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.

fix(emotes): let a failed clip-manifest load be retried instead of caching the failure

nirholasff8faa6c6252
diff --git a/src/voice/talk-emotes.js b/src/voice/talk-emotes.js
index 59c823c0b..7691a53b5 100644
--- a/src/voice/talk-emotes.js
+++ b/src/voice/talk-emotes.js
@@ -86,6 +86,12 @@ export class TalkEmotes {
*/
loadManifest() {
if (this._manifestPromise) return this._manifestPromise;
+ // Only a SUCCESSFUL load is memoized. Caching the failure too meant one
+ // dropped request disabled every clip for the rest of the session and no
+ // retry could ever reach the network again, so the caller's "try again"
+ // was a button that could not work. Clearing the memo on failure keeps
+ // concurrent callers sharing the in-flight request while still letting
+ // the next attempt re-fetch.
this._manifestPromise = (async () => {
try {
const r = await fetch('/animations/manifest.json');
@@ -100,7 +106,16 @@ export class TalkEmotes {
log.warn('[talk-emotes] manifest fetch failed:', err?.message);
return false;
}
- })();
+ })().then(
+ (ok) => {
+ if (!ok) this._manifestPromise = null;
+ return ok;
+ },
+ (err) => {
+ this._manifestPromise = null;
+ throw err;
+ },
+ );
return this._manifestPromise;
}