File size: 19,513 Bytes
02d640a
 
b07e47b
02d640a
b07e47b
02d640a
b07e47b
763898d
4f8a105
b07e47b
ffea40d
763898d
 
 
b07e47b
7b71415
 
763898d
 
 
ffea40d
 
69af1c5
 
ffea40d
 
 
763898d
ffea40d
763898d
7b71415
 
 
 
 
 
763898d
7b71415
 
 
763898d
 
7b71415
 
763898d
7b71415
 
 
 
 
763898d
7b71415
 
 
4f8a105
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ffea40d
4f8a105
ffea40d
4f8a105
763898d
 
 
 
 
 
 
4f8a105
69af1c5
4f8a105
 
 
ffea40d
 
763898d
ffea40d
763898d
69af1c5
ffea40d
763898d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ffea40d
 
4f8a105
763898d
 
 
ffea40d
4f8a105
763898d
4f8a105
ffea40d
69af1c5
4f8a105
 
 
ffea40d
 
4f8a105
763898d
 
 
ffea40d
763898d
 
4f8a105
b07e47b
69af1c5
4f8a105
 
763898d
4f8a105
763898d
 
 
 
 
 
 
 
 
4f8a105
ffea40d
69af1c5
4f8a105
763898d
 
 
69af1c5
763898d
 
4f8a105
69af1c5
 
4f8a105
 
 
69af1c5
4f8a105
763898d
 
 
69af1c5
763898d
 
4f8a105
69af1c5
642e9cc
4f8a105
 
 
69af1c5
ffea40d
4f8a105
763898d
 
 
ffea40d
763898d
4f8a105
b07e47b
69af1c5
4f8a105
 
 
 
763898d
ffea40d
4f8a105
763898d
4f8a105
763898d
4f8a105
 
 
763898d
4f8a105
 
 
 
 
 
 
763898d
 
 
4f8a105
 
 
 
763898d
4f8a105
 
 
763898d
4f8a105
 
 
 
 
 
 
 
 
 
 
 
763898d
4f8a105
763898d
 
 
 
 
4f8a105
 
 
 
 
 
763898d
 
 
 
 
 
4f8a105
 
 
763898d
4f8a105
 
 
 
 
 
763898d
 
 
 
ffea40d
763898d
ffea40d
4f8a105
763898d
4f8a105
 
763898d
 
 
4f8a105
 
 
763898d
4f8a105
763898d
4f8a105
763898d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4f8a105
763898d
4f8a105
763898d
 
 
ffea40d
69af1c5
4f8a105
ffea40d
69af1c5
763898d
 
 
 
ffea40d
763898d
ffea40d
763898d
 
ffea40d
 
763898d
ffea40d
763898d
b07e47b
 
763898d
 
 
ffea40d
763898d
ffea40d
763898d
4b2e1da
763898d
 
 
 
 
4b2e1da
763898d
 
 
 
69af1c5
 
39c710d
69af1c5
 
 
763898d
39c710d
763898d
69af1c5
39c710d
 
4f8a105
ffea40d
4f8a105
763898d
 
 
 
ffea40d
 
69af1c5
 
 
39c710d
763898d
 
 
4f8a105
763898d
b07e47b
ffea40d
69af1c5
ffea40d
 
763898d
 
 
 
69af1c5
763898d
 
 
 
 
 
 
ffea40d
b07e47b
763898d
 
69af1c5
b07e47b
69af1c5
4b2e1da
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
import os
import asyncio
import logging
import threading
import queue
import gradio as gr
import httpx
from typing import Generator, Any, Dict, List, Optional
from functools import lru_cache

# -------------------- Configuration --------------------
logging.basicConfig(
    level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s"
)

# -------------------- External Model Call (with Caching and Retry) --------------------
async def call_model(prompt: str, model: str = "gpt-4o", api_key: str = None, max_retries: int = 3) -> str:
    """
    Sends a prompt to the OpenAI API endpoint with retries and exponential backoff.
    """
    if api_key is None:
        api_key = os.getenv("OPENAI_API_KEY")
        if api_key is None:
            raise ValueError("OpenAI API key not found.")
    url = "https://api.openai.com/v1/chat/completions"
    headers = {
        "Authorization": f"Bearer {api_key}",
        "Content-Type": "application/json",
    }
    payload = {"model": model, "messages": [{"role": "user", "content": prompt}]}

    for attempt in range(max_retries):
        try:
            async with httpx.AsyncClient(timeout=httpx.Timeout(300.0)) as client:
                response = await client.post(url, headers=headers, json=payload)
                response.raise_for_status()
                response_json = response.json()  # Synchronous parsing is acceptable here
                return response_json["choices"][0]["message"]["content"]
        except httpx.HTTPStatusError as e:
            logging.error(f"HTTP error (attempt {attempt + 1}/{max_retries}): {e}")
            if e.response.status_code in (502, 503, 504):
                await asyncio.sleep(2 ** attempt)
                continue
            else:
                raise
        except httpx.RequestError as e:
            logging.error(f"Request error (attempt {attempt + 1}/{max_retries}): {e}")
            await asyncio.sleep(2 ** attempt)
            continue
        except Exception as e:
            logging.error(f"Unexpected error (attempt {attempt+1}/{max_retries}): {e}")
            raise
    raise Exception(f"Failed to get response from OpenAI API after {max_retries} attempts.")

# -------------------- Shared Context --------------------
class Context:
    def __init__(self, original_task: str, optimized_task: Optional[str] = None,
                 plan: Optional[str] = None, code: Optional[str] = None,
                 review_comments: Optional[List[Dict[str, str]]] = None,
                 test_cases: Optional[str] = None, test_results: Optional[str] = None,
                 documentation: Optional[str] = None, conversation_history: Optional[List[Dict[str, str]]] = None):
        self.original_task = original_task
        self.optimized_task = optimized_task
        self.plan = plan
        self.code = code
        self.review_comments = review_comments or []
        self.test_cases = test_cases
        self.test_results = test_results
        self.documentation = documentation
        self.conversation_history = conversation_history or []

    def add_conversation_entry(self, agent_name: str, message: str):
        self.conversation_history.append({"agent": agent_name, "message": message})

# -------------------- Agent Classes --------------------

class PromptOptimizerAgent:
    async def optimize_prompt(self, context: Context, api_key: str) -> Context:
        """
        Optimizes the user’s original prompt.
        """
        system_prompt = (
            "Improve the prompt. Be clear, specific, and complete. "
            "Keep original intent. Return ONLY the revised prompt."
        )
        full_prompt = f"{system_prompt}\n\nUser's prompt:\n{context.original_task}"
        optimized = await call_model(full_prompt, model="gpt-4o", api_key=api_key)
        context.optimized_task = optimized
        context.add_conversation_entry("Prompt Optimizer", f"Optimized Task:\n{optimized}")
        return context

class OrchestratorAgent:
    def __init__(self, log_queue: queue.Queue, human_event: threading.Event, human_input_queue: queue.Queue) -> None:
        self.log_queue = log_queue
        self.human_event = human_event
        self.human_input_queue = human_input_queue

    async def generate_plan(self, context: Context, api_key: str) -> Context:
        """
        Generates (or revises) a plan using human feedback if necessary.
        Uses an iterative approach instead of recursion.
        """
        while True:
            if context.plan:
                prompt = (
                    f"You are a planner. Revise/complete the plan for '{context.original_task}' using feedback:\n"
                    f"{context.plan}\n\n"
                    "If unsure, output 'REQUEST_HUMAN_FEEDBACK\\n[Question]'"
                )
            else:
                prompt = (
                    f"You are a planner. Create a plan for: '{context.optimized_task}'. "
                    "Break down the task and assign sub-tasks to: Coder, Code Reviewer, Quality Assurance Tester, and Documentation Agent. "
                    "Include review/revision steps, error handling, and documentation instructions.\n\n"
                    "If unsure, output 'REQUEST_HUMAN_FEEDBACK\\n[Question]'"
                )

            plan = await call_model(prompt, model="gpt-4o", api_key=api_key)
            context.add_conversation_entry("Orchestrator", f"Plan:\n{plan}")

            # Check if human feedback is requested.
            if "REQUEST_HUMAN_FEEDBACK" in plan:
                question = plan.split("REQUEST_HUMAN_FEEDBACK\n", 1)[1].strip()
                self.log_queue.put("[Orchestrator]: Requesting human feedback...")
                self.log_queue.put(f"[Orchestrator]: Question for human: {question}")

                # Prepare feedback context and trigger the human feedback event.
                feedback_request_context = (
                    f"The orchestrator agent is requesting feedback on the following task:\n"
                    f"**{context.optimized_task}**\n\n"
                    f"Current plan:\n**{context.plan or 'None'}**\n\n"
                    f"Question:\n**{question}**"
                )
                self.human_event.set()
                # Pass the context to the human input handler.
                self.human_input_queue.put(feedback_request_context)
                human_response = self.human_input_queue.get()  # Blocking call for human response.
                self.human_event.clear()

                self.log_queue.put(f"[Orchestrator]: Received human feedback: {human_response}")
                # Incorporate human feedback into the plan and loop again.
                context.plan = context.plan + "\n" + human_response if context.plan else human_response
            else:
                context.plan = plan
                break  # Exit loop when no feedback is requested.
        return context

class CoderAgent:
    async def generate_code(self, context: Context, api_key: str, model: str = "gpt-4o") -> Context:
        """
        Generates code based on the provided plan.
        """
        prompt = (
            "You are a coding agent. Output ONLY the code. "
            "Adhere to best practices and include error handling.\n\n"
            f"Instructions:\n{context.plan}"
        )
        code = await call_model(prompt, model=model, api_key=api_key)
        context.code = code
        context.add_conversation_entry("Coder", f"Code:\n{code}")
        return context

class CodeReviewerAgent:
    async def review_code(self, context: Context, api_key: str) -> Context:
        """
        Reviews the generated code and returns either actionable feedback or 'APPROVE'.
        """
        prompt = (
            "You are a code reviewer. Provide CONCISE feedback focusing on correctness, efficiency, readability, error handling, and security. "
            "If the code is acceptable, respond with ONLY 'APPROVE'. Do NOT generate code.\n\n"
            f"Task: {context.optimized_task}\n\nCode:\n{context.code}"
        )
        review = await call_model(prompt, model="gpt-4o", api_key=api_key)
        context.add_conversation_entry("Code Reviewer", f"Review:\n{review}")

        # Check for approval; if not approved, parse feedback.
        if "APPROVE" not in review.upper():
            structured_review = {"comments": []}
            for line in review.splitlines():
                if line.strip():
                    structured_review["comments"].append({
                        "issue": line.strip(),
                        "line_number": "N/A",
                        "severity": "Medium"
                    })
            context.review_comments.append(structured_review)
        return context

class QualityAssuranceTesterAgent:
    async def generate_test_cases(self, context: Context, api_key: str) -> Context:
        """
        Generates test cases considering edge and error cases.
        """
        prompt = (
            "You are a testing agent. Generate comprehensive test cases considering edge cases and error scenarios. "
            "Output in a clear format.\n\n"
            f"Task: {context.optimized_task}\n\nCode:\n{context.code}"
        )
        test_cases = await call_model(prompt, model="gpt-4o", api_key=api_key)
        context.test_cases = test_cases
        context.add_conversation_entry("QA Tester", f"Test Cases:\n{test_cases}")
        return context

    async def run_tests(self, context: Context, api_key: str) -> Context:
        """
        Runs the generated test cases and compares expected vs. actual outcomes.
        """
        prompt = (
            "Run the test cases. Compare actual vs expected outputs and state any discrepancies. "
            "If all tests pass, output 'TESTS PASSED'.\n\n"
            f"Code:\n{context.code}\n\nTest Cases:\n{context.test_cases}"
        )
        test_results = await call_model(prompt, model="gpt-4o", api_key=api_key)
        context.test_results = test_results
        context.add_conversation_entry("QA Tester", f"Test Results:\n{test_results}")
        return context

class DocumentationAgent:
    async def generate_documentation(self, context: Context, api_key: str) -> Context:
        """
        Generates concise documentation including a --help message.
        """
        prompt = (
            "Generate clear documentation including a brief description, explanation, and a --help message.\n\n"
            f"Code:\n{context.code}"
        )
        documentation = await call_model(prompt, model="gpt-4o", api_key=api_key)
        context.documentation = documentation
        context.add_conversation_entry("Documentation Agent", f"Documentation:\n{documentation}")
        return context

# -------------------- Agent Dispatcher --------------------

class AgentDispatcher:
    def __init__(self, log_queue: queue.Queue, human_event: threading.Event, human_input_queue: queue.Queue):
        self.log_queue = log_queue
        self.human_event = human_event
        self.human_input_queue = human_input_queue
        self.agents = {
            "prompt_optimizer": PromptOptimizerAgent(),
            "orchestrator": OrchestratorAgent(log_queue, human_event, human_input_queue),
            "coder": CoderAgent(),
            "code_reviewer": CodeReviewerAgent(),
            "qa_tester": QualityAssuranceTesterAgent(),
            "documentation_agent": DocumentationAgent(),
        }

    async def dispatch(self, agent_name: str, context: Context, api_key: str, **kwargs) -> Context:
        """
        Dispatches the task to the specified agent.
        """
        agent = self.agents.get(agent_name)
        if not agent:
            raise ValueError(f"Unknown agent: {agent_name}")

        self.log_queue.put(f"[{agent_name.replace('_', ' ').title()}]: Starting task...")
        if agent_name == "prompt_optimizer":
            context = await agent.optimize_prompt(context, api_key)
        elif agent_name == "orchestrator":
            context = await agent.generate_plan(context, api_key)
        elif agent_name == "coder":
            context = await agent.generate_code(context, api_key, **kwargs)
        elif agent_name == "code_reviewer":
            context = await agent.review_code(context, api_key)
        elif agent_name == "qa_tester":
            if kwargs.get("generate_tests", False):
                context = await agent.generate_test_cases(context, api_key)
            elif kwargs.get("run_tests", False):
                context = await agent.run_tests(context, api_key)
        elif agent_name == "documentation_agent":
            context = await agent.generate_documentation(context, api_key)
        else:
            raise ValueError(f"Unknown Agent Name: {agent_name}")
        return context

    async def determine_next_agent(self, context: Context, api_key: str) -> str:
        """
        Determines the next agent to run based on the current context.
        """
        if not context.optimized_task:
            return "prompt_optimizer"
        if not context.plan:
            return "orchestrator"
        if not context.code:
            return "coder"
        # Check if any review comment lacks an APPROVE.
        if not any(
            "APPROVE" in comment.get("issue", "").upper()
            for review in context.review_comments
            for comment in review.get("comments", [])
        ):
            return "code_reviewer"
        if not context.test_cases:
            return "qa_tester"
        if not context.test_results or "TESTS PASSED" not in context.test_results.upper():
            return "qa_tester"
        if not context.documentation:
            return "documentation_agent"

        return "done"  # All tasks are complete

# -------------------- Multi-Agent Conversation --------------------

async def multi_agent_conversation(task_message: str, log_queue: queue.Queue, api_key: str,
                                   human_event: threading.Event, human_input_queue: queue.Queue) -> None:
    """
    Orchestrates the multi-agent conversation.
    """
    context = Context(original_task=task_message)
    dispatcher = AgentDispatcher(log_queue, human_event, human_input_queue)

    next_agent = await dispatcher.determine_next_agent(context, api_key)
    # Prevent endless revisions by tracking coder iterations.
    coder_iterations = 0

    while next_agent != "done":
        if next_agent == "qa_tester":
            if not context.test_cases:
                context = await dispatcher.dispatch(next_agent, context, api_key, generate_tests=True)
            else:
                context = await dispatcher.dispatch(next_agent, context, api_key, run_tests=True)
        elif next_agent == "coder" and (context.review_comments or context.test_results):
            coder_iterations += 1
            # Switch to a different model after the first iteration.
            context = await dispatcher.dispatch(next_agent, context, api_key, model="gpt-3.5-turbo-16k")
        else:
            context = await dispatcher.dispatch(next_agent, context, api_key)
        
        # Check for approval in code review if applicable.
        if next_agent == "code_reviewer":
            approved = any(
                "APPROVE" in comment.get("issue", "").upper()
                for review in context.review_comments
                for comment in review.get("comments", [])
            )
            if not approved:
                # If not approved, we continue with coder to improve the code.
                next_agent = "coder"
            else:
                next_agent = await dispatcher.determine_next_agent(context, api_key)
        else:
            next_agent = await dispatcher.determine_next_agent(context, api_key)

        if next_agent == "coder" and coder_iterations > 5:
            log_queue.put("Maximum revision iterations reached. Exiting.")
            break

    log_queue.put("Conversation complete.")
    log_queue.put(("result", context.conversation_history))

# -------------------- Process Generator and Human Input --------------------

def process_conversation_generator(task_message: str, api_key: str,
                                   human_event: threading.Event, human_input_queue: queue.Queue,
                                   log_queue: queue.Queue) -> Generator[str, None, None]:
    """
    Runs the conversation and yields log messages.
    """
    # Run the conversation asynchronously.
    asyncio.run(multi_agent_conversation(task_message, log_queue, api_key, human_event, human_input_queue))

    final_result = None
    while True:
        try:
            msg = log_queue.get_nowait()
            if isinstance(msg, tuple) and msg[0] == "result":
                final_result = msg[1]
                yield gr.Chatbot.update(value=final_result, visible=True)
                yield "Conversation complete."
                break
            else:
                yield msg
        except queue.Empty:
            pass

        # If human feedback is requested, yield an appropriate message.
        if human_event.is_set():
            yield "Waiting for human feedback..."
        # Use a short asynchronous sleep to avoid busy-waiting.
        asyncio.run(asyncio.sleep(0.1))

def get_human_feedback(placeholder_text: str, human_input_queue: queue.Queue) -> gr.Blocks:
    """
    Constructs the Gradio interface to collect human feedback.
    """
    with gr.Blocks() as human_feedback_interface:
        with gr.Row():
            human_input = gr.Textbox(lines=4, label="Human Feedback", placeholder=placeholder_text)
        with gr.Row():
            submit_button = gr.Button("Submit Feedback")

        def submit_feedback(input_text: str):
            human_input_queue.put(input_text)
            return ""

        submit_button.click(fn=submit_feedback, inputs=human_input, outputs=human_input)
    return human_feedback_interface

# -------------------- Chat Function for Gradio --------------------

def multi_agent_chat(message: str, history: List[Any], openai_api_key: str = None) -> Generator[Any, None, None]:
    """
    Gradio chat function that runs the multi-agent conversation.
    """
    if not openai_api_key:
        openai_api_key = os.getenv("OPENAI_API_KEY")
        if not openai_api_key:
            yield "Error: API key not provided."
            return

    human_event = threading.Event()
    human_input_queue = queue.Queue()
    log_queue = queue.Queue()

    yield from process_conversation_generator(message, openai_api_key, human_event, human_input_queue, log_queue)

# -------------------- Launch the Chatbot --------------------

iface = gr.ChatInterface(
    fn=multi_agent_chat,
    chatbot=gr.Chatbot(type="messages"),
    additional_inputs=[
        gr.Textbox(label="OpenAI API Key (optional)", type="password", placeholder="Leave blank to use env variable")
    ],
    title="Multi-Agent Task Solver with Human-in-the-Loop",
    description=(
        "- Collaborative workflow with Human-in-the-Loop.\n"
        "- Orchestrator can ask for human feedback.\n"
        "- Enter a task; agents will work on it. You may be prompted for input.\n"
        "- Max 5 revisions.\n"
        "- Provide API Key."
    )
)

# Dummy interface to prevent Gradio errors.
dummy_iface = gr.Interface(lambda x: x, "textbox", "textbox")

if __name__ == "__main__":
    demo = gr.TabbedInterface([iface, dummy_iface], ["Chatbot", "Dummy"])
    demo.launch(share=True)