switchroom 0.19.19 → 0.19.22

package/dist/auth-broker/index.js

@@ -18352,6 +18352,47 @@ for (const name of SHARED_FRAGMENTS) {
   }
 }
 
+// src/litellm/timeout-budget.ts
+var LITELLM_ROUTER_MARGIN_S = 10;
+var LITELLM_TIMEOUT_TIERS = {
+  interactive: {
+    group: "gpt-oss-20b",
+    localTimeoutS: 90,
+    fallbackGroup: "gpt-oss-20b-openrouter",
+    fallbackTimeoutS: 60
+  },
+  retain: {
+    group: "gpt-oss-20b-retain",
+    localTimeoutS: 200,
+    fallbackGroup: "gpt-oss-20b-retain-openrouter",
+    fallbackTimeoutS: 90
+  },
+  consolidation: {
+    group: "gpt-oss-20b-consolidation",
+    localTimeoutS: 200,
+    fallbackGroup: "gpt-oss-20b-consolidation-openrouter",
+    fallbackTimeoutS: 90
+  }
+};
+function litellmChainSeconds(tier) {
+  assertPositive(tier.localTimeoutS, `${tier.group} localTimeoutS`);
+  if (tier.fallbackGroup === null)
+    return tier.localTimeoutS;
+  assertPositive(tier.fallbackTimeoutS, `${tier.fallbackGroup} fallbackTimeoutS`);
+  return tier.localTimeoutS + tier.fallbackTimeoutS;
+}
+function minimumClientBudgetSeconds(tier, marginS = LITELLM_ROUTER_MARGIN_S) {
+  return litellmChainSeconds(tier) + marginS;
+}
+function clientBudgetSeconds(tier, floorS = 0, marginS = LITELLM_ROUTER_MARGIN_S) {
+  return Math.max(Math.ceil(floorS), minimumClientBudgetSeconds(tier, marginS));
+}
+function assertPositive(value, label) {
+  if (!Number.isFinite(value) || value <= 0) {
+    throw new Error(`litellm timeout budget: ${label} must be a positive number, got ${value}`);
+  }
+}
+
 // src/setup/hindsight.ts
 var HINDSIGHT_DEFAULT_API_PORT = 18888;
 var HINDSIGHT_DEFAULT_MCP_URL = `http://127.0.0.1:${HINDSIGHT_DEFAULT_API_PORT}/mcp/`;
@@ -18361,6 +18402,18 @@ var HINDSIGHT_IMAGE_REPO = "ghcr.io/switchroom/switchroom-hindsight";
 var HINDSIGHT_IMAGE = `${HINDSIGHT_IMAGE_REPO}:latest`;
 var HINDSIGHT_BROKER_SOCK_VOLUME = `auth-broker-${HINDSIGHT_CONSUMER_NAME}-sock`;
 var HINDSIGHT_CREDS_MIRROR_VOLUME = `consumer-creds-${HINDSIGHT_CONSUMER_NAME}`;
+var HINDSIGHT_DEFAULT_RETAIN_MAX_COMPLETION_TOKENS = 16384;
+var HINDSIGHT_RETAIN_MIN_TOKENS_PER_SECOND = 80.6;
+function hindsightRetainLlmTimeoutSeconds(maxCompletionTokens = HINDSIGHT_DEFAULT_RETAIN_MAX_COMPLETION_TOKENS, tokensPerSecond = HINDSIGHT_RETAIN_MIN_TOKENS_PER_SECOND) {
+  if (!(maxCompletionTokens > 0) || !(tokensPerSecond > 0)) {
+    throw new Error(`hindsight retain budget: maxCompletionTokens (${maxCompletionTokens}) and ` + `tokensPerSecond (${tokensPerSecond}) must both be positive`);
+  }
+  return Math.ceil(maxCompletionTokens / tokensPerSecond);
+}
+function hindsightRetainClientTimeoutSeconds() {
+  return clientBudgetSeconds(LITELLM_TIMEOUT_TIERS.retain, hindsightRetainLlmTimeoutSeconds());
+}
+var HINDSIGHT_RETAIN_CLIENT_DEADLINE_S = hindsightRetainClientTimeoutSeconds() + LITELLM_ROUTER_MARGIN_S;
 var HINDSIGHT_HEALTHCHECK_PY = 'import urllib.request,sys; sys.exit(0 if urllib.request.urlopen("http://localhost:8888/health",timeout=4).getcode()==200 else 1)';
 var HINDSIGHT_HEALTHCHECK_CMD = `python3 -c '${HINDSIGHT_HEALTHCHECK_PY}'`;
 var DOCKER_PROBE_TIMEOUT_MS = 60 * 1000;