diff --git a/packages/ai/src/evals/eval.ts b/packages/ai/src/evals/eval.ts
index bf253472..15647703 100644
--- a/packages/ai/src/evals/eval.ts
+++ b/packages/ai/src/evals/eval.ts
@@ -1,4 +1,5 @@
 import { afterAll, beforeAll, describe, inject, it } from 'vitest';
+import type { RunnerTestFile, RunnerTestSuite } from 'vitest';
 import { context, SpanStatusCode, trace, type Context } from '@opentelemetry/api';
 import { customAlphabet } from 'nanoid';
 import { withEvalContext, getEvalContext, getConfigScope } from './context/storage';
@@ -57,8 +58,27 @@ type RunTaskFailureDetails = {
   overrides?: Record<string, any>;
 };
 
+type EvalHookSuite = (RunnerTestSuite | RunnerTestFile) & {
+  meta: Record<string, any>;
+  tasks: Array<{ meta: { case?: EvalCaseReport } }>;
+};
+
+type CompatibleSuiteHook = {
+  (suite: RunnerTestSuite | RunnerTestFile): Promise<void> | void;
+  (context: unknown, suite: RunnerTestSuite | RunnerTestFile): Promise<void> | void;
+};
+
 const RUN_TASK_FAILURE_DETAILS = Symbol.for('axiom.eval.runTaskFailureDetails');
 
+function withCompatibleSuiteHook(
+  fn: (suite: EvalHookSuite) => Promise<void> | void,
+): CompatibleSuiteHook {
+  return async function ({}: any, maybeSuite?: RunnerTestSuite | RunnerTestFile): Promise<void> {
+    const suite = (maybeSuite ?? arguments[0]) as EvalHookSuite;
+    await fn(suite);
+  } as CompatibleSuiteHook;
+}
+
 function attachRunTaskFailureDetails(
   error: unknown,
   details: RunTaskFailureDetails,
@@ -238,7 +258,7 @@ async function registerEval<
         | { flags: Record<string, any>; pickedFlags?: string[]; overrides?: Record<string, any> }
         | undefined;
 
-      beforeAll(async (suite) => {
+      const handleBeforeAll = async (suite: EvalHookSuite) => {
         // Ensure worker process knows CLI overrides
         if (injectedOverrides && Object.keys(injectedOverrides).length > 0) {
           try {
@@ -353,9 +373,9 @@ async function registerEval<
         };
 
         suiteStart = performance.now();
-      });
+      };
 
-      afterAll(async (suite) => {
+      const handleAfterAll = async (suite: EvalHookSuite) => {
         if (instrumentationError) {
           throw instrumentationError;
         }
@@ -417,10 +437,10 @@ async function registerEval<
         const durationMs = Math.round(performance.now() - suiteStart);
 
         const successCases = suite.tasks.filter(
-          (task) => task.meta.case.status === 'success',
+          (task) => task.meta.case?.status === 'success',
         ).length;
         const erroredCases = suite.tasks.filter(
-          (task) => task.meta.case.status === 'fail' || task.meta.case.status === 'pending',
+          (task) => task.meta.case?.status === 'fail' || task.meta.case?.status === 'pending',
         ).length;
 
         // signal Axiom that evaluation finished to kick of summary calculations
@@ -434,7 +454,10 @@ async function registerEval<
             durationMs,
           });
         }
-      });
+      };
+
+      beforeAll(withCompatibleSuiteHook(handleBeforeAll));
+      afterAll(withCompatibleSuiteHook(handleAfterAll));
 
       type CollectionRecordWithIndex = { index: number } & CollectionRecord<TInput, TExpected>;