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';
/**
* What the agents actually did, and what it cost.
*
* `GET /api/v1/costs/dashboard` joins `aisoc_run_costs` to `investigation_runs`
* — real recorded token counts and spend from real runs, not a projection. The
* client existed (`costsApi.dashboard`) and was used only by the admin cost
* page; the throughput reading it supports belongs next to pipeline health,
* because "the agents triaged nothing today" and "the agents cost nothing
* today" are the same row of data read two ways.
*
* `total_runs` is investigation runs, not alerts triaged. The panel labels it
* as runs. Calling it "alerts triaged" would be a different measurement than
* the one the endpoint makes.
*/
import useSWR from 'swr';
import { costsApi } from '@/lib/api';
import { EmptyState } from '@/components/ui/EmptyState';
import { ErrorState } from '@/components/ui/ErrorState';
const WINDOW_DAYS = 7;
function formatUsd(value: number): string {
if (!Number.isFinite(value)) return '—';
return value < 1 ? `$${value.toFixed(4)}` : `$${value.toFixed(2)}`;
}
function formatCompact(value: number): string {
if (!Number.isFinite(value)) return '—';
if (value >= 1_000_000) return `${(value / 1_000_000).toFixed(1)}M`;
if (value >= 1_000) return `${(value / 1_000).toFixed(1)}k`;
return value.toLocaleString();
}
export function AgentThroughputPanel() {
const { data, error, isLoading, mutate } = useSWR(
`ops-agent-cost:${WINDOW_DAYS}`,
() => costsApi.dashboard({ window_days: WINDOW_DAYS }),
{ refreshInterval: 120_000, revalidateOnFocus: false },
);
const headline = data?.headline;
const byModel = data?.by_model ?? [];
const maxModelCost = Math.max(...byModel.map((m) => m.total_cost_usd), 0.0001);
// A bar can only be drawn from a measured figure. Scaling an unmeasured
// zero still draws nothing, but the label beside it would have read
// "$0.00" — a claim the panel has no basis for.
//
// A *missing* count is treated as zero, i.e. not measured. This is the
// opposite of what the MTTR tiles do with a missing sample count, and
// deliberately so: an older API that omits a sample size still sent a real
// mean, whereas an older API that omits these counts sent a figure we now
// know was a list-price guess keyed on a gateway alias. Rendering the
// familiar number is the safe default there and the unsafe one here.
const measuredCalls = headline?.measured_call_count ?? 0;
const estimatedCalls = headline?.estimated_call_count ?? 0;
const anyMeasured = measuredCalls > 0;
return (
<section
aria-labelledby="ops-agent-heading"
className="rounded-xl border border-gray-800/60 bg-gray-900/40"
>
<div className="px-4 py-3">
<h2 id="ops-agent-heading" className="text-sm font-medium text-gray-300">
Agent throughput and spend
</h2>
<p className="mt-0.5 text-[11px] text-gray-500">
Recorded token usage from the last {WINDOW_DAYS} days of investigation runs.
</p>
</div>
{isLoading && !data ? (
<div className="m-4 h-28 animate-pulse rounded bg-gray-800/40" />
) : error ? (
<ErrorState
title="Agent telemetry unavailable"
error={error}
onRetry={() => {
void mutate();
}}
className="m-4"
/>
) : !headline || headline.total_runs === 0 ? (
<EmptyState
title="No agent runs in this window"
description={`Nothing has been investigated in the last ${WINDOW_DAYS} days, so there is no token usage to report.`}
className="m-4"
/>
) : (
<div className="space-y-4 px-4 pb-4">
<div className="grid grid-cols-2 gap-3 sm:grid-cols-4">
<Stat label="Runs" value={headline.total_runs.toLocaleString()} />
<Stat label="LLM calls" value={formatCompact(headline.total_calls)} />
<Stat label="Tokens" value={formatCompact(headline.total_tokens)} />
<Stat
label="Measured spend"
value={anyMeasured ? formatUsd(headline.total_cost_usd) : '—'}
/>
</div>
{/* Only render the derived per-run figure the server computed. When
it is null the server had no basis for it, and dividing here
would invent one. */}
{anyMeasured && headline.avg_cost_per_run_usd != null ? (
<p className="text-[11px] text-gray-500">
{`${formatUsd(headline.avg_cost_per_run_usd)} per run on average, measured over ${measuredCalls.toLocaleString()} calls.`}
</p>
) : (
<p className="text-[11px] text-gray-500">
{estimatedCalls > 0
? `Spend not measured — the gateway reported no cost. List price for the models involved is ~${formatUsd(headline.estimated_cost_usd)}.`
: 'Spend not measured: no call in this window reported a cost.'}
</p>
)}
{byModel.length > 0 && (
<div>
<p className="mb-2 text-[11px] uppercase tracking-wider text-gray-500">
By model
</p>
<ul className="space-y-2">
{byModel.map((m) => (
<li key={m.model}>
<div className="mb-1 flex items-baseline justify-between gap-3">
<span className="truncate font-mono text-[11px] text-gray-300">
{m.model}
</span>
<span className="shrink-0 text-[11px] text-gray-500">
{`${m.runs} runs · ${
(m.measured_call_count ?? 0) > 0
? formatUsd(m.total_cost_usd)
: (m.estimated_call_count ?? 0) > 0
? `~${formatUsd(m.estimated_cost_usd)} est.`
: 'cost not measured'
}`}
</span>
</div>
{anyMeasured && (
<div className="h-1 rounded-full bg-gray-800">
<div
className="h-1 rounded-full bg-indigo-500/60"
style={{
width: `${Math.round((m.total_cost_usd / maxModelCost) * 100)}%`,
}}
/>
</div>
)}
</li>
))}
</ul>
</div>
)}
</div>
)}
</section>
);
}
function Stat({ label, value }: { label: string; value: string }) {
return (
<div>
<p className="text-[11px] uppercase tracking-wider text-gray-500">{label}</p>
<p className="text-xl font-semibold text-gray-100">{value}</p>
</div>
);
}