutput] = await self._execute_tools(
selection_plan.tasks
)
# await self._log_message(f"Execute_tool_results : {results}")
else:
self.should_continue = False
# await self._log_message(
# "=== IterativeResearcher Marked As Complete - Finalizing Output ==="
# )
# if not self._check_constraints():
# await self._log_message("\n=== Ending Research Loop ===")
# Create final report # outline is the final_detailed_outline
# report = await self._create_final_report(
# query, length=output_length, instructions=output_instructions
# )
report = await self._create_review_section(
query,
length=output_length,
instructions=output_instructions,
)
# check_section = await self._check_section(report)
# elapsed_time = time.time() - self.start_time
# await self._log_message(
# f"IterativeResearcher completed in {int(elapsed_time // 60)} minutes and {int(elapsed_time % 60)} seconds after {self.iteration} iterations."
# )
if self.tracing:
workflow_trace.finish(reset_current=True)
return report, self.references
def _check_constraints(self) -> bool:
"""Check if we've exceeded our constraints (max iterations or time)."""
if self.iteration >= self.max_iterations:
# self._log_message("\n=== Ending Research Loop ===")
# self._log_message(f"Reached maximum iterations ({self.max_iterations})")
return False
elapsed_minutes = (time.time() - self.start_time) / 60
if elapsed_minutes >= self.max_time_minutes:
# self._log_message("\n=== Ending Research Loop ===")
# self._log_message(f"Reached maximum time ({self.max_time_minutes} minutes)")
return False
return True
async def _evaluate_gaps(
self, query: str, background_context: str = ""
) -> KnowledgeGapOutput:
"""Evaluate the current state of research and identify knowledge gaps."""
background = (
f"BACKGROUND CONTEXT:\n{background_context}" if background_context else ""
)
input_str = f"""
Current Iteration Number: {self.iteration}
Time Elapsed: {(time.time() - self.start_time) / 60:.2f} minutes of maximum {self.max_time_minutes} minutes
ORIGINAL QUERY:
{query}
{background}
HISTORY OF ACTIONS, FINDINGS AND THOUGHTS:
{self.conversation.compile_conversation_history() or "No previous actions, findings or thoughts available."}
"""
result = await ResearchRunner.run(
knowledge_gap_agent, input_str, hooks=self.hooks
)
evaluation = result.final_output_as(KnowledgeGapOutput)
if not evaluation.research_complete:
next_gap = evaluation.outstanding_gaps[0]
self.conversation.set_latest_gap(next_gap)
# await self._log_message(self.conversation.latest_task_string())
return evaluation
async def _select_agents(
self, gap: str, query: str, background_context: str = ""
) -> AgentSelectionPlan:
"""Select agents to address the identified knowledge gap."""
background = (
f"BACKGROUND CONTEXT:\n{background_context}" if background_context else ""
)
input_str = f"""
ORIGINAL QUERY:
{query}
KNOWLEDGE GAP TO ADDRESS:
{gap}
{background}
HISTORY OF ACTIONS, FINDINGS AND THOUGHTS:
{self.conversation.compile_conversation_history() or "No previous actions, findings or thoughts available."}
"""
result = await ResearchRunner.run(
tool_selector_agent,
input_str,
hooks=self.hooks,
)
selection_plan = result.final_output_as(AgentSelectionPlan)
# Add the tool calls to the conversation
self.conversation.set_latest_tool_calls(
[
f"[Agent] {task.agent} [Query] {task.query} [Entity] {task.entity_website if task.entity_website else 'null'}"
for task in selection_plan.tasks
]
)
# await self._log_message(self.conversation.latest_action_string())
return selection_plan
async def _execute_tools(
self, tasks: List[AgentTask]
) -> Dict[str, ToolAgentOutput]:
"""Execute the selected tools concurrently to gather information."""
with custom_span("Execute Tool Agents"):
# Create a task for each agent
async_tasks = []
sem = asyncio.Semaphore(1) # Limit concurrency to 5
async def limited_task(task):
async with sem: # Acquire semaphore on entry, release on exit
return await self._run_agent_task(task)
for task in tasks:
# await self._log_message(f"\ntask is runing: {task} \n")
await self._log_message(
formate_message(
type="search", message=f"Searching articles by {task.query}..."
)
)
async_tasks.append(limited_task(task))
# Run all tasks concurrently
num_completed = 0
results = {}
for future in asyncio.as_completed(async_tasks):
gap, agent_name, result = await future
results[f"{agent_name}_{gap}"] = result
num_completed += 1
# await self._log_message(
# f"<processing>\nTool execution progress: {num_completed}/{len(async_tasks)}\n</processing>"
# )
# Add findings from the tool outputs to the conversation
findings = []
for tool_output in results.values():
findings.append(tool_output.output)
self.conversation.set_latest_findings(findings)
return results
async def _run_agent_task(
self, task: AgentTask
) -> tuple[str, str, ToolAgentOutput]:
"""Run a single agent task and return the result."""
try:
agent_name = task.agent
agent = TOOL_AGENTS.get(agent_name)
if agent:
# result = await ResearchRunner.run(
# agent,
# task.model_dump_json(),
# hooks=self.hooks,
# )
# output = result.final_output_as(ToolAgentOutput)
## stream-output
# await self._log_message(
# formate_message(
# type="search",
# message="Searching articles by Articles_search_tool...",
# )
# )
input_call = InputCallbackTool(
# thoughts_callback=self.thoughts_callback,
u_id=str(self.u_id),
is_pkb=False,
results_callback=self.thoughts_callback,
# c_id=str(c_id),
)
synthesis_streamed_result = ResearchRunner.run_streamed(
agent,
task.model_dump_json(),
context=input_call,
hooks=self.hooks,
)
full_response = ""
def get_references(articles: List[SimpleArticle]):
for article in articles:
self.references.append(f"<{article.hash_id}> {article.source}")
async for event in synthesis_streamed_result.stream_events():
if event.type == "raw_response_event" and isinstance(
event.data, ResponseTextDeltaEvent
):
token = event.data.delta
full_response += token
elif event.type == "run_item_stream_event":
if event.item.type == "tool_call_output_item":
tool_call_output = event.item.output
# print(f"########## tool_call_output {tool_call_output}")
# await self._log_message(f"########## tool_call_output {type(tool_call_output)},isinstance {isinstance(tool_call_output,list )}")
if (
isinstance(tool_call_output, list)
and len(tool_call_output) > 0
and isinstance(tool_call_output[0], SimpleArticle)
):
get_references(tool_call_output)
# print(f"########## referencfull_responsees {full_response}")
result = ToolAgentOutput(output=full_response, sources=[])
# Extract ToolAgentOutput from RunResult
output = result
else:
output = ToolAgentOutput(
output=f"No implementation found for agent {agent_name}", sources=[]
)
return task.gap, agent_name, output
except Exception as e:
error_output = ToolAgentOutput(
output=f"Error executing {task.agent} for gap '{task.gap}': {str(e)}",
sources=[],
)
return task.gap, task.agent, error_output
async def _generate_observations(
self, query: str, background_context: str = ""
) -> str:
"""Generate observations from the current state of the research."""
background = (
f"BACKGROUND CONTEXT:\n{background_context}" if background_context else ""
)
input_str = f"""
ORIGINAL QUERY:
{query}
{background}
HISTORY OF ACTIONS, FINDINGS AND THOUGHTS:
{self.conversation.compile_conversation_history() or "No previous actions, findings or thoughts available."}
"""
result = await ResearchRunner.run(thinking_agent, input_str, hooks=self.hooks)
# Add the observations to the conversation
observations = result.final_output
self.conversation.set_latest_thought(observations)
# await self._log_message(self.conversation.latest_thought_string())
return observations
# async def _create_final_report(
# self, query: str, length: str = "", instructions: str = ""
# ) -> str:
# """Create the final response from the completed draft."""
# # await self._log_message("=== Drafting Final Response ===")
# length_str = (
# f"* The full response should be approximately {length}.\n" if length else ""
# )
# instructions_str = f"* {instructions}" if instructions else ""
# guidelines_str = (
# ("\n\nGUIDELINES:\n" + length_str + instructions_str).strip("\n")
# if length or instructions
# else ""
# )
# all_findings = (
# "\n\n".join(self.conversation.get_all_findings())
# or "No findings available yet."
# )
# input_str = f"""
# Provide a response based on the query and findings below with as much detail as possible. {guidelines_str}
# QUERY: {query}
# FINDINGS:
# {all_findings}
# """
# # await self._log_message(
# # input_str
# # )
# # result = await ResearchRunner.run(
# # writer_agent,
# # input_str,
# # )
# # return result.final_output
# # await self._log_message(
# # formate_message(
# # type="file", message="Generating final report by writer_agent..."
# # )
# # )
# ## use the stream response
# synthesis_streamed_result = ResearchRunner.run_streamed(
# starting_agent=writer_agent, input=input_str
# )
# full_response = ""
# try:
# async for event in synthesis_streamed_result.stream_events():
# # Check for cancellation
# # if stop_event and stop_event.is_set():
# # await thoughts_callback("Operation cancelled during synthesis")
# # return "Operation cancelled"
# # Process different event types
# if event.type == "raw_response_event" and isinstance(
# event.data, ResponseTextDeltaEvent
# ):
# token = event.data.delta
# full_response += token
# # Stream token to the results callback
# # Stream agent updates
# # elif event.type == "agent_updated_stream_event":
# # await self._log_message(f"Agent updated: {event.new_agent.name}")
# # await self._log_message(
# # "\nFinal response from IterativeResearcher created successfully\n"
# # )
# except Exception as e:
# logger.error(f"IterativeResearcher create report error: {e} ")
# logger.info(f"#############all_findings: {len(self.conversation.get_all_findings())} \n ####full_response: {full_response[:100]}")
# return full_response
async def _log_message(self, message: str) -> None:
"""Log a message if verbose is True"""
if self.verbose:
# if self.thoughts_callback:
await self.thoughts_callback(message)
else:
print(message)
async def _create_review_section(
self,
query: str,
length: str = "",
instructions: CoreSection = None,
) -> LongWriterOutput:
length_str = (
f"* The full response should be approximately {length}.\n" if length else ""
)
instructions_str = f"* {instructions}" if instructions else ""
guidelines_str = (
("\n\nGUIDELINES:\n" + length_str + instructions_str).strip("\n")
if length or instructions
else ""
)
all_findings = (
"\n\n".join(self.conversation.get_all_findings())
or "No findings available yet."
)
input_str = f"""
Provide a response based on the query and findings below with as much detail as possible.
SECTION OUTLINE:
{instructions.description}
SECTION Title"
{instructions.title}
RAW QUERY: {query}
FINDINGS:
{all_findings}
"""
max_iter = 3
iter_num = 0
temp_agent_type = ""
while iter_num < max_iter:
full_response = ""
try:
result = ResearchRunner.run_streamed(
starting_agent=writer_section_agent, input=input_str
)
async for event in result.stream_events():
# Process different event types
if event.type == "raw_response_event" and isinstance(
event.data, ResponseTextDeltaEvent
):
full_response += event.data.delta
elif event.type == "agent_updated_stream_event":
if event.new_agent.name != temp_agent_type:
temp_agent_type = event.new_agent.name
final_response = result.final_output
try:
cleaned_response = clean_json_response(final_response)
resf = create_type_parser(LongWriterOutput)
res = resf(cleaned_response)
return res
except Exception as parse_error:
# If JSON parsing fails, try manual extraction
logger.warning(
f"Failed to parse output as JSON in write_next_section ,try extract from failed json: {str(parse_error)[:200]}"
)
try:
manual_result = extract_from_failed_json(full_response)
if manual_result:
return manual_result
except Exception as manual_error:
logger.error(
f"Manual extraction also failed: {str(manual_error)[:100]}"
)
# Increment iteration counter and continue the loop instead of returning empty references
iter_num += 1
logger.error(
f"Parse error occurred: {parse_error}. Retrying {iter_num}/{max_iter}..."
)
continue
except ValidationError:
resf = create_type_parser(LongWriterOutput)
res = resf(full_response)
return res
except Exception as e:
logger.error(f"Write review section error: {e}")
iter_num += 1
logger.error(f"Error occurred: {e}. Retrying {iter_num}/{max_iter}...")
# If all retries fail, return an error output
return LongWriterOutput(
next_section_markdown="The section generate error", references=[]
)
|