Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

test(vision): cover the empty reply, which is how the reasoning lane actually fails

nirholas7c5326b4157d
diff --git a/tests/vision-json-lane-failover.test.js b/tests/vision-json-lane-failover.test.js
index 04389bc8c..d8a888163 100644
--- a/tests/vision-json-lane-failover.test.js
+++ b/tests/vision-json-lane-failover.test.js
@@ -82,6 +82,22 @@ describe('describeImageJson lane acceptance', () => {
expect(out.json).toEqual(VERDICT);
});
+ it('falls through an empty reply, which is how a reasoning model fails in production', async () => {
+ // Observed live on 2026-09-10 against /api/vision: the winning lane was
+ // openrouter nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free and it
+ // returned text: ''. The model spends its budget on reasoning tokens and
+ // leaves message.content empty, which extractText faithfully reports as an
+ // empty string. An empty reply is a failed lane, not an answer.
+ fetchSpy
+ .mockResolvedValueOnce(reply(''))
+ .mockResolvedValueOnce(reply(JSON.stringify(VERDICT)));
+
+ const out = await call();
+
+ expect(fetchSpy).toHaveBeenCalledTimes(2);
+ expect(out.json).toEqual(VERDICT);
+ });
+
it('throws only once every rung has failed the shape test', async () => {
fetchSpy.mockResolvedValue(reply('not json at all'));