forked from frdel/agent-zero
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathagent.py
516 lines (436 loc) · 19.8 KB
/
agent.py
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
import asyncio
from dataclasses import dataclass, field
import time, importlib, inspect, os, json
from typing import Any, Optional, Dict
import uuid
from python.helpers import extract_tools, rate_limiter, files, errors
from python.helpers.print_style import PrintStyle
from langchain.schema import AIMessage
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.messages import HumanMessage, SystemMessage
from langchain_core.language_models.chat_models import BaseChatModel
from langchain_core.language_models.llms import BaseLLM
from langchain_core.embeddings import Embeddings
import python.helpers.log as Log
from python.helpers.dirty_json import DirtyJson
from python.helpers.defer import DeferredTask
class AgentContext:
_contexts: dict[str, "AgentContext"] = {}
_counter: int = 0
def __init__(
self, config: "AgentConfig", id: str | None = None, agent0: "Agent|None" = None
):
# build context
self.id = id or str(uuid.uuid4())
self.config = config
self.log = Log.Log()
self.agent0 = agent0 or Agent(0, self.config, self)
self.paused = False
self.streaming_agent: Agent | None = None
self.process: DeferredTask | None = None
AgentContext._counter += 1
self.no = AgentContext._counter
self._contexts[self.id] = self
@staticmethod
def get(id: str):
return AgentContext._contexts.get(id, None)
@staticmethod
def first():
if not AgentContext._contexts:
return None
return list(AgentContext._contexts.values())[0]
@staticmethod
def remove(id: str):
context = AgentContext._contexts.pop(id, None)
if context and context.process:
context.process.kill()
return context
def reset(self):
if self.process:
self.process.kill()
self.log.reset()
self.agent0 = Agent(0, self.config, self)
self.streaming_agent = None
self.paused = False
def communicate(self, msg: str, broadcast_level: int = 1):
self.paused = False # unpause if paused
if self.process and self.process.is_alive():
if self.streaming_agent:
current_agent = self.streaming_agent
else:
current_agent = self.agent0
# set intervention messages to agent(s):
intervention_agent = current_agent
while intervention_agent and broadcast_level != 0:
intervention_agent.intervention_message = msg
broadcast_level -= 1
intervention_agent = intervention_agent.data.get("superior", None)
else:
self.process = DeferredTask(self.agent0.message_loop, msg)
return self.process
@dataclass
class AgentConfig:
chat_model: BaseChatModel | BaseLLM
utility_model: BaseChatModel | BaseLLM
embeddings_model: Embeddings
prompts_subdir: str = ""
memory_subdir: str = ""
knowledge_subdir: str = ""
auto_memory_count: int = 3
auto_memory_skip: int = 2
rate_limit_seconds: int = 60
rate_limit_requests: int = 15
rate_limit_input_tokens: int = 0
rate_limit_output_tokens: int = 0
msgs_keep_max: int = 25
msgs_keep_start: int = 5
msgs_keep_end: int = 10
response_timeout_seconds: int = 60
max_tool_response_length: int = 3000
code_exec_docker_enabled: bool = True
code_exec_docker_name: str = "agent-zero-exe"
code_exec_docker_image: str = "frdel/agent-zero-exe:latest"
code_exec_docker_ports: dict[str, int] = field(
default_factory=lambda: {"22/tcp": 50022}
)
code_exec_docker_volumes: dict[str, dict[str, str]] = field(
default_factory=lambda: {
files.get_abs_path("work_dir"): {"bind": "/root", "mode": "rw"}
}
)
code_exec_ssh_enabled: bool = True
code_exec_ssh_addr: str = "localhost"
code_exec_ssh_port: int = 50022
code_exec_ssh_user: str = "root"
code_exec_ssh_pass: str = "toor"
additional: Dict[str, Any] = field(default_factory=dict)
# intervention exception class - skips rest of message loop iteration
class InterventionException(Exception):
pass
# repairable exception class - forwarded to LLM, may be fixed on its own
class RepairableException(Exception):
pass
class Agent:
def __init__(
self, number: int, config: AgentConfig, context: AgentContext | None = None
):
# agent config
self.config = config
# agent context
self.context = context or AgentContext(config)
# non-config vars
self.number = number
self.agent_name = f"Agent {self.number}"
self.history = []
self.last_message = ""
self.intervention_message = ""
self.rate_limiter = rate_limiter.RateLimiter(
self.context.log,
max_calls=self.config.rate_limit_requests,
max_input_tokens=self.config.rate_limit_input_tokens,
max_output_tokens=self.config.rate_limit_output_tokens,
window_seconds=self.config.rate_limit_seconds,
)
self.data = {} # free data object all the tools can use
async def message_loop(self, msg: str):
try:
printer = PrintStyle(italic=True, font_color="#b3ffd9", padding=False)
user_message = self.read_prompt("fw.user_message.md", message=msg)
await self.append_message(
user_message, human=True
) # Append the user's input to the history
memories = await self.fetch_memories(True)
while (
True
): # let the agent iterate on his thoughts until he stops by using a tool
self.context.streaming_agent = self # mark self as current streamer
agent_response = ""
try:
system = (
self.read_prompt("agent.system.md", agent_name=self.agent_name)
+ "\n\n"
+ self.read_prompt("agent.tools.md")
)
memories = await self.fetch_memories()
if memories:
system += "\n\n" + memories
prompt = ChatPromptTemplate.from_messages(
[
SystemMessage(content=system),
MessagesPlaceholder(variable_name="messages"),
]
)
inputs = {"messages": self.history}
chain = prompt | self.config.chat_model
formatted_inputs = prompt.format(messages=self.history)
tokens = int(len(formatted_inputs) / 4)
self.rate_limiter.limit_call_and_input(tokens)
# output that the agent is starting
PrintStyle(
bold=True,
font_color="green",
padding=True,
background_color="white",
).print(f"{self.agent_name}: Generating:")
log = self.context.log.log(
type="agent", heading=f"{self.agent_name}: Generating:"
)
async for chunk in chain.astream(inputs):
await self.handle_intervention(
agent_response
) # wait for intervention and handle it, if paused
if isinstance(chunk, str):
content = chunk
elif hasattr(chunk, "content"):
content = str(chunk.content)
else:
content = str(chunk)
if content:
printer.stream(content) # output the agent response stream
agent_response += (
content # concatenate stream into the response
)
self.log_from_stream(agent_response, log)
self.rate_limiter.set_output_tokens(
int(len(agent_response) / 4)
) # rough estimation
await self.handle_intervention(agent_response)
if (
self.last_message == agent_response
): # if assistant_response is the same as last message in history, let him know
await self.append_message(
agent_response
) # Append the assistant's response to the history
warning_msg = self.read_prompt("fw.msg_repeat.md")
await self.append_message(
warning_msg, human=True
) # Append warning message to the history
PrintStyle(font_color="orange", padding=True).print(warning_msg)
self.context.log.log(type="warning", content=warning_msg)
else: # otherwise proceed with tool
await self.append_message(
agent_response
) # Append the assistant's response to the history
tools_result = await self.process_tools(
agent_response
) # process tools requested in agent message
if tools_result: # final response of message loop available
return (
tools_result # break the execution if the task is done
)
except InterventionException as e:
pass # intervention message has been handled in handle_intervention(), proceed with conversation loop
except asyncio.CancelledError as e:
PrintStyle(
font_color="white", background_color="red", padding=True
).print(f"Context {self.context.id} terminated during message loop")
raise e # process cancelled from outside, kill the loop
except RepairableException as e: # Forward repairable errors to the LLM, maybe it can fix them
error_message = errors.format_error(e)
msg_response = self.read_prompt(
"fw.error.md", error=error_message
) # error message template
await self.append_message(msg_response, human=True)
PrintStyle(font_color="red", padding=True).print(msg_response)
self.context.log.log(type="error", content=msg_response)
except Exception as e: # Other exception kill the loop
error_message = errors.format_error(e)
PrintStyle(font_color="red", padding=True).print(error_message)
self.context.log.log(type="error", content=error_message)
raise e # kill the loop
finally:
self.context.streaming_agent = None # unset current streamer
def read_prompt(self, file: str, **kwargs):
content = ""
if self.config.prompts_subdir:
try:
content = files.read_file(
files.get_abs_path(
f"./prompts/{self.config.prompts_subdir}/{file}"
),
**kwargs,
)
except Exception as e:
pass
if not content:
content = files.read_file(
files.get_abs_path(f"./prompts/default/{file}"), **kwargs
)
return content
def get_data(self, field: str):
return self.data.get(field, None)
def set_data(self, field: str, value):
self.data[field] = value
async def append_message(self, msg: str, human: bool = False):
message_type = "human" if human else "ai"
if self.history and self.history[-1].type == message_type:
self.history[-1].content += "\n\n" + msg
else:
new_message = HumanMessage(content=msg) if human else AIMessage(content=msg)
self.history.append(new_message)
await self.cleanup_history(
self.config.msgs_keep_max,
self.config.msgs_keep_start,
self.config.msgs_keep_end,
)
if message_type == "ai":
self.last_message = msg
def concat_messages(self, messages):
return "\n".join([f"{msg.type}: {msg.content}" for msg in messages])
async def send_adhoc_message(self, system: str, msg: str, output_label: str):
prompt = ChatPromptTemplate.from_messages(
[SystemMessage(content=system), HumanMessage(content=msg)]
)
chain = prompt | self.config.utility_model
response = ""
printer = None
logger = None
if output_label:
PrintStyle(
bold=True, font_color="orange", padding=True, background_color="white"
).print(f"{self.agent_name}: {output_label}:")
printer = PrintStyle(italic=True, font_color="orange", padding=False)
logger = self.context.log.log(
type="adhoc", heading=f"{self.agent_name}: {output_label}:"
)
formatted_inputs = prompt.format()
tokens = int(len(formatted_inputs) / 4)
self.rate_limiter.limit_call_and_input(tokens)
async for chunk in chain.astream({}):
if self.handle_intervention():
break # wait for intervention and handle it, if paused
if isinstance(chunk, str):
content = chunk
elif hasattr(chunk, "content"):
content = str(chunk.content)
else:
content = str(chunk)
if printer:
printer.stream(content)
response += content
if logger:
logger.update(content=response)
self.rate_limiter.set_output_tokens(int(len(response) / 4))
return response
def get_last_message(self):
if self.history:
return self.history[-1]
async def replace_middle_messages(self, middle_messages):
cleanup_prompt = self.read_prompt("fw.msg_cleanup.md")
summary = await self.send_adhoc_message(
system=cleanup_prompt,
msg=self.concat_messages(middle_messages),
output_label="Mid messages cleanup summary",
)
new_human_message = HumanMessage(content=summary)
return [new_human_message]
async def cleanup_history(self, max: int, keep_start: int, keep_end: int):
if len(self.history) <= max:
return self.history
first_x = self.history[:keep_start]
last_y = self.history[-keep_end:]
# Identify the middle part
middle_part = self.history[keep_start:-keep_end]
# Ensure the first message in the middle is "human", if not, move one message back
if middle_part and middle_part[0].type != "human":
if len(first_x) > 0:
middle_part.insert(0, first_x.pop())
# Ensure the middle part has an odd number of messages
if len(middle_part) % 2 == 0:
middle_part = middle_part[:-1]
# Replace the middle part using the replacement function
new_middle_part = await self.replace_middle_messages(middle_part)
self.history = first_x + new_middle_part + last_y
return self.history
async def handle_intervention(self, progress: str = ""):
while self.context.paused:
await asyncio.sleep(0.1) # wait if paused
if (
self.intervention_message
): # if there is an intervention message, but not yet processed
msg = self.intervention_message
self.intervention_message = "" # reset the intervention message
if progress.strip():
await self.append_message(
progress
) # append the response generated so far
user_msg = self.read_prompt(
"fw.intervention.md", user_message=msg
) # format the user intervention template
await self.append_message(
user_msg, human=True
) # append the intervention message
raise InterventionException(msg)
async def process_tools(self, msg: str):
# search for tool usage requests in agent message
tool_request = extract_tools.json_parse_dirty(msg)
if tool_request is not None:
tool_name = tool_request.get("tool_name", "")
tool_args = tool_request.get("tool_args", {})
tool = self.get_tool(tool_name, tool_args, msg)
await self.handle_intervention() # wait if paused and handle intervention message if needed
await tool.before_execution(**tool_args)
await self.handle_intervention() # wait if paused and handle intervention message if needed
response = await tool.execute(**tool_args)
await self.handle_intervention() # wait if paused and handle intervention message if needed
await tool.after_execution(response)
await self.handle_intervention() # wait if paused and handle intervention message if needed
if response.break_loop:
return response.message
else:
msg = self.read_prompt("fw.msg_misformat.md")
await self.append_message(msg, human=True)
PrintStyle(font_color="red", padding=True).print(msg)
self.context.log.log(
type="error", content=f"{self.agent_name}: Message misformat:"
)
def get_tool(self, name: str, args: dict, message: str, **kwargs):
from python.tools.unknown import Unknown
from python.helpers.tool import Tool
tool_class = Unknown
if files.exists("python/tools", f"{name}.py"):
module = importlib.import_module(
"python.tools." + name
) # Import the module
class_list = inspect.getmembers(
module, inspect.isclass
) # Get all functions in the module
for cls in class_list:
if cls[1] is not Tool and issubclass(cls[1], Tool):
tool_class = cls[1]
break
return tool_class(agent=self, name=name, args=args, message=message, **kwargs)
async def fetch_memories(self, reset_skip=False):
if self.config.auto_memory_count <= 0:
return ""
if reset_skip:
self.memory_skip_counter = 0
if self.memory_skip_counter > 0:
self.memory_skip_counter -= 1
return ""
else:
self.memory_skip_counter = self.config.auto_memory_skip
from python.tools import memory_tool
messages = self.concat_messages(self.history)
memories = memory_tool.search(self, messages)
input = {"conversation_history": messages, "raw_memories": memories}
cleanup_prompt = self.read_prompt("msg.memory_cleanup.md").replace(
"{", "{{"
)
clean_memories = await self.send_adhoc_message(
cleanup_prompt, json.dumps(input), output_label="Memory injection"
)
return clean_memories
def log_from_stream(self, stream: str, logItem: Log.LogItem):
try:
if len(stream) < 25:
return # no reason to try
response = DirtyJson.parse_string(stream)
if isinstance(response, dict):
logItem.update(
content=stream, kvps=response
) # log if result is a dictionary already
except Exception as e:
pass
def call_extension(self, name: str, **kwargs) -> Any:
pass