Spaces:
Runtime error
Runtime error
import asyncio | |
import base64 | |
import json | |
import shlex | |
import gradio as gr | |
from gradio.oauth import attach_oauth, OAuthToken | |
from huggingface_hub import HfApi | |
from src.team import TeamChatSession | |
from src.db import list_sessions_info | |
# Store active chat sessions | |
_SESSIONS: dict[tuple[str, str], TeamChatSession] = {} | |
_API = HfApi() | |
def _username(token: OAuthToken) -> str: | |
"""Return the username for the given token.""" | |
info = _API.whoami(token.token) | |
return info.get("name") or info.get("user", "unknown") | |
async def _get_chat(user: str, session: str) -> TeamChatSession: | |
"""Return an active :class:`TeamChatSession` for ``user`` and ``session``.""" | |
key = (user, session) | |
chat = _SESSIONS.get(key) | |
if chat is None: | |
chat = TeamChatSession(user=user, session=session) | |
await chat.__aenter__() | |
_SESSIONS[key] = chat | |
return chat | |
async def _vm_execute(user: str, session: str, command: str) -> str: | |
"""Execute ``command`` inside the user's VM and return output.""" | |
chat = await _get_chat(user, session) | |
vm = getattr(chat.senior, "_vm", None) | |
if vm is None: | |
raise RuntimeError("VM not running") | |
return await vm.execute_async(command, timeout=5) | |
async def send_message( | |
message: str, history: list[dict] | None, session: str, token: OAuthToken | |
): | |
user = _username(token) | |
chat = await _get_chat(user, session) | |
history = history or [] | |
# user turn | |
history.append({"role": "user", "content": message}) | |
yield history # show immediately | |
# stream assistant turns | |
async for part in chat.chat_stream(message): | |
if history[-1]["role"] == "assistant": | |
history[-1]["content"] += part | |
else: | |
history.append({"role": "assistant", "content": part}) | |
yield history | |
def load_sessions(token: OAuthToken): | |
user = _username(token) | |
infos = list_sessions_info(user) | |
names = [info["name"] for info in infos] | |
table = [[info["name"], info["last_message"]] for info in infos] | |
value = names[0] if names else "default" | |
return gr.update(choices=names or ["default"], value=value), table | |
async def list_dir(path: str, session: str, token: OAuthToken): | |
user = _username(token) | |
cmd = f"ls -1ap {shlex.quote(path)}" | |
output = await _vm_execute(user, session, cmd) | |
if output.startswith("ls:"): | |
return [] | |
rows = [] | |
for line in output.splitlines(): | |
line = line.strip() | |
if not line or line in (".", ".."): | |
continue | |
is_dir = line.endswith("/") | |
name = line[:-1] if is_dir else line | |
rows.append([name, is_dir]) | |
return rows | |
async def read_file(path: str, session: str, token: OAuthToken): | |
user = _username(token) | |
cmd = f"cat {shlex.quote(path)}" | |
return await _vm_execute(user, session, cmd) | |
async def save_file(path: str, content: str, session: str, token: OAuthToken): | |
user = _username(token) | |
encoded = base64.b64encode(content.encode()).decode() | |
cmd = ( | |
f"python -c 'import base64,os; " | |
f'open({json.dumps(path)}, "wb").write(base64.b64decode({json.dumps(encoded)}))\'' | |
) | |
await _vm_execute(user, session, cmd) | |
return "Saved" | |
async def delete_path(path: str, session: str, token: OAuthToken): | |
user = _username(token) | |
cmd = ( | |
f"bash -c 'if [ -d {shlex.quote(path)} ]; then rm -rf {shlex.quote(path)} && echo Deleted; " | |
f"elif [ -e {shlex.quote(path)} ]; then rm -f {shlex.quote(path)} && echo Deleted; " | |
f"else echo File not found; fi'" | |
) | |
return await _vm_execute(user, session, cmd) | |
with gr.Blocks(theme=gr.themes.Soft()) as demo: | |
attach_oauth(demo.app) | |
login_btn = gr.LoginButton() | |
with gr.Tab("Chat"): | |
session_dd = gr.Dropdown(["default"], label="Session", value="default") | |
refresh = gr.Button("Refresh Sessions") | |
chatbox = gr.Chatbot(type="messages") | |
msg = gr.Textbox(label="Message") | |
send = gr.Button("Send") | |
gr.Markdown( | |
""" | |
This is a demo app of [llmOS](https://github.com/starsnatched/llmOS-Agent), an agent framework for building AI assistants that can interact with a Linux VM to accomplish tasks. | |
""" | |
) | |
with gr.Tab("Files"): | |
dir_path = gr.Textbox(label="Directory", value="/") | |
list_btn = gr.Button("List") | |
table = gr.Dataframe(headers=["name", "is_dir"], datatype=["str", "bool"]) | |
file_path = gr.Textbox(label="File Path") | |
load_btn = gr.Button("Load") | |
content = gr.Code(label="Content", language=None) | |
save_btn = gr.Button("Save") | |
del_btn = gr.Button("Delete") | |
refresh.click(load_sessions, outputs=[session_dd, table]) | |
send_click = send.click( | |
send_message, | |
inputs=[msg, chatbox, session_dd], | |
outputs=chatbox, | |
) | |
list_btn.click(list_dir, inputs=[dir_path, session_dd], outputs=table) | |
load_btn.click(read_file, inputs=[file_path, session_dd], outputs=content) | |
save_btn.click(save_file, inputs=[file_path, content, session_dd], outputs=content) | |
del_btn.click(delete_path, inputs=[file_path, session_dd], outputs=content) | |
send_click.then(lambda: "", None, msg) | |
demo.queue() | |
if __name__ == "__main__": | |
demo.launch() | |