🩹 fix: Repaired Output in Gauge, Cache-Rate Keys, Config Gate, Usage Cleanup

- live/completed gauge counts the repaired completion (normalized output),
  so under-reporting providers don't drop the response from used context
- translate static tokenConfig cacheWrite/cacheRead onto the write/read
  keys getCacheMultiplier reads, so cache tokens bill at the configured
  rate instead of the prompt-rate fallback
- clear the token index and usage atoms when leaving a conversation, so
  visited histories don't accumulate in memory for the tab's lifetime
- wait for startupConfig before mounting the gauge, so a deployment with
  contextUsage disabled never briefly mounts it or fires the token-config
  query on first load

packages/api/src/endpoints/custom/initialize.spec.ts

@@ -376,9 +376,21 @@ describe('initializeCustom – token-config fetch header forwarding', () => {
       } as unknown as BaseInitializeParams['db'],
     };
 
-    const result = (await initializeCustom(params)) as { endpointTokenConfig?: unknown };
+    const result = (await initializeCustom(params)) as {
+      endpointTokenConfig?: Record<string, Record<string, number>>;
+    };
 
     expect(fetchModels).not.toHaveBeenCalled();
-    expect(result.endpointTokenConfig).toEqual(tokenConfig);
+    /** Original rates pass through, plus the billing-shape cache keys so
+     *  getCacheMultiplier (which reads `write`/`read`) finds them */
+    expect(result.endpointTokenConfig?.['gpt-4']).toEqual({
+      prompt: 1.5,
+      completion: 4.5,
+      context: 32000,
+      cacheRead: 0.3,
+      cacheWrite: 1.8,
+      write: 1.8,
+      read: 0.3,
+    });
   });
 });

packages/api/src/endpoints/custom/initialize.ts

@@ -36,6 +36,29 @@ export function getTokenConfigKey(
     : endpoint;
 }
 
+/**
+ * Maps an admin-facing static `tokenConfig` to the billing shape: the UI uses
+ * `cacheWrite`/`cacheRead`, but `getCacheMultiplier` indexes `write`/`read`.
+ * Adds those keys (preserving the originals) so cache tokens bill at the
+ * configured rate instead of the prompt-rate fallback.
+ */
+function toBillingTokenConfig(
+  tokenConfig: Record<string, Record<string, number>>,
+): EndpointTokenConfig {
+  const result: EndpointTokenConfig = {};
+  for (const [model, rates] of Object.entries(tokenConfig)) {
+    const mapped = { ...rates } as Record<string, number>;
+    if (rates.cacheWrite != null) {
+      mapped.write = rates.cacheWrite;
+    }
+    if (rates.cacheRead != null) {
+      mapped.read = rates.cacheRead;
+    }
+    result[model] = mapped as EndpointTokenConfig[string];
+  }
+  return result;
+}
+
 /**
  * Builds custom options from endpoint configuration
  */
@@ -161,8 +184,12 @@ export async function initializeCustom({
 
   if (hasTokenConfig) {
     /** A static override is authoritative — use it for the agent's billing
-     *  and balance checks, not just the advertised UI token config */
-    endpointTokenConfig = endpointConfig.tokenConfig as EndpointTokenConfig;
+     *  and balance checks, not just the advertised UI token config. Mirror
+     *  the admin-facing `cacheWrite`/`cacheRead` keys onto the `write`/`read`
+     *  keys the billing multiplier reads. */
+    endpointTokenConfig = toBillingTokenConfig(
+      endpointConfig.tokenConfig as Record<string, Record<string, number>>,
+    );
   } else {
     const cachedConfig =
       FetchTokenConfig[endpoint.toLowerCase() as keyof typeof FetchTokenConfig] &&