File size: 7,406 Bytes
63274c0 | 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 | import asyncio
import json
import os
import time
import threading
import contextvars
from pathlib import Path
from fastapi import FastAPI, Request
from fastapi.responses import HTMLResponse, StreamingResponse
from fastapi.staticfiles import StaticFiles
app = FastAPI()
SYSTEM_PROMPT = Path("system.md").read_text()
MODEL = os.environ.get("SYNC_PLANNER_MODEL", "zai-org/GLM-5.1")
BASE_URL = os.environ.get("SYNC_BASE_URL", "https://router.huggingface.co/v1")
API_KEY = os.environ.get("SYNC_API_KEY", "")
if not API_KEY:
try:
from huggingface_hub import get_token
API_KEY = get_token() or ""
except ImportError:
pass
from openai import AsyncOpenAI
client = AsyncOpenAI(base_url=BASE_URL, api_key=API_KEY)
# --- Minimal sync runtime (inlined) ---
_current_ctx = contextvars.ContextVar("ctx")
class Result:
def __init__(self, messages):
self.messages = messages
@property
def last(self):
return self.messages[-1]["content"]
class _Context:
def __init__(self):
self._out = threading.Event()
self._in = threading.Event()
self._batch = None
self._results = None
self._done = False
self._done_value = None
def _call_many(self, calls):
self._batch = ("calls", calls)
self._out.set()
self._in.wait()
self._in.clear()
return self._results
def _update(self, message):
self._batch = ("update", message)
self._out.set()
self._in.wait()
self._in.clear()
def _finish(self, value):
self._done = True
self._done_value = value
self._out.set()
def agent(name, message, **_kw):
ctx = _current_ctx.get()
if isinstance(message, list):
return ctx._call_many([(name, m) for m in message])
return ctx._call_many([(name, message)])[0]
def update(message):
ctx = _current_ctx.get()
ctx._update(message)
def load_plan(source, task=""):
ns = {"agent": agent, "update": update, "TASK": task}
exec(source, ns)
return ns.get("AGENTS", []), ns.get("plan")
# --- SSE endpoint ---
@app.get("/", response_class=HTMLResponse)
async def index():
return Path("static/index.html").read_text()
@app.get("/generate")
async def generate(task: str):
messages = [
{"role": "system", "content": SYSTEM_PROMPT},
{"role": "user", "content": task},
]
resp = await client.chat.completions.create(
model=MODEL, messages=messages)
code = resp.choices[0].message.content
code = (code.strip()
.removeprefix("```python")
.removeprefix("```")
.removesuffix("```")
.strip())
return {"code": code}
@app.get("/execute")
async def execute(task: str, code: str):
async def stream():
def send(event, data):
return f"event: {event}\ndata: {json.dumps(data)}\n\n"
try:
agent_defs, plan_fn = load_plan(code, task=task)
except Exception as e:
yield send("error", {"msg": f"Plan load error: {e}"})
return
if not plan_fn:
yield send("error", {"msg": "No plan() found"})
return
ctx = _Context()
def run_plan():
_current_ctx.set(ctx)
try:
result = plan_fn()
ctx._finish(result)
except Exception as e:
ctx._finish(f"Error: {e}")
thread = threading.Thread(target=run_plan, daemon=True)
thread.start()
t0 = time.time()
while True:
await asyncio.get_event_loop().run_in_executor(None, ctx._out.wait)
ctx._out.clear()
if ctx._done:
elapsed = time.time() - t0
yield send("log", {"msg": f"done — {elapsed:.1f}s total"})
yield send("result", {"text": str(ctx._done_value or "")})
return
kind, payload = ctx._batch
if kind == "update":
elapsed = time.time() - t0
yield send("log", {"msg": f"[{elapsed:.1f}s] {payload}"})
ctx._in.set()
continue
if kind == "calls":
names = {}
for name, msg in payload:
names[name] = names.get(name, 0) + 1
parts = []
for n, c in names.items():
if c > 1:
parts.append(f"[{c}x] {n}")
else:
parts.append(n)
desc = ", ".join(parts)
elapsed = time.time() - t0
yield send("log", {"msg": f"[{elapsed:.1f}s] ▶ {desc}"})
results = []
tasks = []
for name, msg in payload:
ad = next((a for a in agent_defs if a["name"] == name), {})
tasks.append(_call_agent(ad, msg))
agent_results = await asyncio.gather(*tasks)
for (name, msg), messages in zip(payload, agent_results):
if isinstance(messages, list):
results.append(Result(messages))
else:
results.append(Result([
{"role": "user", "content": msg},
{"role": "assistant", "content": str(messages)},
]))
elapsed2 = time.time() - t0
yield send("log", {"msg": f"[{elapsed2:.1f}s] ✓ {desc}"})
ctx._results = results
ctx._in.set()
return StreamingResponse(
stream(), media_type="text/event-stream",
headers={"Cache-Control": "no-cache", "X-Accel-Buffering": "no"})
async def _call_agent(agent_def, message):
model = agent_def.get("model", MODEL)
system = agent_def.get("system", "")
tools = {fn.__name__: fn for fn in agent_def.get("tools", []) if callable(fn)}
messages = []
if system:
messages.append({"role": "system", "content": system})
messages.append({"role": "user", "content": message})
kwargs = {"model": model, "messages": messages}
if tools:
kwargs["tools"] = list(tools.values())
try:
while True:
resp = await client.chat.completions.create(**kwargs)
choice = resp.choices[0]
if choice.finish_reason == "tool_calls":
messages.append(choice.message.model_dump())
for tc in choice.message.tool_calls:
fn = tools.get(tc.function.name)
if fn:
import json as j
output = fn(**j.loads(tc.function.arguments))
else:
output = f"Unknown tool: {tc.function.name}"
messages.append({
"role": "tool",
"tool_call_id": tc.id,
"content": str(output),
})
continue
messages.append({
"role": "assistant",
"content": choice.message.content,
})
return messages
except Exception as e:
return [
{"role": "user", "content": message},
{"role": "assistant", "content": f"Error: {e}"},
]
|