daemon_main.py

Go to the documentation of this file.

# Copyright (c) 2025-present Aquin Labs Private Limited. All Rights Reserved.
# This file is part of the Aquin Engine. Unauthorized copying, modification,
# distribution, or use of this file, via any medium, is strictly prohibited.
# Proprietary and confidential. See LICENSE for terms.
 
"""
Standalone persistent model daemon.
 
Started detached by `aquin load --model` (see aquin.engine.model_daemon). It holds
one model resident in VRAM across CLI invocations and serves generic tool dispatch
plus model lifecycle over localhost, so commands like `aquin attention` do not
reload the model every time.
 
Run:  python -m aquin.engine.daemon_main [--port PORT] [--model <id>]
Default port: AQUIN_ENGINE_PORT → ~/.aquin/engine-info.json → 17832 (not 8002).
"""
from __future__ import annotations
 
import argparse
import os
import sys
 
 
def main() -> None:
    from aquin.engine.engine_info import resolve_engine_port, set_daemon_state
 
    parser = argparse.ArgumentParser(prog="aquin-daemon")
    parser.add_argument("--port", type=int, default=None)
    parser.add_argument("--model", default=None, help="optional model to preload")
    ns = parser.parse_args()
    port = ns.port if ns.port is not None else resolve_engine_port()
 
    # Mark this process as the daemon so in-process loaders do not try to free
    # "another" daemon's VRAM (that guard is meant for external short-lived procs).
    os.environ["AQUIN_DAEMON"] = "1"
 
    try:
        from aquin.quiet import install_cli_quiet_mode
        install_cli_quiet_mode()
    except Exception:
        pass
 
    # Ensure tool stubs are registered and the loader shim is applied.
    try:
        from aquin.engine.tools.registry import _load_stubs
        _load_stubs()
    except Exception as exc:
        print(f"[daemon] tool registry load warning: {exc}", flush=True)
    try:
        from aquin.compute.loader_shim import apply as _shim_apply
        _shim_apply()
    except Exception as exc:
        print(f"[daemon] loader shim warning: {exc}", flush=True)
 
    if ns.model:
        try:
            from aquin.compute.model_runtime import load_weights
 
            print(f"[daemon] preloading {ns.model}...", flush=True)
            load_weights(ns.model)
            print("[daemon] preload complete", flush=True)
        except Exception as exc:
            print(f"[daemon] preload failed: {exc}", flush=True)
 
    try:
        set_daemon_state(
            status="running",
            pid=os.getpid(),
            port=port,
            model_id=ns.model,
        )
    except Exception as exc:
        print(f"[daemon] engine-info write warning: {exc}", flush=True)
 
    from aquin.engine.local_server import run_forever
    try:
        run_forever(port)
    finally:
        try:
            set_daemon_state(status="stopped", pid=None, started_at=None, model_id=None)
        except Exception:
            pass
 
 
if __name__ == "__main__":
    sys.exit(main())