Agent Settings

Agent metadata and optional runtime options live in settings.json at the project root. The file is created by genie setup. Edit it directly, then run genie update to push metadata changes to the AsterQuanta platform without restarting the agent.

Standard fields

Field	Description
name	Agent name shown on the platform.
description	Short description of the agent.
hyper_parameters	Default hyper-parameters copied into new models and used as a baseline for training.
is_public	Whether the agent is visible to other users on the platform.

Example:

{
    "name": "MyAgent",
    "description": "RL optimizer for analog circuits",
    "hyper_parameters": {
        "learning_rate": 0.0003
    },
    "is_public": false
}

Resource usage logging

You can enable periodic CPU and memory logging for a running agent. When enabled, the ADK writes one JSON line per sample to a local log file while the agent is connected to the platform. Samples are taken on the agent heartbeat, so logging respects the configured interval rather than writing on every internal event.

This is useful when running agents on shared machines, in Docker, or when you want a local record of load during long optimizations.

Configuration

Add or edit resource_logging_options in settings.json:

{
    "name": "MyAgent",
    "resource_logging_options": {
        "enabled": true,
        "file_name": "logs/resources.log",
        "interval_seconds": 10,
        "max_size": 10000000,
        "backup_count": 10,
        "child_process_names": []
    }
}

Option	Default	Description
enabled	false	Turn resource logging on or off.
file_name	logs/resources.log	Path to the JSONL log file (parent directories are created automatically).
interval_seconds	10	Minimum seconds between samples.
max_size	10000000	Maximum log file size in bytes before rotation.
backup_count	10	Number of rotated backup files to keep.
child_process_names	[]	Optional process name substrings. Matching child processes are included in memory totals (for example simulator or worker processes started by your agent).
service_name	(agent name)	Label written into each log record. Defaults to the agent name when omitted.

Changes take effect the next time you start the agent (genie run or python src/main.py).

Log format

Each line is a JSON object. Typical fields include:

Field	Description
ts	UTC timestamp of the sample.
cpu_pct	Process CPU usage percentage.
load_1m	System load average (1 minute).
process_rss_mb	Resident memory of the agent process (MB).
children_rss_mb	Resident memory of matched child processes (MB).
child_count	Number of matched child processes.
service	Service name (usually your agent name).
agent_name	Agent name from settings.
active_optimizations	Number of optimizations currently running in this process.

Example record:

{"ts":"2026-06-23T12:00:00Z","cpu_pct":12.5,"load_1m":0.42,"process_rss_mb":256.0,"children_rss_mb":0.0,"child_count":0,"service":"MyAgent","agent_name":"MyAgent","active_optimizations":1}

tip

Add logs/ to your .gitignore if you enable resource logging in development and do not want log files committed to version control.