feat(vertex): self-tracked USD spend since account added (#3724)

Integrated into release/v3.8.23

open-sse/services/usage.ts

@@ -1477,6 +1477,8 @@ export const USAGE_FETCHER_PROVIDERS = [
   "opencode",
   "opencode-zen",
   "xiaomi-mimo",
+  "vertex",
+  "vertex-partner",
 ] as const;
 
 export type UsageFetcherProvider = (typeof USAGE_FETCHER_PROVIDERS)[number];
@@ -1516,6 +1518,9 @@ export async function getUsageForProvider(
     case "kiro":
     case "amazon-q":
       return await getKiroUsage(accessToken, providerSpecificData);
+    case "vertex":
+    case "vertex-partner":
+      return await getVertexUsage(id || "", provider);
     case "kimi-coding":
       return await getKimiUsage(accessToken);
     case "qwen":
@@ -3103,6 +3108,53 @@ async function getKiroUsage(accessToken?: string, providerSpecificData?: JsonRec
   }
 }
 
+/**
+ * Vertex AI — SELF-TRACKED spend.
+ *
+ * Vertex AI exposes no usage/quota API for an API key or Service Account (billing/credit balance
+ * lives behind the Cloud Billing API, which the proxy credential can't reach). Instead we report
+ * the USD that OmniRoute has spent through this connection since the account was added — summed
+ * from `usage_history` and priced via the backend pricing table. Returns a `message` (with the $
+ * figure) plus a `spend` quota entry so the limits cache persists it (a message-only result is
+ * treated as a transient error and not cached).
+ */
+async function getVertexUsage(connectionId: string, provider: string) {
+  if (!connectionId) {
+    return { message: "Vertex connected. Connection id unavailable for usage tracking." };
+  }
+  try {
+    const { getConnectionSpendUsdSinceAdded } = await import("@/lib/usage/usageStats");
+    const { costUsd, requests } = await getConnectionSpendUsdSinceAdded(provider, connectionId);
+
+    const spend: JsonRecord = {
+      used: Number(costUsd.toFixed(6)),
+      displayName: "Spend (USD)",
+      quotaSource: "localUsageHistory",
+      resetAt: null,
+      unlimited: false,
+    };
+
+    if (requests === 0) {
+      return {
+        plan: "Vertex AI",
+        message: "Vertex connected. No usage recorded through OmniRoute yet for this account.",
+        quotas: { spend },
+      };
+    }
+
+    const costStr = costUsd >= 1 ? costUsd.toFixed(2) : costUsd.toFixed(4);
+    return {
+      plan: "Vertex AI",
+      message: `$${costStr} used since this account was added \u00b7 ${requests} request${
+        requests === 1 ? "" : "s"
+      }`,
+      quotas: { spend },
+    };
+  } catch (error) {
+    return { message: `Vertex usage tracking error: ${(error as Error).message}` };
+  }
+}
+
 /**
  * Map Kimi membership level to display name
  * LEVEL_BASIC = Moderato, LEVEL_INTERMEDIATE = Allegretto,
@@ -3331,6 +3383,7 @@ export const __testing = {
   getMiniMaxRemainingPercent,
   getMiniMaxUsage,
   getXiaomiMimoUsage,
+  getVertexUsage,
   getMiniMaxAuthErrorMessage,
   getMiniMaxErrorSummary,
   mapCodeAssistSubscriptionToPlanLabel,

src/lib/usage/providerLimits.ts

@@ -70,6 +70,8 @@ const PROVIDER_LIMITS_APIKEY_PROVIDERS = new Set([
   "nanogpt",
   "deepseek",
   "xiaomi-mimo",
+  "vertex",
+  "vertex-partner",
 ]);
 const DEFAULT_PROVIDER_LIMITS_SYNC_INTERVAL_MINUTES = 70;
 const PROVIDER_LIMITS_AUTO_SYNC_SETTING_KEY = "provider_limits_auto_sync_last_run";

src/lib/usage/usageStats.ts

@@ -219,6 +219,65 @@ export function getMonthlyProviderTokensForConnection(
   return Math.max(0, Number(row?.total ?? 0));
 }
 
+/**
+ * Total USD spend OmniRoute has recorded for a single provider connection, across all time
+ * (i.e. "since the account was added" — usage_history rows only exist from first use onward).
+ *
+ * Sums per-model token usage from `usage_history` for the connection and prices each model via the
+ * backend pricing table (`calculateCost`). Scoped to the given `provider` and to **successful**
+ * requests (`success = 1`) so failed/errored calls and any cross-provider rows can't inflate the
+ * total. Only reflects traffic that went THROUGH OmniRoute, not the provider's own dashboard. Used
+ * to surface a "$X used since added" figure for providers that expose no native usage/quota API
+ * (e.g. Vertex AI).
+ */
+export async function getConnectionSpendUsdSinceAdded(
+  provider: string,
+  connectionId: string
+): Promise<{ costUsd: number; requests: number }> {
+  if (!provider || !connectionId) return { costUsd: 0, requests: 0 };
+
+  const db = getDbInstance();
+  const rows = db
+    .prepare(
+      `SELECT model,
+          COALESCE(SUM(tokens_input), 0) AS input,
+          COALESCE(SUM(tokens_output), 0) AS output,
+          COALESCE(SUM(tokens_cache_read), 0) AS cacheRead,
+          COALESCE(SUM(tokens_cache_creation), 0) AS cacheCreation,
+          COALESCE(SUM(tokens_reasoning), 0) AS reasoning,
+          COUNT(*) AS requests
+       FROM usage_history
+       WHERE connection_id = ? AND provider = ? AND success = 1
+       GROUP BY model`
+    )
+    .all(connectionId, provider) as Array<{
+    model?: string;
+    input?: number;
+    output?: number;
+    cacheRead?: number;
+    cacheCreation?: number;
+    reasoning?: number;
+    requests?: number;
+  }>;
+
+  let costUsd = 0;
+  let requests = 0;
+  for (const row of rows) {
+    requests += Math.max(0, Number(row.requests ?? 0));
+    const model = typeof row.model === "string" ? row.model : "";
+    const tokens = {
+      input: Number(row.input ?? 0),
+      output: Number(row.output ?? 0),
+      cacheRead: Number(row.cacheRead ?? 0),
+      cacheCreation: Number(row.cacheCreation ?? 0),
+      reasoning: Number(row.reasoning ?? 0),
+    };
+    costUsd += await calculateCost(provider, model, tokens, { provider, model });
+  }
+
+  return { costUsd: Math.max(0, costUsd), requests };
+}
+
 /**
  * Get aggregated usage stats.
  * Uses UNION of recent raw data and older aggregated data when aggregation is enabled.

src/shared/constants/providers.ts

@@ -3127,6 +3127,8 @@ export const USAGE_SUPPORTED_PROVIDERS = [
   "nanogpt",
   "deepseek",
   "xiaomi-mimo",
+  "vertex",
+  "vertex-partner",
 ];
 
 // ── Zod validation at module load (Phase 7.2) ──

tests/unit/vertex-spend-usage.test.ts

@@ -0,0 +1,105 @@
+/**
+ * tests/unit/vertex-spend-usage.test.ts
+ *
+ * Vertex AI exposes no native usage/quota API for an API key or Service Account, so OmniRoute
+ * SELF-TRACKS spend: it sums the tokens it routed to the connection (usage_history) and prices
+ * them via the backend pricing table, surfacing a "$X used since this account was added" figure.
+ * These tests cover the aggregation helper + the fetcher response shape with a real temp DB.
+ */
+
+import { describe, it, before, after } from "node:test";
+import assert from "node:assert/strict";
+import os from "node:os";
+import path from "node:path";
+import fs from "node:fs";
+
+// DATA_DIR must be set before any module that opens the DB is imported.
+const TMP = fs.mkdtempSync(path.join(os.tmpdir(), "omni-vertex-"));
+process.env.DATA_DIR = TMP;
+
+const core = await import("../../src/lib/db/core.ts");
+const { getConnectionSpendUsdSinceAdded } = await import("../../src/lib/usage/usageStats.ts");
+const { __testing } = await import("../../open-sse/services/usage.ts");
+const { getVertexUsage } = __testing;
+
+function insertUsage(
+  connectionId: string,
+  provider: string,
+  model: string,
+  tokensIn: number,
+  tokensOut: number,
+  success = 1
+) {
+  const db = core.getDbInstance();
+  db.prepare(
+    `INSERT INTO usage_history (provider, model, connection_id, tokens_input, tokens_output, success, timestamp)
+     VALUES (?, ?, ?, ?, ?, ?, ?)`
+  ).run(provider, model, connectionId, tokensIn, tokensOut, success, new Date().toISOString());
+}
+
+describe("vertex self-tracked spend", () => {
+  before(() => {
+    core.getDbInstance(); // trigger migrations
+    // conn-v: two SUCCESSFUL priced requests across two models.
+    insertUsage("conn-v", "vertex", "gemini-2.5-flash", 1_000_000, 500_000, 1);
+    insertUsage("conn-v", "vertex", "gemini-3-pro-image-preview", 200_000, 100_000, 1);
+    // a FAILED request on the same connection must NOT count toward spend.
+    insertUsage("conn-v", "vertex", "gemini-2.5-flash", 5_000_000, 5_000_000, 0);
+    // a row with a different provider on the same connection id must NOT bleed in.
+    insertUsage("conn-v", "vertex-partner", "claude-opus-4-7", 9_000_000, 9_000_000, 1);
+    // a different connection must not bleed in
+    insertUsage("conn-other", "vertex", "gemini-2.5-flash", 9_000_000, 9_000_000, 1);
+  });
+
+  after(() => {
+    core.resetDbInstance();
+    try {
+      fs.rmSync(TMP, { recursive: true, force: true });
+    } catch {
+      // best-effort temp cleanup
+    }
+  });
+
+  it("counts only the connection's successful, same-provider requests", async () => {
+    const { costUsd, requests } = await getConnectionSpendUsdSinceAdded("vertex", "conn-v");
+    assert.equal(
+      requests,
+      2,
+      "only the two successful vertex rows count (failed + vertex-partner + other-conn excluded)"
+    );
+    assert.ok(Number.isFinite(costUsd) && costUsd >= 0, "cost is a finite, non-negative number");
+  });
+
+  it("returns 0/0 for an unknown connection (no bleed)", async () => {
+    const { costUsd, requests } = await getConnectionSpendUsdSinceAdded("vertex", "conn-none");
+    assert.equal(requests, 0);
+    assert.equal(costUsd, 0);
+  });
+
+  it("getVertexUsage returns a spend quota + $ message for a used connection", async () => {
+    const r = (await getVertexUsage("conn-v", "vertex")) as {
+      plan?: string;
+      message?: string;
+      quotas?: Record<string, { used: number; quotaSource?: string; displayName?: string }>;
+    };
+    assert.ok(r.quotas?.spend, "spend quota present (so the limits cache persists it)");
+    assert.equal(r.quotas!.spend.quotaSource, "localUsageHistory");
+    assert.ok(typeof r.quotas!.spend.used === "number" && r.quotas!.spend.used >= 0);
+    assert.ok(r.message && r.message.includes("$"), "message carries the dollar figure");
+    assert.ok(r.message!.includes("2 requests"), "message reports the request count");
+  });
+
+  it("getVertexUsage reports no-usage cleanly when nothing was routed", async () => {
+    const r = (await getVertexUsage("conn-empty", "vertex")) as {
+      message?: string;
+      quotas?: Record<string, { used: number }>;
+    };
+    assert.ok(r.message && /no usage/i.test(r.message), "informative no-usage message");
+    assert.equal(r.quotas?.spend.used, 0);
+  });
+
+  it("getVertexUsage returns a message when connection id is missing", async () => {
+    const r = (await getVertexUsage("", "vertex")) as { message?: string; quotas?: unknown };
+    assert.ok(r.message && !r.quotas, "no spend quota without a connection id");
+  });
+});