File size: 15,990 Bytes
02d640a
 
 
b07e47b
02d640a
b07e47b
02d640a
b07e47b
69af1c5
b07e47b
ffea40d
 
b07e47b
ffea40d
69af1c5
ffea40d
69af1c5
ffea40d
 
 
69af1c5
 
ffea40d
 
 
 
 
 
69af1c5
ffea40d
 
 
 
 
 
 
 
 
 
 
69af1c5
b07e47b
69af1c5
 
 
ffea40d
69af1c5
 
ffea40d
 
 
69af1c5
ffea40d
69af1c5
 
ffea40d
69af1c5
ffea40d
69af1c5
ffea40d
69af1c5
 
 
 
 
 
 
 
 
 
 
ffea40d
69af1c5
 
 
 
 
 
 
 
ffea40d
69af1c5
 
 
 
 
 
 
 
 
 
 
 
 
ffea40d
 
 
69af1c5
 
ffea40d
69af1c5
 
 
ffea40d
69af1c5
ffea40d
 
 
 
69af1c5
ffea40d
69af1c5
 
 
 
 
b07e47b
69af1c5
ffea40d
 
69af1c5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ffea40d
 
69af1c5
ffea40d
69af1c5
 
 
b07e47b
69af1c5
ffea40d
 
 
69af1c5
ffea40d
69af1c5
ffea40d
69af1c5
ffea40d
69af1c5
 
ffea40d
 
 
 
 
69af1c5
 
 
ffea40d
 
 
 
69af1c5
ffea40d
69af1c5
 
ffea40d
 
 
 
69af1c5
ffea40d
69af1c5
ffea40d
 
 
69af1c5
ffea40d
 
69af1c5
b07e47b
ffea40d
 
69af1c5
b07e47b
ffea40d
 
69af1c5
 
 
 
 
 
 
ffea40d
69af1c5
 
 
 
 
 
 
 
ffea40d
69af1c5
 
ffea40d
69af1c5
ffea40d
69af1c5
ffea40d
 
 
 
69af1c5
ffea40d
 
69af1c5
 
ffea40d
69af1c5
ffea40d
 
 
 
69af1c5
ffea40d
 
 
 
 
 
 
b07e47b
 
 
69af1c5
ffea40d
 
 
 
b07e47b
ffea40d
 
69af1c5
ffea40d
 
 
 
69af1c5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ffea40d
 
69af1c5
ffea40d
 
69af1c5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b07e47b
ffea40d
69af1c5
 
ffea40d
 
 
69af1c5
ffea40d
69af1c5
 
 
 
 
ffea40d
 
b07e47b
69af1c5
 
 
b07e47b
69af1c5
 
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
import os
import sys
import asyncio
import logging
import threading
import queue
import gradio as gr
import httpx
from typing import Generator, Any, Dict, List, Optional

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

# -------------------- External Model Call --------------------
async def call_model(prompt: str, model: str = "gpt-4o", api_key: str = None) -> str:
    """
    Sends a prompt to the OpenAI API endpoint.
    """
    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}],
    }
    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()
        return response_json["choices"][0]["message"]["content"]

# -------------------- Agent Classes --------------------
class PromptOptimizerAgent:
    async def optimize_prompt(self, user_prompt: str, api_key: str) -> str:
        """Optimizes the user's initial prompt."""
        system_prompt = (
            "You are a prompt optimization expert. Improve the given user prompt. "
            "Be clear, specific, and complete. Maintain the user's original intent."
            "Return ONLY the revised prompt."
        )
        full_prompt = f"{system_prompt}\n\nUser's initial prompt:\n{user_prompt}"
        optimized = await call_model(full_prompt, model="gpt-4o", api_key=api_key)
        return optimized

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

    async def generate_plan(self, task: str, api_key: str, human_feedback: Optional[str] = None) -> str:
        """
        Generates a plan, potentially requesting human feedback.
        """
        if human_feedback:  # Use human feedback if provided
            prompt = (
                f"You are a master planner. You previously generated a partial plan for the task: '{task}'.\n"
                "You requested human feedback, and here's the feedback you received:\n"
                f"{human_feedback}\n\n"
                "Now, complete or revise the plan, incorporating the human feedback. "
                "Output the plan as a numbered list."
            )
            plan = await call_model(prompt, model="gpt-4o", api_key=api_key)
            return plan

        prompt = (
            f"You are a master planner. Given the task: '{task}', create a detailed, step-by-step plan. "
            "Break down the task into sub-tasks. Assign each sub-task to agents: Coder, Code Reviewer, Quality Assurance Tester, and Documentation Agent. "
            "Include steps for review and revision. Consider potential issues and error handling. "
            "Include instructions for documentation.\n\n"
            "HOWEVER, if at ANY point you are unsure how to proceed, you can request human feedback.  "
            "To do this, output ONLY the following phrase (and nothing else): 'REQUEST_HUMAN_FEEDBACK'\n"
            "Followed by a newline and a clear and concise question for the human. Example:\n\nREQUEST_HUMAN_FEEDBACK\nShould the output be in JSON or XML format?"
            "\n\nOutput the plan as a numbered list (or as much as you can before requesting feedback)."
        )
        plan = await call_model(prompt, model="gpt-4o", api_key=api_key)

        if "REQUEST_HUMAN_FEEDBACK" in plan:
            self.log_queue.put("[Orchestrator]: Requesting human feedback...")
            question = plan.split("REQUEST_HUMAN_FEEDBACK\n", 1)[1].strip()
            self.log_queue.put(f"[Orchestrator]: Question for human: {question}")
            self.human_in_the_loop_event.set()  # Signal the human input thread
            human_response = self.human_input_queue.get()  # Wait for human input
            self.human_in_the_loop_event.clear()  # Reset the event
            self.log_queue.put(f"[Orchestrator]: Received human feedback: {human_response}")
            return await self.generate_plan(task, api_key, human_response) # Recursive call with feedback


        return plan

class CoderAgent:
    async def generate_code(self, instructions: str, api_key: str, model: str = "gpt-4o") -> str:
        """Generates code based on instructions."""
        prompt = (
            "You are a highly skilled coding agent. Output ONLY the code. "
            "Adhere to best practices. Include error handling.\n\n"
            f"Instructions:\n{instructions}"
        )
        code = await call_model(prompt, model=model, api_key=api_key)
        return code

class CodeReviewerAgent:
    async def review_code(self, code: str, task: str, api_key: str) -> str:
        """Reviews code. Provides concise, actionable feedback or 'APPROVE'."""
        prompt = (
            "You are a meticulous code reviewer. Provide CONCISE feedback. "
            "Focus on correctness, efficiency, readability, error handling, security, and adherence to the task. "
            "Suggest improvements. If acceptable, respond with ONLY 'APPROVE'. "
            "Do NOT generate code.\n\n"
            f"Task: {task}\n\nCode:\n{code}"
        )
        review = await call_model(prompt, model="gpt-4o", api_key=api_key)
        return review

class QualityAssuranceTesterAgent:
    async def generate_test_cases(self, code: str, task: str, api_key: str) -> str:
        """Generates test cases."""
        prompt = (
            "You are a quality assurance testing agent. Generate test cases. "
            "Consider edge cases and error scenarios. Output in a clear format.\n\n"
            f"Task: {task}\n\nCode:\n{code}"
        )
        test_cases = await call_model(prompt, model="gpt-4o", api_key=api_key)
        return test_cases

    async def run_tests(self, code:str, test_cases:str, api_key:str) -> str:
        """Runs tests and reports results."""
        prompt = (
            "Run the generated test cases. Compare actual vs expected output. "
            "State discrepancies. If all pass, output 'TESTS PASSED'.\n\n"
            f"Code:\n{code}\n\nTest Cases:\n{test_cases}"
        )
        test_results = await call_model(prompt, model="gpt-4o", api_key=api_key)
        return test_results

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

# -------------------- Multi-Agent Conversation --------------------
async def multi_agent_conversation(task_message: str, log_queue: queue.Queue, api_key: str, human_in_the_loop_event: threading.Event, human_input_queue: queue.Queue) -> None:
    """
    Conducts the multi-agent conversation.
    """
    conversation: List[Dict[str, str]] = []

    # Step 0: Optimize Prompt
    log_queue.put("[Prompt Optimizer]: Optimizing prompt...")
    prompt_optimizer = PromptOptimizerAgent()
    optimized_task = await prompt_optimizer.optimize_prompt(task_message, api_key=api_key)
    conversation.append({"agent": "Prompt Optimizer", "message": f"Optimized Task:\n{optimized_task}"})
    log_queue.put(f"[Prompt Optimizer]: Optimized task prompt:\n{optimized_task}")

    # Step 1: Generate Plan
    log_queue.put("[Orchestrator]: Generating plan...")
    orchestrator = OrchestratorAgent(log_queue, human_in_the_loop_event, human_input_queue)
    plan = await orchestrator.generate_plan(optimized_task, api_key=api_key)
    conversation.append({"agent": "Orchestrator", "message": f"Plan:\n{plan}"})
    log_queue.put(f"[Orchestrator]: Plan generated:\n{plan}")

    # Step 2: Generate Code
    coder = CoderAgent()
    coder_instructions = f"Implement the task:\n{plan}"
    log_queue.put("[Coder]: Generating code...")
    code = await coder.generate_code(coder_instructions, api_key=api_key)
    conversation.append({"agent": "Coder", "message": f"Code:\n{code}"})
    log_queue.put(f"[Coder]: Code generated:\n{code}")

    # Step 3: Code Review and Revision
    reviewer = CodeReviewerAgent()
    tester = QualityAssuranceTesterAgent()
    approval_keyword = "approve"
    revision_iteration = 0
    while True:
        log_queue.put(f"[Code Reviewer]: Reviewing code (Iteration {revision_iteration})...")
        review = await reviewer.review_code(code, optimized_task, api_key=api_key)
        conversation.append({"agent": "Code Reviewer", "message": f"Review (Iteration {revision_iteration}):\n{review}"})
        log_queue.put(f"[Code Reviewer]: Review (Iteration {revision_iteration}):\n{review}")

        if approval_keyword in review.lower():
            log_queue.put("[Code Reviewer]: Code approved.")
            break

        revision_iteration += 1
        if revision_iteration >= 5:
            log_queue.put("Unable to solve task satisfactorily.")
            sys.exit("Unable to solve task satisfactorily.")

        log_queue.put("[QA Tester]: Generating test cases...")
        test_cases = await tester.generate_test_cases(code, optimized_task, api_key=api_key)
        conversation.append({"agent": "QA Tester", "message": f"Test Cases:\n{test_cases}"})
        log_queue.put(f"[QA Tester]: Test Cases:\n{test_cases}")

        log_queue.put("[QA Tester]: Running tests...")
        test_results = await tester.run_tests(code, test_cases, api_key)
        conversation.append({"agent": "QA Tester", "message": f"Test Results:\n{test_results}"})
        log_queue.put(f"[QA Tester]: Test Results:\n{test_results}")

        log_queue.put(f"[Orchestrator]: Revising code (Iteration {revision_iteration})...")
        update_instructions = f"Revise:\nReview:\n{review}\nTests:\n{test_results}\nPlan:\n{plan}"
        revised_code = await coder.generate_code(update_instructions, api_key=api_key, model="gpt-3.5-turbo-16k")
        conversation.append({"agent": "Coder", "message": f"Revised Code (Iteration {revision_iteration}):\n{revised_code}"})
        log_queue.put(f"[Coder]: Revised (Iteration {revision_iteration}):\n{revised_code}")
        code = revised_code

    # Step 4: Generate Documentation
    doc_agent = DocumentationAgent()
    log_queue.put("[Documentation Agent]: Generating documentation...")
    documentation = await doc_agent.generate_documentation(code, api_key=api_key)
    conversation.append({"agent": "Documentation Agent", "message": f"Documentation:\n{documentation}"})
    log_queue.put(f"[Documentation Agent]: Documentation generated:\n{documentation}")

    log_queue.put("Conversation complete.")
    log_queue.put(("result", conversation))

# -------------------- Process Generator and Human Input --------------------
def process_conversation_generator(task_message: str, api_key: str, human_in_the_loop_event: threading.Event, human_input_queue: queue.Queue) -> Generator[str, None, None]:
    """
    Wraps the conversation and yields log messages. Handles human input.
    """
    log_q: queue.Queue = queue.Queue()

    def run_conversation() -> None:
        asyncio.run(multi_agent_conversation(task_message, log_q, api_key, human_in_the_loop_event, human_input_queue))

    thread = threading.Thread(target=run_conversation)
    thread.start()

    final_result = None
    while thread.is_alive() or not log_q.empty():
        try:
            msg = log_q.get(timeout=0.1)
            if isinstance(msg, tuple) and msg[0] == "result":
                final_result = msg[1]
                yield "Conversation complete."
            else:
                yield msg
        except queue.Empty:
            continue

    thread.join()
    if final_result:
        conv_text = "\n=== Conversation ===\n"
        for entry in final_result:
            conv_text += f"[{entry['agent']}]: {entry['message']}\n\n"
        yield conv_text

def get_human_feedback(placeholder_text):
    """Gets human input using a Gradio Textbox."""
    with gr.Blocks() as human_feedback_interface:
        with gr.Row():
            human_input = gr.Textbox(lines=4, placeholder=placeholder_text, label="Human Feedback")
        with gr.Row():
            submit_button = gr.Button("Submit Feedback")

        feedback_queue = queue.Queue()

        def submit_feedback(input_text):
            feedback_queue.put(input_text)
            return ""

        submit_button.click(submit_feedback, inputs=human_input, outputs=human_input)
        human_feedback_interface.load(None, [], [])  # This is needed to keep the interface alive

    return human_feedback_interface, feedback_queue
# -------------------- Chat Function for Gradio --------------------
def multi_agent_chat(message: str, history: List[Any], openai_api_key: str = None) -> Generator[str, None, None]:
    """Chat function for Gradio."""
    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_in_the_loop_event = threading.Event()
    human_input_queue = queue.Queue()

    yield from process_conversation_generator(message, openai_api_key, human_in_the_loop_event, human_input_queue)

    while human_in_the_loop_event.is_set():
        yield "Waiting for human feedback..."
        placeholder = "Please provide your feedback."
        human_interface, feedback_queue = get_human_feedback(placeholder)
        #This is a hacky but currently only working way to make this work with gradio
        yield gr.Textbox.update(visible=False), gr.update(visible=True)
        try:
            human_feedback = feedback_queue.get(timeout=300)  # Wait for up to 5 minutes
            human_input_queue.put(human_feedback)
            human_in_the_loop_event.clear()
            yield gr.Textbox.update(visible=True), human_interface.close()
            yield from process_conversation_generator(message, openai_api_key, human_in_the_loop_event, human_input_queue)

        except queue.Empty:
            human_input_queue.put("No feedback provided.") #Timeout
            human_in_the_loop_event.clear()
            yield gr.Textbox.update(visible=True), human_interface.close()
            yield from process_conversation_generator(message, openai_api_key, human_in_the_loop_event, human_input_queue)

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

# Create the main chat interface
iface = gr.ChatInterface(
    fn=multi_agent_chat,
    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 capability.
        - The Orchestrator can ask for human feedback if needed.
        - Enter a task, and the agents will work on it.  You may be prompted for input.
        - Max 5 revision iterations.
        - Provide your OpenAI API Key below.
        """
)

#Need a dummy interface to make the human feedback interface update
dummy_iface = gr.Interface(lambda x:x, "textbox", "textbox")

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