feat: orchestrate goals with local and cloud models

This commit is contained in:
thanhnv
2026-07-10 23:35:30 +09:00
parent 1345d4c930
commit 48b1186439
11 changed files with 920 additions and 2 deletions
@@ -12,7 +12,9 @@ import json
import os
import shutil
import subprocess
import tempfile
import threading
import time
from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
from typing import Dict, Optional
@@ -33,6 +35,7 @@ PROVIDERS = {
}
RUNNING: Dict[str, subprocess.Popen] = {}
LOCK = threading.Lock()
MODEL_LOCKS = {provider: threading.Lock() for provider in PROVIDERS}
def command_status(provider: str) -> dict:
@@ -109,6 +112,66 @@ def start_login(provider: str) -> dict:
return {"success": True, "reason": "browser_login_started", "provider": command_status(provider)}
def generate_with_account(provider: str, prompt: str) -> dict:
status = command_status(provider)
if not status["available"] or not status["loggedIn"]:
return {"success": False, "reason": "provider_not_logged_in"}
if not prompt or len(prompt) > 24000:
return {"success": False, "reason": "prompt_length_invalid"}
started = time.monotonic()
with MODEL_LOCKS[provider], tempfile.TemporaryDirectory(prefix="casan-account-model-") as directory:
try:
if provider == "codex":
output_path = os.path.join(directory, "last-message.txt")
command = [
"codex", "exec", "--ephemeral", "--ignore-user-config", "--ignore-rules",
"--skip-git-repo-check", "--sandbox", "read-only", "--color", "never",
"--cd", directory, "--output-last-message", output_path, "-",
]
result = subprocess.run(
command, input=prompt, capture_output=True, text=True, timeout=300,
env={**os.environ, "NO_COLOR": "1"},
)
text = ""
if result.returncode == 0 and os.path.isfile(output_path):
with open(output_path, encoding="utf-8") as handle:
text = handle.read().strip()
usage = {}
model = "codex-account-default"
else:
command = [
"claude", "--print", "--output-format", "json", "--permission-mode", "plan",
"--tools", "", "--safe-mode", "--no-session-persistence",
]
result = subprocess.run(
command, input=prompt, capture_output=True, text=True, timeout=300, cwd=directory,
env={**os.environ, "NO_COLOR": "1"},
)
try:
payload = json.loads(result.stdout or "{}")
except ValueError:
payload = {}
text = str(payload.get("result") or "").strip()
raw_usage = payload.get("usage") if isinstance(payload.get("usage"), dict) else {}
usage = {
"input_tokens": int(raw_usage.get("input_tokens") or 0),
"output_tokens": int(raw_usage.get("output_tokens") or 0),
}
model = str(payload.get("model") or "claude-account-default")
except (OSError, subprocess.TimeoutExpired):
return {"success": False, "reason": "account_model_unreachable"}
if result.returncode != 0 or not text:
return {"success": False, "reason": "account_model_failed"}
return {
"success": True,
"provider": provider,
"model": model,
"text": text,
"usage": usage,
"latency_ms": int((time.monotonic() - started) * 1000),
}
class BridgeHandler(BaseHTTPRequestHandler):
server_version = "CASANAuthBridge/1.0"
@@ -130,6 +193,19 @@ class BridgeHandler(BaseHTTPRequestHandler):
supplied = self.headers.get("X-CASAN-Bridge-Token", "")
return bool(expected) and hmac.compare_digest(expected, supplied)
def read_json(self) -> dict:
try:
length = int(self.headers.get("Content-Length", "0"))
except ValueError:
return {}
if length < 1 or length > 100000:
return {}
try:
payload = json.loads(self.rfile.read(length).decode("utf-8"))
except (UnicodeDecodeError, ValueError):
return {}
return payload if isinstance(payload, dict) else {}
def do_GET(self) -> None:
if self.path == "/healthz":
self.send_json(200, {"status": "ok"})
@@ -151,6 +227,11 @@ class BridgeHandler(BaseHTTPRequestHandler):
result = start_login(parts[2])
self.send_json(202 if result["success"] else 503, result)
return
if len(parts) == 4 and parts[:2] == ["v1", "models"] and parts[3] == "generate" and parts[2] in PROVIDERS:
body = self.read_json()
result = generate_with_account(parts[2], str(body.get("prompt") or ""))
self.send_json(200 if result["success"] else 503, result)
return
self.send_json(404, {"success": False, "reason": "not_found"})