Open-source AI Security Operations Center: alert fusion, LLM-agent triage, MITRE ATT&CK investigation, and a replayable decision ledger for every agent step. Self-hostable, runs with no API keys, MIT licensed. Ships an MCP server for Claude, Cursor and Continue.
Python agentic-aiai-securityalert-triageblue-teamcybersecuritydetection-engineeringincident-responselanggraphllm-agentsmcpmitre-attackmodel-context-protocolpurple-teamsecurity-operationsself-hostedsiemsoarsocthreat-detectionthreat-intelligence
'use client';
/**
* Where AI triage is running, and the ways to change it.
*
* Shown inside the setup checklist's "Choose where the AI runs" step. It
* reports rather than configures, for the two local options, because it has to:
* switching the bundled Ollama onto a GPU means restarting that container with
* different compose arguments, and a web service that could do that would need
* the Docker socket -- which is a container-escape path, not a feature.
*
* The hosted-provider option *is* configurable from here, because that is a
* database row rather than a container restart. It links to the existing BYOK
* form rather than duplicating it.
*/
import Link from 'next/link';
import useSWR from 'swr';
import { llmApi, type LlmRuntime } from '@/lib/api';
/** How each placement reads to someone deciding what to do about it. */
const TONE: Record<string, { label: string; className: string }> = {
gpu: { label: 'On a GPU', className: 'border-emerald-700/60 bg-emerald-950/30 text-emerald-200' },
partial: { label: 'Partly on a GPU', className: 'border-amber-700/60 bg-amber-950/30 text-amber-200' },
cpu: { label: 'On the CPU', className: 'border-slate-600 bg-slate-800/40 text-slate-200' },
unknown: { label: 'Not loaded right now', className: 'border-slate-600 bg-slate-800/40 text-slate-300' },
unreachable: { label: 'No local model', className: 'border-slate-600 bg-slate-800/40 text-slate-300' },
};
export function ModelPlacement() {
const { data, error, isLoading } = useSWR<LlmRuntime>(
'onboarding:llm-runtime',
() => llmApi.runtime(),
// No `fallbackData`: supplying one disables revalidation, so a placeholder
// would be what this panel *shows* rather than a first paint. There is no
// invented placement better than saying the probe failed.
{ revalidateOnFocus: false, shouldRetryOnError: false },
);
if (isLoading) {
return <p className="mt-2 text-sm text-slate-500">Asking the model where it is running…</p>;
}
if (error || !data) {
return (
<p className="mt-2 text-sm text-slate-400">
Could not determine where the model is running. Triage is unaffected; this panel is
informational.{' '}
<Link href="/settings" className="underline hover:text-slate-200">
Configure a provider
</Link>
</p>
);
}
const tone = TONE[data.placement] ?? TONE.unknown;
return (
<div className="mt-2 space-y-2">
<div className={`inline-flex items-center gap-2 rounded-md border px-2.5 py-1 text-xs ${tone.className}`}>
<span className="font-medium">{tone.label}</span>
{typeof data.vram_bytes === 'number' && data.vram_bytes > 0 && (
<span className="opacity-80">{(data.vram_bytes / 1024 ** 3).toFixed(1)} GB in VRAM</span>
)}
</div>
<p className="text-sm text-slate-400">{data.detail}</p>
{data.options.length > 0 && (
<ul className="space-y-1 text-sm text-slate-400">
{data.options.map((option) => (
<li key={option} className="flex gap-2">
<span aria-hidden className="text-slate-600">
·
</span>
<span>{option}</span>
</li>
))}
</ul>
)}
<Link
href="/settings?tab=deployment"
className="mt-1 inline-block rounded-md border border-slate-600 px-3 py-1.5 text-sm text-slate-200 transition hover:border-slate-400 hover:bg-slate-800"
>
Use my own provider
</Link>
</div>
);
}
export default ModelPlacement;