@nick3/copilot-api 1.4.5 → 1.4.6

package/dist/{server-D_7gI9hx.js → server-Cln-GCLU.js}

@@ -1,8 +1,8 @@
 import { PATHS } from "./paths-DoT4SZ8f.js";
 import { listAccountsFromRegistry } from "./accounts-registry-c7rs5Ed9.js";
-import { HTTPError, accountFromState, cacheModels, copilotBaseUrl, copilotHeaders, forwardError, generateRequestIdFromPayload, getCopilotUsage, getRootSessionId, getUUID, isNullish, parseUserIdMetadata, prepareForCompact, prepareInteractionHeaders, sleep, state } from "./utils-BIK3ym34.js";
-import "./get-copilot-token-BySQCue6.js";
-import { PROVIDER_TYPE_ANTHROPIC, accountsManager, getAliasTargetSet, getAnthropicApiKey, getConfig, getExtraPromptForModel, getModelAliases, getModelAliasesInfo, getModelRefreshIntervalMs, getProviderConfig, getReasoningEffortForModel, getSmallModel, isForceAgentEnabled, isFreeModelLoadBalancingEnabled, isMessageStartInputTokensFallbackEnabled, isMessagesApiEnabled, isResponsesApiContextManagementModel, mergeConfigWithDefaults, shouldCompactUseSmallModel } from "./accounts-manager-BsGuQhKM.js";
+import { HTTPError, accountFromState, cacheModels, copilotBaseUrl, copilotHeaders, forwardError, generateRequestIdFromPayload, getCopilotUsage, getRootSessionId, getUUID, isNullish, parseUserIdMetadata, prepareForCompact, prepareInteractionHeaders, sleep, state } from "./utils-SV0FyqXk.js";
+import "./get-copilot-token-Q4FXtOE9.js";
+import { PROVIDER_TYPE_ANTHROPIC, accountsManager, getAliasTargetSet, getAnthropicApiKey, getConfig, getExtraPromptForModel, getModelAliases, getModelAliasesInfo, getModelRefreshIntervalMs, getProviderConfig, getReasoningEffortForModel, getSmallModel, isAccountAffinityEnabled, isForceAgentEnabled, isMessageStartInputTokensFallbackEnabled, isMessagesApiEnabled, isResponsesApiContextManagementModel, isResponsesApiWebSearchEnabled, mergeConfigWithDefaults, shouldCompactUseSmallModel } from "./accounts-manager-B8Y2Y4ab.js";
 import consola from "consola";
 import fs, { readFile } from "node:fs/promises";
 import * as path$1 from "node:path";
@@ -189,66 +189,69 @@ function getAdminDbUserVersion(db = getAdminDb()) {
 		return 0;
 	}
 }
-function migrateAdminDb(db) {
-	const current = db.query("PRAGMA user_version;").get()?.user_version ?? 0;
-	if (current >= 4) return;
-	if (current < 1) db.run(`
-      CREATE TABLE IF NOT EXISTS request_log (
-        id INTEGER PRIMARY KEY AUTOINCREMENT,
-        request_id TEXT NOT NULL UNIQUE,
+function migrateV1(db) {
+	db.run(`
+    CREATE TABLE IF NOT EXISTS request_log (
+      id INTEGER PRIMARY KEY AUTOINCREMENT,
+      request_id TEXT NOT NULL UNIQUE,
 
-        started_at_ms INTEGER NOT NULL,
-        finished_at_ms INTEGER,
-        duration_ms INTEGER,
-        ttfb_ms INTEGER,
+      started_at_ms INTEGER NOT NULL,
+      finished_at_ms INTEGER,
+      duration_ms INTEGER,
+      ttfb_ms INTEGER,
 
-        method TEXT NOT NULL,
-        path TEXT NOT NULL,
-        upstream_endpoint TEXT,
-        stream INTEGER NOT NULL DEFAULT 0,
+      method TEXT NOT NULL,
+      path TEXT NOT NULL,
+      upstream_endpoint TEXT,
+      stream INTEGER NOT NULL DEFAULT 0,
 
-        account_id TEXT,
-        account_type TEXT,
-        cost_units REAL,
-        client_model TEXT,
-        upstream_model TEXT,
+      account_id TEXT,
+      account_type TEXT,
+      cost_units REAL,
+      client_model TEXT,
+      upstream_model TEXT,
 
-        client_ip TEXT,
-        client_ip_source TEXT,
-        user_agent TEXT,
+      client_ip TEXT,
+      client_ip_source TEXT,
+      user_agent TEXT,
 
-        tokens_input INTEGER,
-        tokens_output INTEGER,
-        tokens_total INTEGER,
-        tokens_cached_input INTEGER,
-        usage_json TEXT,
+      tokens_input INTEGER,
+      tokens_output INTEGER,
+      tokens_total INTEGER,
+      tokens_cached_input INTEGER,
+      usage_json TEXT,
 
-        premium_remaining_before REAL,
-        premium_remaining_after REAL,
-        premium_remaining_diff REAL,
-        premium_unlimited_before INTEGER,
-        premium_unlimited_after INTEGER,
+      premium_remaining_before REAL,
+      premium_remaining_after REAL,
+      premium_remaining_diff REAL,
+      premium_unlimited_before INTEGER,
+      premium_unlimited_after INTEGER,
 
-        http_status INTEGER,
-        error_name TEXT,
-        error_status INTEGER,
-        error_message TEXT,
-        selection_failure_reason TEXT
-      );
+      http_status INTEGER,
+      error_name TEXT,
+      error_status INTEGER,
+      error_message TEXT,
+      selection_failure_reason TEXT
+    );
 
-      CREATE INDEX IF NOT EXISTS idx_request_log_started_at
-        ON request_log(started_at_ms DESC);
-      CREATE INDEX IF NOT EXISTS idx_request_log_account_started_at
-        ON request_log(account_id, started_at_ms DESC);
-      CREATE INDEX IF NOT EXISTS idx_request_log_model_started_at
-        ON request_log(upstream_model, started_at_ms DESC);
-      CREATE INDEX IF NOT EXISTS idx_request_log_endpoint_started_at
-        ON request_log(upstream_endpoint, started_at_ms DESC);
-      CREATE INDEX IF NOT EXISTS idx_request_log_status_started_at
-        ON request_log(http_status, started_at_ms DESC);
+    CREATE INDEX IF NOT EXISTS idx_request_log_started_at
+      ON request_log(started_at_ms DESC);
+    CREATE INDEX IF NOT EXISTS idx_request_log_account_started_at
+      ON request_log(account_id, started_at_ms DESC);
+    CREATE INDEX IF NOT EXISTS idx_request_log_model_started_at
+      ON request_log(upstream_model, started_at_ms DESC);
+    CREATE INDEX IF NOT EXISTS idx_request_log_endpoint_started_at
+      ON request_log(upstream_endpoint, started_at_ms DESC);
+    CREATE INDEX IF NOT EXISTS idx_request_log_status_started_at
+      ON request_log(http_status, started_at_ms DESC);
 
-      PRAGMA user_version = 1;
-    `);
+    PRAGMA user_version = 1;
+  `);
+}
+function migrateAdminDb(db) {
+	const current = db.query("PRAGMA user_version;").get()?.user_version ?? 0;
+	if (current >= 5) return;
+	if (current < 1) migrateV1(db);
 	if (current < 2) db.run(`
       ALTER TABLE request_log ADD COLUMN user_id TEXT;
       ALTER TABLE request_log ADD COLUMN safety_identifier TEXT;
@@ -283,6 +286,12 @@ function migrateAdminDb(db) {
 
       PRAGMA user_version = 4;
     `);
+	if (current < 5) db.run(`
+      ALTER TABLE request_log ADD COLUMN affinity_hit INTEGER;
+      ALTER TABLE request_log ADD COLUMN affinity_cache_key TEXT;
+
+      PRAGMA user_version = 5;
+    `);
 }
 
 //#endregion
@@ -434,13 +443,15 @@ var RequestHistoryStore = class {
         error_name,
         error_status,
         error_message,
-        selection_failure_reason
+        selection_failure_reason,
+        affinity_hit,
+        affinity_cache_key
       ) VALUES (
         ?,?,?,?,?,?,?,?,
         ?,?,?,?,?,?,?,?,
         ?,?,?,?,?,?,?,?,
         ?,?,?,?,?,?,?,?,
-        ?,?,?,?,?
+        ?,?,?,?,?,?,?
       );
     `);
 		this.getByRequestIdStmt = db.query("SELECT * FROM request_log WHERE request_id = ? LIMIT 1;");
@@ -499,7 +510,9 @@ var RequestHistoryStore = class {
 				toDbNull(record.errorName),
 				toDbNull(record.errorStatus),
 				toDbNull(record.errorMessage),
-				toDbNull(record.selectionFailureReason)
+				toDbNull(record.selectionFailureReason),
+				toDbBool(record.affinityHit),
+				toDbNull(record.affinityCacheKey)
 			];
 			this.insertStmt.run(...args);
 		} catch (error) {
@@ -763,7 +776,7 @@ const CONFIG_KEYS = new Set([
 	"auth",
 	"extraPrompts",
 	"smallModel",
-	"freeModelLoadBalancing",
+	"accountAffinity",
 	"apiKey",
 	"anthropicApiKey",
 	"providers",
@@ -776,7 +789,8 @@ const CONFIG_KEYS = new Set([
 	"compactUseSmallModel",
 	"messageStartInputTokensFallback",
 	"modelRefreshIntervalHours",
-	"useMessagesApi"
+	"useMessagesApi",
+	"useResponsesApiWebSearch"
 ]);
 const REASONING_EFFORTS = new Set([
 	"none",
@@ -1124,7 +1138,7 @@ const CONFIG_PATCH_HANDLERS = {
 	auth: applyAuthConfig,
 	extraPrompts: applyExtraPrompts,
 	smallModel: (next, value) => applyOptionalString(next, "smallModel", value),
-	freeModelLoadBalancing: (next, value) => applyOptionalBoolean(next, "freeModelLoadBalancing", value),
+	accountAffinity: (next, value) => applyOptionalBoolean(next, "accountAffinity", value),
 	apiKey: (next, value) => applyOptionalString(next, "apiKey", value),
 	anthropicApiKey: (next, value) => applyOptionalString(next, "anthropicApiKey", value),
 	providers: applyProvidersConfig,
@@ -1137,7 +1151,8 @@ const CONFIG_PATCH_HANDLERS = {
 	compactUseSmallModel: (next, value) => applyOptionalBoolean(next, "compactUseSmallModel", value),
 	messageStartInputTokensFallback: (next, value) => applyOptionalBoolean(next, "messageStartInputTokensFallback", value),
 	modelRefreshIntervalHours: (next, value) => applyOptionalNumber(next, "modelRefreshIntervalHours", value),
-	useMessagesApi: (next, value) => applyOptionalBoolean(next, "useMessagesApi", value)
+	useMessagesApi: (next, value) => applyOptionalBoolean(next, "useMessagesApi", value),
+	useResponsesApiWebSearch: (next, value) => applyOptionalBoolean(next, "useResponsesApiWebSearch", value)
 };
 function applyConfigPatch(base, input) {
 	const next = { ...base };
@@ -1215,7 +1230,7 @@ adminApiRoutes.post("/config", async (c) => {
 	try {
 		await writeConfigFile(result.config);
 		const merged = mergeConfigWithDefaults();
-		accountsManager.setFreeModelLoadBalancingEnabled(isFreeModelLoadBalancingEnabled());
+		accountsManager.setAccountAffinityEnabled(isAccountAffinityEnabled());
 		accountsManager.setModelsRefreshIntervalMs(getModelRefreshIntervalMs());
 		return c.json({
 			...merged,
@@ -2520,7 +2535,10 @@ async function handleCompletion$1(c) {
 	const selection = await accountsManager.selectAccountForRequest([{
 		modelId: clientModel,
 		endpoint: CHAT_COMPLETIONS_ENDPOINT$1
-	}]);
+	}], {
+		promptCacheKey: normalizedPromptCacheKey,
+		safetyIdentifier: normalizedSafetyIdentifier
+	});
 	if (!selection.ok) {
 		recordSelectionFailure$2(store, {
 			request,
@@ -2534,6 +2552,8 @@ async function handleCompletion$1(c) {
 		});
 	}
 	const { account, selectedModel } = selection;
+	request.affinityHit = selection.affinityHit;
+	request.affinityCacheKey = selection.affinityCacheKey;
 	const upstreamPayload = {
 		...payload,
 		model: selectedModel.id
@@ -2605,6 +2625,8 @@ function insertRequestLog$2(store, request, record) {
 		promptCacheKey: request.promptCacheKey,
 		initiator: request.initiator,
 		upstreamRequestId: request.upstreamRequestId,
+		affinityHit: request.affinityHit,
+		affinityCacheKey: request.affinityCacheKey,
 		...record
 	});
 }
@@ -2657,6 +2679,7 @@ async function handleStreamingRequest(params) {
 			upstreamRequestId: request.upstreamRequestId,
 			sessionId: request.upstreamSessionId
 		});
+		selection.confirmAffinity?.();
 	} catch (error) {
 		return handleUpstreamCreateError$1({
 			store,
@@ -2844,6 +2867,7 @@ async function handleNonStreamingRequest(params) {
 			upstreamRequestId: request.upstreamRequestId,
 			sessionId: request.upstreamSessionId
 		});
+		selection.confirmAffinity?.();
 		finishedAtMs = Date.now();
 		if (!isNonStreaming$1(response)) {
 			logger$6.debug("Unexpected streaming response");
@@ -3239,6 +3263,21 @@ const maybeBlockOriginalModelName = (context) => {
 		}
 	});
 };
+const compactSystemPromptStart = "You are a helpful AI assistant tasked with summarizing conversations";
+const compactUserMessageStart = "CRITICAL: Respond with TEXT ONLY. Do NOT call any tools.";
+const hasCompactUserMessage = (messages) => {
+	const lastMsg = messages.at(-1);
+	if (!lastMsg || lastMsg.role !== "user") return false;
+	if (typeof lastMsg.content === "string") return lastMsg.content.startsWith(compactUserMessageStart);
+	if (!Array.isArray(lastMsg.content)) return false;
+	return lastMsg.content.some((block) => block.type === "text" && typeof block.text === "string" && block.text.startsWith(compactUserMessageStart));
+};
+const isCompactRequest = (anthropicPayload) => {
+	const system = anthropicPayload.system;
+	if (typeof system === "string" && system.startsWith(compactSystemPromptStart)) return true;
+	if (Array.isArray(system) && system.some((msg) => typeof msg.text === "string" && msg.text.startsWith(compactSystemPromptStart))) return true;
+	return hasCompactUserMessage(anthropicPayload.messages);
+};
 
 //#endregion
 //#region src/routes/messages/non-stream-translation.ts
@@ -4482,7 +4521,15 @@ const containsVisionContent = (value) => {
 
 //#endregion
 //#region src/services/copilot/create-messages.ts
+const isAgentMessage = (msg) => {
+	if (msg.role === "assistant") return true;
+	if (Array.isArray(msg.content)) {
+		if (msg.content.every((block) => block.type === "tool_result") && msg.content.length > 0) return true;
+	}
+	return false;
+};
 const getMessagesInitiator = (payload) => {
+	if (isForceAgentEnabled()) return payload.messages.some((msg) => isAgentMessage(msg)) ? "agent" : "user";
 	const lastMessage = payload.messages.at(-1);
 	if (!lastMessage || lastMessage.role !== "user") return "agent";
 	if (!Array.isArray(lastMessage.content)) return "user";
@@ -4839,7 +4886,6 @@ const logger$5 = createHandlerLogger("messages-handler");
 const CHAT_COMPLETIONS_ENDPOINT = "/chat/completions";
 const RESPONSES_ENDPOINT$1 = "/responses";
 const MESSAGES_ENDPOINT = "/v1/messages";
-const compactSystemPromptStart = "You are a helpful AI assistant tasked with summarizing conversations";
 async function handleCompletion(c) {
 	await checkRateLimit(state);
 	const store = getRequestHistoryStore();
@@ -4907,7 +4953,11 @@ async function handleCompletion(c) {
 		modelId: endpointModel?.id ?? openAIPayload.model,
 		endpoint: CHAT_COMPLETIONS_ENDPOINT
 	});
-	const selection = await accountsManager.selectAccountForRequest(candidates);
+	const selection = await accountsManager.selectAccountForRequest(candidates, {
+		promptCacheKey: normalizedPromptCacheKey,
+		sessionId,
+		safetyIdentifier: normalizedSafetyIdentifier
+	});
 	if (!selection.ok) return handleSelectionFailure({
 		c,
 		store,
@@ -4952,7 +5002,10 @@ async function handleCompletion(c) {
 		costUnits,
 		upstreamRequestId,
 		premiumRemainingBefore,
-		premiumUnlimitedBefore
+		premiumUnlimitedBefore,
+		confirmAffinity: selection.confirmAffinity,
+		affinityHit: selection.affinityHit,
+		affinityCacheKey: selection.affinityCacheKey
 	};
 	if (endpoint === MESSAGES_ENDPOINT) return await handleWithMessagesApi({
 		c,
@@ -5002,6 +5055,7 @@ const handleWithChatCompletions = async (params) => {
 			sessionId,
 			isCompact
 		});
+		instr.confirmAffinity?.();
 	} catch (error) {
 		return await handleChatCompletionsCreateError({
 			error,
@@ -5050,6 +5104,7 @@ const handleWithResponsesApi = async (params) => {
 			sessionId,
 			isCompact
 		}, ctx);
+		instr.confirmAffinity?.();
 	} catch (error) {
 		return await handleResponsesCreateError({
 			error,
@@ -5095,6 +5150,8 @@ function insertRequestLog$1(instr, record) {
 		promptCacheKey: instr.promptCacheKey,
 		initiator: instr.initiator,
 		upstreamRequestId: instr.upstreamRequestId,
+		affinityHit: instr.affinityHit,
+		affinityCacheKey: instr.affinityCacheKey,
 		clientModel,
 		upstreamEndpoint,
 		accountId: account.id,
@@ -5526,6 +5583,7 @@ const handleWithMessagesApi = async (params) => {
 			sessionId,
 			isCompact
 		});
+		instr.confirmAffinity?.();
 	} catch (error) {
 		return await handleMessagesCreateError({
 			error,
@@ -5555,12 +5613,6 @@ const getAnthropicEffortForModel = (model) => {
 	if (reasoningEffort === "none" || reasoningEffort === "minimal") return "low";
 	return reasoningEffort;
 };
-const isCompactRequest = (anthropicPayload) => {
-	const system = anthropicPayload.system;
-	if (typeof system === "string") return system.startsWith(compactSystemPromptStart);
-	if (!Array.isArray(system)) return false;
-	return system.some((msg) => typeof msg.text === "string" && msg.text.startsWith(compactSystemPromptStart));
-};
 
 //#endregion
 //#region src/routes/messages/route.ts
@@ -5896,7 +5948,7 @@ const handleResponses = async (c) => {
 	const payload = await c.req.json();
 	const clientModel = payload.model;
 	logger$1.debug("Responses request payload:", JSON.stringify(payload));
-	removeWebSearchTool(payload);
+	if (!isResponsesApiWebSearchEnabled()) removeWebSearchTool(payload);
 	compactInputByLatestCompaction(payload);
 	const streamRequested = Boolean(payload.stream);
 	const { initiator: initialInitiator } = getResponsesRequestOptions(payload);
@@ -5923,7 +5975,10 @@ const handleResponses = async (c) => {
 	const selection = await accountsManager.selectAccountForRequest([{
 		modelId: clientModel,
 		endpoint: RESPONSES_ENDPOINT
-	}]);
+	}], {
+		promptCacheKey: normalizedPromptCacheKey,
+		safetyIdentifier: normalizedSafetyIdentifier
+	});
 	if (!selection.ok) {
 		recordSelectionFailure(store, {
 			request,
@@ -5934,6 +5989,8 @@ const handleResponses = async (c) => {
 		return selectionFailureResponse(c, { reason: selection.reason });
 	}
 	const { account, selectedModel } = selection;
+	request.affinityHit = selection.affinityHit;
+	request.affinityCacheKey = selection.affinityCacheKey;
 	const upstreamPayload = {
 		...payload,
 		model: selectedModel.id
@@ -6009,6 +6066,8 @@ function insertRequestLog(store, request, record) {
 		promptCacheKey: request.promptCacheKey,
 		initiator: request.initiator,
 		upstreamRequestId: request.upstreamRequestId,
+		affinityHit: request.affinityHit,
+		affinityCacheKey: request.affinityCacheKey,
 		...record
 	});
 }
@@ -6064,6 +6123,7 @@ async function handleStreamingResponses(params) {
 			upstreamRequestId: request.upstreamRequestId,
 			sessionId: request.upstreamSessionId
 		}, accountCtx);
+		selection.confirmAffinity?.();
 	} catch (error) {
 		return handleUpstreamCreateError({
 			store,
@@ -6250,6 +6310,7 @@ async function handleNonStreamingResponses(params) {
 			upstreamRequestId: request.upstreamRequestId,
 			sessionId: request.upstreamSessionId
 		}, accountCtx);
+		selection.confirmAffinity?.();
 		finishedAtMs = Date.now();
 		const streamResponse = handleUnexpectedResponsesStream(c, response);
 		if (streamResponse) return streamResponse;
@@ -6434,4 +6495,4 @@ server.route("/:provider/v1/models", providerModelRoutes);
 
 //#endregion
 export { server };
-//# sourceMappingURL=server-D_7gI9hx.js.map
+//# sourceMappingURL=server-Cln-GCLU.js.map