fix(ai-review): cap BYOK daily review usage

src/db/repositories.ts

@@ -1,4 +1,4 @@
 import { and, desc, eq, gte, inArray, not, or, sql, type SQL } from "drizzle-orm";
 import { getDb } from "./client";
 import {
   advisories,
@@ -2145,6 +2145,24 @@
   return Number(row?.total ?? 0);
 }
 
+export async function countByokAiReviewEventsForRepoSince(env: Env, repoFullName: string, sinceIso: string): Promise<number> {
+  const db = getDb(env.DB);
+  const [row] = await db
+    .select({ total: sql<number>`count(*)` })
+    .from(aiUsageEvents)
+    .where(
+      and(
+        gte(aiUsageEvents.createdAt, sinceIso),
+        eq(aiUsageEvents.feature, "ai_review_pr"),
+        eq(aiUsageEvents.status, "ok"),
+        sql`${aiUsageEvents.model} like 'byok:%'`,
+        sql`json_extract(${aiUsageEvents.metadataJson}, '$.repoFullName') = ${repoFullName}`,
+      ),
+    );
+  /* v8 ignore next -- SQL aggregate count always returns one row; fallback protects D1 driver anomalies. */
+  return Number(row?.total ?? 0);
+}
+
 export async function upsertContributorScoringProfile(env: Env, profile: ContributorScoringProfileRecord): Promise<void> {
   const db = getDb(env.DB);
   await db

src/env.d.ts

@@ -1,4 +1,4 @@
 declare global {
   interface Env {
     DB: D1Database;
     JOBS: Queue;
@@ -10,6 +10,8 @@
     AI_PUBLIC_COMMENTS_ENABLED?: string;
     WORKERS_AI_SUMMARY_MODEL?: string;
     AI_DAILY_NEURON_BUDGET?: string;
+    /** Per-repository/day cap for maintainer-paid BYOK AI review provider calls. */
+    AI_BYOK_DAILY_REPO_LIMIT?: string;
     AI_MAX_OUTPUT_TOKENS?: string;
     /** Optional Cloudflare AI Gateway id. When set, free Workers-AI review calls route through the gateway
      *  for caching, rate-limiting, request logging, and fallback. Unset = direct binding calls (unchanged). */

src/services/ai-review.ts

@@ -1,4 +1,4 @@
 // Gittensory AI maintainer review (the `aiReview` capability).
 //
 // Two layers, both opt-in and both fail-safe (no AI / errors / over-budget / unsafe output → no public
 // text and no gate finding; gittensory NEVER blocks because the model spoke):
@@ -13,9 +13,10 @@
 //     confirmed Gittensor contributors (the gate enforces that downstream).
 //
 // Every public string (notes + defect title/detail) is forced through `sanitizePublicComment`; anything
-// that trips the public/private boundary is dropped, not published. Every model call is metered against
-// the shared daily neuron budget and audited via `recordAiUsageEvent`.
-import { recordAiUsageEvent, sumAiEstimatedNeuronsSince } from "../db/repositories";
+// that trips the public/private boundary is dropped, not published. Free Workers-AI calls are metered against
+// the shared daily neuron budget; maintainer-paid BYOK calls have a separate repo/day cap. All calls
+// are audited via `recordAiUsageEvent`.
+import { countByokAiReviewEventsForRepoSince, recordAiUsageEvent, sumAiEstimatedNeuronsSince } from "../db/repositories";
 import { sanitizePublicComment } from "../queue-intelligence";
 
 /**
@@ -227,6 +228,9 @@
  *  the existing fail-safe null path. Mirrors the github/gittensor fetch-timeout convention. */
 const AI_PROVIDER_TIMEOUT_MS = 20_000;
 
+/** Default per-repository/day cap for maintainer-paid BYOK advisory calls. */
+const DEFAULT_BYOK_DAILY_REPO_LIMIT = 25;
+
 /** Why a BYOK advisory call produced no review — surfaced in the audit event for observability (never a key). */
 type ProviderFailure = "timeout" | "http_error" | "exception";
 type ProviderReviewOutcome = { review: ModelReview | null; failure?: ProviderFailure };
@@ -328,6 +332,15 @@
     return { status: "quota_exceeded", estimatedNeurons, remainingBudget };
   }
 
+  if (input.providerKey) {
+    const byokDailyLimit = clampNumber(Number(env.AI_BYOK_DAILY_REPO_LIMIT || DEFAULT_BYOK_DAILY_REPO_LIMIT), 0, 10_000);
+    const byokUsed = await countByokAiReviewEventsForRepoSince(env, input.repoFullName, utcDayStartIso());
+    if (byokUsed >= byokDailyLimit) {
+      await record(env, input, "quota_exceeded", 0, `BYOK daily repo limit ${byokDailyLimit} reached`);
+      return { status: "quota_exceeded", estimatedNeurons, remainingBudget };
+    }
+  }
+
   // Advisory write-up: BYOK frontier model if configured, else the free Workers-AI primary (with fallback).
   let byokFailure: ProviderFailure | undefined;
   let advisoryReview: ModelReview | null;

test/unit/ai-review.test.ts

@@ -1,4 +1,4 @@
 import { afterEach, describe, expect, it, vi } from "vitest";
 import {
   __aiReviewInternals,
   AI_CONSENSUS_FLOOR,
@@ -64,15 +64,30 @@
     const fetchMock = vi.fn(async () => new Response(JSON.stringify({ content: [{ type: "text", text: reviewJson({ assessment: "BYOK advisory." }) }] }), { status: 200 }));
     vi.stubGlobal("fetch", fetchMock);
     const run = vi.fn();
-    // Free budget is exhausted (1 neuron), but a BYOK advisory bills the maintainer's account, so it still runs.
-    const env = createTestEnv({ AI: { run } as unknown as Ai, AI_SUMMARIES_ENABLED: "true", AI_PUBLIC_COMMENTS_ENABLED: "true", AI_DAILY_NEURON_BUDGET: "1" });
+    // Free budget is exhausted (1 neuron), but a BYOK advisory bills the maintainer's account, so it still runs
+    // while the separate BYOK repo/day quota has capacity.
+    const env = createTestEnv({ AI: { run } as unknown as Ai, AI_SUMMARIES_ENABLED: "true", AI_PUBLIC_COMMENTS_ENABLED: "true", AI_DAILY_NEURON_BUDGET: "1", AI_BYOK_DAILY_REPO_LIMIT: "1" });
     const result = await runGittensoryAiReview(env, { ...baseInput, providerKey: { provider: "anthropic", key: "sk-ant-secret" } });
     expect(result.status).toBe("ok");
     expect(result.status === "ok" && result.advisoryNotes).toContain("BYOK advisory.");
     expect(result.status === "ok" && result.estimatedNeurons).toBe(0); // advisory-only BYOK consumes no free budget
     expect(fetchMock).toHaveBeenCalled();
     expect(run).not.toHaveBeenCalled();
   });
+
+  it("enforces a separate per-repo daily quota before BYOK provider calls", async () => {
+    const fetchMock = vi.fn(async () => new Response(JSON.stringify({ content: [{ type: "text", text: reviewJson({ assessment: "BYOK advisory." }) }] }), { status: 200 }));
+    vi.stubGlobal("fetch", fetchMock);
+    const run = vi.fn();
+    const env = createTestEnv({ AI: { run } as unknown as Ai, AI_SUMMARIES_ENABLED: "true", AI_PUBLIC_COMMENTS_ENABLED: "true", AI_DAILY_NEURON_BUDGET: "0", AI_BYOK_DAILY_REPO_LIMIT: "1" });
+    const providerKey = { provider: "anthropic" as const, key: "sk-ant-secret" };
+
+    await expect(runGittensoryAiReview(env, { ...baseInput, providerKey })).resolves.toMatchObject({ status: "ok" });
+    await expect(runGittensoryAiReview(env, { ...baseInput, prNumber: 8, providerKey })).resolves.toMatchObject({ status: "quota_exceeded" });
+
+    expect(fetchMock).toHaveBeenCalledTimes(1);
+    expect(run).not.toHaveBeenCalled();
+  });
 });
 
 describe("AI Gateway routing for free Workers-AI calls", () => {

wrangler.jsonc

@@ -1,4 +1,4 @@
 {
   "$schema": "node_modules/wrangler/config-schema.json",
   // Auto-deploys to production via Cloudflare Workers Builds on push to `main`
   // (deploy command: `npm run deploy:api` — applies pending D1 migrations, then `wrangler deploy`).
@@ -38,6 +38,7 @@
     "AI_PUBLIC_COMMENTS_ENABLED": "false",
     "WORKERS_AI_SUMMARY_MODEL": "@cf/meta/llama-3.1-8b-instruct-fp8-fast",
     "AI_DAILY_NEURON_BUDGET": "10000",
+    "AI_BYOK_DAILY_REPO_LIMIT": "25",
     "AI_MAX_OUTPUT_TOKENS": "256",
     "AI_GATEWAY_ID": "",
     "ADMIN_GITHUB_LOGINS": "JSONbored",