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
|
#!/bin/python3
import os
from pathlib import Path
from openai import OpenAI
from typing import Optional
import argparse
import subprocess
import readline
import json
import logging
logger = logging.getLogger(__name__)
logging.basicConfig(filename="thoughts.log",
level=logging.INFO)
client = OpenAI(api_key = os.environ["AALTO_LLM_KEY"],
base_url = "https://llm-gateway.k8s.aalto.fi/api/v1")
def direct_query(query: str):
logger.info(f"send: {query}")
res = client.responses.create(model = "RedHatAI/gemma-4-31B-it-FP8-Dynamic",
input = query)
logger.info(f"receive: {res.output_text}")
return res.output_text
# def reason_query(input):
# # TODO some models may natively support reasoning
# reasoning = direct_query(input + ({"role": "system",
# "content": "Reason about the user's query step by step.\n\nAn answer will be generated based on your analysis."},))
# summary = direct_query(input + ({"role": "assistant", "content": reasoning},
# {"role": "system",
# "content": "Formulate your previous reasoning as a concise answer to the user."}))
# return summary
# # kinda funky
# # def flexible_query(input):
# # reasoning_required = direct_query(input[-1:] + ({"role": "system", "content": "Does answering the user's question require reasoning? (y|N)"},))
# # if reasoning_required == 'y':
# # return reason_query(input)
# # else:
# # return direct_query(input)
def chat(content, state = (), query_method = direct_query, role = "user"):
state = state + ({"role": role, "content": content},)
response = query_method(state)
state = state + ({"role": "assistant", "content": response},)
return (response, state)
def strip_enclosing(s: str, start="```json", end="```") -> str:
s = s.strip()
return s[len(start):-len(end)] if s[:len(start)] == start and s[-len(end):] == end else s
def path_checked(path: Path, cwd = Path.cwd()) -> Optional[Path]:
"""Resolves `path`, returning `None` if it's not a subdirectory of `cwd`."""
path = path.resolve()
cwd = cwd.resolve()
return path if path.is_relative_to(cwd) else None
def inner_query_agent(actions: list, query: str, state = ()):
state = state + ({"role": "user", "content": query},)
(res, state) = chat(f"Based on the current state, what should you do next?\n"
"Respond JSON indicating an action, or respond in natural language if you are finished.",
state)
try:
action = json.loads(strip_enclosing(res))
except json.JSONDecodeError:
return (res, state)
for f in actions:
result_json = f(action)
if result_json is not None:
(res, state) = inner_query_agent(actions, json.dumps(result_json), state)
return (res, state)
logger.info(f"valid json '{res}' but no actions matched")
return (res, state)
def query_agent(query: str, state = ()):
def path_helper(path_str: str, callback):
path = path_checked(Path(path_str))
if path is None:
return {"status": "error", "reason": f"Permission denied: {path_str}"}
else:
try:
return {"status": "success", **callback(path)}
except Exception as e:
return {"status": "error", "reason": f"Failure: {e}"}
def ls(args):
'{"action": "ls", "path": PATH} - list files at PATH'
def inner(path: Path):
return {"content": list(map(lambda s: str(s), path.iterdir()))}
match args:
case {"action": "ls", "path": path_str}:
return path_helper(path_str, inner)
def read(args):
'{"action": "read", "path": PATH} - read file at PATH'
def inner(path: Path):
contents = path.read_text(encoding="utf-8")
return {"content": contents}
match args:
case {"action": "read", "path": path_str}:
return path_helper(path_str, inner)
def write(args):
'{"action": "write", "path": PATH, "content": CONTENT} - write CONTENT into file at PATH'
def inner(path: Path, text: str):
n_bytes = path.write_text(text)
return {"bytes": n_bytes}
match args:
case {"action": "write", "path": path_str, "content": content}:
return path_helper(path_str, lambda path: inner(path, content))
def tldr(args):
'{"action": "tldr", "program": PROGRAM} - read tldr docs for PROGRAM'
match args:
case {"action": "tldr", "program": program_str}:
try:
# Run man command and capture output
# -P cat is used to avoid paging and just print the text
result = subprocess.run(
["tldr", program_str],
capture_output=True,
text=True,
check=True
)
return {"status": "success", "content": result.stdout}
except subprocess.CalledProcessError:
return {"status": "error", "reason": f"Tldr for '{program_str}' not found."}
except Exception as e:
return {"status": "error", "reason": f"An unexpected error occurred: {e}"}
actions = [ls, read, write, tldr]
state = state + ({"role": "system",
"content": f"""You are an autonomous coding agent that can help with programming tasks.
You can explore repositories, read files, search for information, and make changes to code.
You have access to the following actions.
{'\n'.join(map(lambda f: '* ' + f.__doc__, actions))}
* RESPONSE - send non-json text RESPONSE to the user once the task is complete
"""},)
state = state + ({"role": "system", "content": "Analyze the following user-given task and break it down into steps."},)
return inner_query_agent(actions, query, state)
def main():
state = ()
while True:
query = input("> ").strip()
(response, state) = query_agent(query, state)
print(response)
if __name__ == "__main__":
main()
|
