@copilotkit/aimock 1.45.1 → 1.46.0

package/dist/server.js

@@ -1,9 +1,12 @@
+import { isAnthropicModelRequest, missingRequiredHeader } from "./response/profiles.js";
+import { renderError } from "./response/errors.js";
 import { DEFAULT_TEST_ID } from "./constants.js";
 import { build, msg, quote } from "./message-text.js";
 import { markAuthenticatedRequest, resolveInboundAuth, validateRequestApiKey, writeApiKeyHttpRejection, writeApiKeyUpgradeRejection } from "./api-key-auth.js";
-import { RequestBodyTooLargeError, buildContentWithToolCallsChunks, buildContentWithToolCallsCompletion, buildOpenAIContentFilterChunks, buildOpenAIContentFilterCompletion, buildOpenAIReasoningChunks, buildOpenAIReasoningCompletion, buildOpenAIRefusalChunks, buildOpenAIRefusalCompletion, buildTextChunks, buildTextCompletion, buildToolCallChunks, buildToolCallCompletion, buildUsageChunk, createServerFixtureList, describeMatch, extractOverrides, fixtureToolCallErrorCode, flattenHeaders, getContext, getTestId, googleFixtureToolCallErrorDetails, isAudioResponse, isCombinedFixtureResponse, isEmbeddingResponse, isErrorResponse, isExtendedResponsesToolsList, isFixtureToolCallError, isImageResponse, isJSONResponse, isJsonObject, isResponseFactory, isServerOwnedFixtures, isTextResponse, isToolCallResponse, isTranscriptionResponse, isVideoResponse, markMintedRequestId, prepareOpenAIChatMisbehavior, readBody, readBodyBufferBounded, requireEmittedFunctionToolCalls, requireServedFunctionToolCalls, resolveFixtureBlockOutcome, resolveOpenAIChatMisbehaviorUsage, resolveReasoningForModel, resolveRequestId, resolveResponse, resolveStrictMode, resolveTestId, resolveUsage, runWithToolArgumentsScope, serializeErrorResponse, strictNoMatchLogLine, strictNoMatchMessage, strictOverrideField, validateToolsField, withoutResponsesToolKeys, wouldProxyMiss } from "./helpers.js";
+import { parsePolicy, resolvePolicy, resolveRequestDecisions, strictOverrideField } from "./response/policy.js";
+import { RequestBodyTooLargeError, buildContentWithToolCallsChunks, buildContentWithToolCallsCompletion, buildOpenAIContentFilterChunks, buildOpenAIContentFilterCompletion, buildOpenAIReasoningChunks, buildOpenAIReasoningCompletion, buildOpenAIRefusalChunks, buildOpenAIRefusalCompletion, buildTextChunks, buildTextCompletion, buildToolCallChunks, buildToolCallCompletion, buildUsageChunk, createServerFixtureList, describeMatch, extractOverrides, fixtureToolCallErrorCode, flattenHeaders, getContext, getTestId, googleFixtureToolCallErrorDetails, isAudioResponse, isCombinedFixtureResponse, isEmbeddingResponse, isErrorResponse, isExtendedResponsesToolsList, isFixtureToolCallError, isImageResponse, isJSONResponse, isJsonObject, isResponseFactory, isServerOwnedFixtures, isTextResponse, isToolCallResponse, isTranscriptionResponse, isVideoResponse, markMintedRequestId, prepareOpenAIChatMisbehavior, readBody, readBodyBufferBounded, requireEmittedFunctionToolCalls, requireServedFunctionToolCalls, resolveFixtureBlockOutcome, resolveOpenAIChatMisbehaviorUsage, resolveReasoningForModel, resolveRequestId, resolveResponse, resolveTestId, resolveUsage, runWithToolArgumentsScope, serializeErrorResponse, strictNoMatchLogLine, strictNoMatchMessage, validateToolsField, withoutResponsesToolKeys, wouldProxyMiss } from "./helpers.js";
 import { normalizeLiveOptions } from "./live-fixture.js";
-import { writeErrorResponse, writeSSEStream } from "./sse-writer.js";
+import { calculateTotalDelay, delay, writeErrorResponse, writeSSEStream } from "./sse-writer.js";
 import { CHAOS_FIELDS, CHAOS_FIELD_NAMES, applyChaosAction, awaitChaosLatency, describeUnwritableReason, evaluateChaos, isChaosScope, parseChaosField, resetChaosWarnings, resolveChaosConfig, responseGoneReason } from "./chaos.js";
 import { MISBEHAVIOR_CATALOG, WIRE_SUPPORT, getFixtureMisbehaviorPosition, parseMisbehavior, parseMisbehaviorHeader, planMisbehavior, recordMisbehaviorOutcome, resolveMisbehaviorSeed, resolveMisbehaviorShortCircuit, setFixtureMisbehaviorPosition } from "./misbehavior.js";
 import { FixtureLoadError, INJECTED_STATUS_RANGE, MisbehaviorConfigError, claimOneShotError, clearFixtureQueue, enableHeldFixtureMisbehavior, entryToFixture, isInjectableStatus, isOneShotError, markFixtureResponsesToolsExtended, queueOneShotError, releaseOneShotError, responsesToolsExtendedView, responsesToolsViewSource, validateFixtures } from "./fixture-loader.js";
@@ -12,10 +15,15 @@ import { proxyAndRecord } from "./recorder.js";
 import { Logger } from "./logger.js";
 import { Journal } from "./journal.js";
 import { createInterruptionSignal } from "./interruption.js";
+import { collect, idsFor } from "./response/catalog/collect.js";
+import { anthropicPage } from "./response/catalog/anthropic-page.js";
+import { geminiPage } from "./response/catalog/gemini-page.js";
+import { coreModelIds, renderAnthropicModel, renderAnthropicModels, renderGeminiModel, renderGeminiModels, renderOllamaTags, renderOpenAIModel, renderOpenAIModels } from "./response/catalog/render.js";
 import { buildOpenRouterCandidates, handleOpenRouterCredits, handleOpenRouterKey, handleOpenRouterModels, isOpenRouterPath, resolveOpenRouterShaping, serializeOpenRouterError, shapeOpenRouterChunks, shapeOpenRouterCompletion } from "./openrouter-chat.js";
 import { handleResponses } from "./responses.js";
 import { handleMessages } from "./messages.js";
 import { handleGemini } from "./gemini.js";
+import { handleCountTokens } from "./count-tokens.js";
 import { handleGeminiEmbedContent } from "./gemini-embeddings.js";
 import { handleBedrock, handleBedrockStream } from "./bedrock.js";
 import { handleConverse, handleConverseStream } from "./bedrock-converse.js";
@@ -47,7 +55,7 @@ import { handleLiveSession } from "./ws-live.js";
 import { handleWebSocketResponses } from "./ws-responses.js";
 import { handleWebSocketRealtime } from "./ws-realtime.js";
 import { handleWebSocketGeminiLive } from "./ws-gemini-live.js";
-import { AZURE_DEPLOYMENT_RE, BATCHES_CANCEL_RE, BATCHES_ID_RE, BATCHES_PATH, BEDROCK_CONVERSE_RE, BEDROCK_CONVERSE_STREAM_RE, BEDROCK_INVOKE_RE, BEDROCK_STREAM_RE, BYTEPLUS_VIDEO_STATUS_RE, BYTEPLUS_VIDEO_SUBMIT_RE, COHERE_CHAT_PATH, COHERE_EMBED_PATH, COMPLETIONS_PATH, CONTROL_PREFIX, ELEVENLABS_MUSIC_RE, ELEVENLABS_SOUND_GENERATION_PATH, ELEVENLABS_TTS_RE, ELEVENLABS_VOICE_CREATE_PATH, ELEVENLABS_VOICE_DESIGN_PATH, ELEVENLABS_VOICE_RE, EMBEDDINGS_PATH, FAL_QUEUE_REQUESTS_RE, FAL_QUEUE_SUBMIT_RE, FAL_ROUTE_RE, FAL_RUN_RE, FILES_CONTENT_RE, FILES_ID_RE, FILES_PATH, FINE_TUNING_CANCEL_RE, FINE_TUNING_EVENTS_RE, FINE_TUNING_ID_RE, FINE_TUNING_JOBS_PATH, GEMINI_EMBED_RE, GEMINI_INTERACTIONS_PATH, GEMINI_LIVE_PATH, GEMINI_PATH_RE, GEMINI_PREDICT_RE, GROK_VIDEO_STATUS_RE, GROK_VIDEO_SUBMIT_PATH, HEALTH_PATH, IMAGES_EDIT_PATH, IMAGES_PATH, IMAGES_VARIATIONS_PATH, LIVE_PATH, MESSAGES_PATH, METRICS_PATH, MODELS_PATH, MODERATIONS_PATH, OLLAMA_CHAT_PATH, OLLAMA_EMBEDDINGS_PATH, OLLAMA_EMBED_PATH, OLLAMA_GENERATE_PATH, OLLAMA_TAGS_PATH, OPENROUTER_CREDITS_PATH, OPENROUTER_KEY_PATH, OPENROUTER_MODELS_PATH, OPENROUTER_VIDEOS_PATH, OPENROUTER_VIDEO_CONTENT_RE, OPENROUTER_VIDEO_MODELS_PATH, OPENROUTER_VIDEO_STATUS_RE, READY_PATH, REALTIME_PATH, REQUESTS_PATH, RERANK_PATH, RESPONSES_PATH, SEARCH_PATH, SPEECH_PATH, TRANSCRIPTIONS_PATH, TRANSLATIONS_PATH, VECTOR_STORES_BATCH_CANCEL_RE, VECTOR_STORES_BATCH_FILES_RE, VECTOR_STORES_BATCH_RE, VECTOR_STORES_FILES_RE, VECTOR_STORES_FILE_BATCHES_RE, VECTOR_STORES_FILE_RE, VECTOR_STORES_ID_RE, VECTOR_STORES_PATH, VECTOR_STORES_SEARCH_RE, VEO_OPERATION_RE, VEO_PREDICT_LRO_RE, VERTEX_AI_RE, VIDEOS_PATH, matchRouteDefinition } from "./route-registry.js";
+import { AZURE_DEPLOYMENT_RE, BATCHES_CANCEL_RE, BATCHES_ID_RE, BATCHES_PATH, BEDROCK_CONVERSE_RE, BEDROCK_CONVERSE_STREAM_RE, BEDROCK_INVOKE_RE, BEDROCK_STREAM_RE, BYTEPLUS_VIDEO_STATUS_RE, BYTEPLUS_VIDEO_SUBMIT_RE, COHERE_CHAT_PATH, COHERE_EMBED_PATH, COMPLETIONS_PATH, CONTROL_PREFIX, ELEVENLABS_MUSIC_RE, ELEVENLABS_SOUND_GENERATION_PATH, ELEVENLABS_TTS_RE, ELEVENLABS_VOICE_CREATE_PATH, ELEVENLABS_VOICE_DESIGN_PATH, ELEVENLABS_VOICE_RE, EMBEDDINGS_PATH, FAL_QUEUE_REQUESTS_RE, FAL_QUEUE_SUBMIT_RE, FAL_ROUTE_RE, FAL_RUN_RE, FILES_CONTENT_RE, FILES_ID_RE, FILES_PATH, FINE_TUNING_CANCEL_RE, FINE_TUNING_EVENTS_RE, FINE_TUNING_ID_RE, FINE_TUNING_JOBS_PATH, GEMINI_COUNT_RE, GEMINI_EMBED_RE, GEMINI_INTERACTIONS_PATH, GEMINI_LIVE_PATH, GEMINI_MODELS_PATH, GEMINI_MODEL_DETAIL_RE, GEMINI_PATH_RE, GEMINI_PREDICT_RE, GROK_VIDEO_STATUS_RE, GROK_VIDEO_SUBMIT_PATH, HEALTH_PATH, IMAGES_EDIT_PATH, IMAGES_PATH, IMAGES_VARIATIONS_PATH, LIVE_PATH, MESSAGES_COUNT_PATH, MESSAGES_PATH, METRICS_PATH, MODELS_DETAIL_RE, MODELS_PATH, MODERATIONS_PATH, OLLAMA_CHAT_PATH, OLLAMA_EMBEDDINGS_PATH, OLLAMA_EMBED_PATH, OLLAMA_GENERATE_PATH, OLLAMA_TAGS_PATH, OPENROUTER_CREDITS_PATH, OPENROUTER_KEY_PATH, OPENROUTER_MODELS_PATH, OPENROUTER_VIDEOS_PATH, OPENROUTER_VIDEO_CONTENT_RE, OPENROUTER_VIDEO_MODELS_PATH, OPENROUTER_VIDEO_STATUS_RE, READY_PATH, REALTIME_PATH, REQUESTS_PATH, RERANK_PATH, RESPONSES_PATH, SEARCH_PATH, SPEECH_PATH, TRANSCRIPTIONS_PATH, TRANSLATIONS_PATH, VECTOR_STORES_BATCH_CANCEL_RE, VECTOR_STORES_BATCH_FILES_RE, VECTOR_STORES_BATCH_RE, VECTOR_STORES_FILES_RE, VECTOR_STORES_FILE_BATCHES_RE, VECTOR_STORES_FILE_RE, VECTOR_STORES_ID_RE, VECTOR_STORES_PATH, VECTOR_STORES_SEARCH_RE, VEO_OPERATION_RE, VEO_PREDICT_LRO_RE, VERTEX_AI_RE, VIDEOS_PATH, matchRouteDefinition } from "./route-registry.js";
 import { CATALOG_ROUTES, buildOpenApiDocument } from "./openapi.js";
 import { DESTROYED_STATUS_LABEL, createMetricsRegistry, normalizePathLabel } from "./metrics.js";
 import { MCP_FAKES_DEFAULT_MOUNT, McpFakeStore, McpFakesAddError, blockIdOf } from "./mcp-fakes.js";
@@ -87,13 +95,6 @@ function normalizeCompatPath(pathname, logger) {
 	}
 	return pathname;
 }
-const DEFAULT_MODELS = [
-	"gpt-4",
-	"gpt-4o",
-	"claude-3-5-sonnet-20241022",
-	"gemini-2.0-flash",
-	"text-embedding-3-small"
-];
 const CORS_HEADERS = {
 	"Access-Control-Allow-Origin": "*",
 	"Access-Control-Allow-Methods": "GET, POST, PUT, DELETE, OPTIONS",
@@ -110,10 +111,10 @@ function handleOptions(res) {
 }
 function handleNotFound(res, message) {
 	setCorsHeaders(res);
-	writeErrorResponse(res, 404, JSON.stringify({ error: {
+	writeErrorResponse(res, 404, renderError("openai", "route-failure", {
 		message,
 		type: "not_found"
-	} }));
+	}).body);
 }
 /** The complete `GET /__aimock/fixtures` query-param vocabulary; anything else 400s. */
 const FIXTURES_PARAMS = new Set(["include"]);
@@ -924,10 +925,10 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 				fixture: null
 			}
 		});
-		writeErrorResponse(res, status, openRouter ? serializeOpenRouterError(status, `Request body read failed: ${msg}`) : JSON.stringify({ error: {
+		writeErrorResponse(res, status, openRouter ? serializeOpenRouterError(status, `Request body read failed: ${msg}`) : renderError("openai", "transport-failure", {
 			message: `Request body read failed: ${msg}`,
 			type: clientFault ? "invalid_request_error" : "server_error"
-		} }));
+		}).body);
 		return;
 	}
 	let body;
@@ -945,12 +946,13 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 				fixture: null
 			}
 		});
-		writeErrorResponse(res, 400, openRouter ? serializeOpenRouterError(400, `Malformed JSON: ${detail}`) : JSON.stringify({ error: {
+		writeErrorResponse(res, 400, openRouter ? serializeOpenRouterError(400, `Malformed JSON: ${detail}`) : renderError("openai", "validation", {
 			message: `Malformed JSON: ${detail}`,
 			type: "invalid_request_error",
 			param: null,
-			code: "invalid_json"
-		} }));
+			code: "invalid_json",
+			includeParam: true
+		}).body);
 		return;
 	}
 	if (!isJsonObject(body)) {
@@ -964,10 +966,10 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 				fixture: null
 			}
 		});
-		writeErrorResponse(res, 400, openRouter ? serializeOpenRouterError(400, "Request body must be a JSON object") : JSON.stringify({ error: {
+		writeErrorResponse(res, 400, openRouter ? serializeOpenRouterError(400, "Request body must be a JSON object") : renderError("openai", "validation", {
 			message: "Request body must be a JSON object",
 			type: "invalid_request_error"
-		} }));
+		}).body);
 		return;
 	}
 	if (modelFallback && !body.model) body.model = modelFallback;
@@ -982,12 +984,13 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 				fixture: null
 			}
 		});
-		writeErrorResponse(res, 400, openRouter ? serializeOpenRouterError(400, "Missing required parameter: 'messages'") : JSON.stringify({ error: {
+		writeErrorResponse(res, 400, openRouter ? serializeOpenRouterError(400, "Missing required parameter: 'messages'") : renderError("openai", "validation", {
 			message: "Missing required parameter: 'messages'",
 			type: "invalid_request_error",
 			param: null,
-			code: null
-		} }));
+			code: null,
+			includeParam: true
+		}).body);
 		return;
 	}
 	const nullMessageIndex = body.messages.findIndex((message) => message === null);
@@ -1003,12 +1006,13 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 				fixture: null
 			}
 		});
-		writeErrorResponse(res, 400, openRouter ? serializeOpenRouterError(400, message) : JSON.stringify({ error: {
+		writeErrorResponse(res, 400, openRouter ? serializeOpenRouterError(400, message) : renderError("openai", "validation", {
 			message,
 			type: "invalid_request_error",
 			param: null,
-			code: null
-		} }));
+			code: null,
+			includeParam: true
+		}).body);
 		return;
 	}
 	if (new URL(req.url ?? COMPLETIONS_PATH, "http://localhost").pathname === COMPLETIONS_PATH) {
@@ -1028,10 +1032,10 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 					fixture: null
 				}
 			});
-			writeErrorResponse(res, 400, JSON.stringify({ error: {
+			writeErrorResponse(res, 400, renderError("openai", "validation", {
 				message: toolsError,
 				type: "invalid_request_error"
-			} }));
+			}).body);
 			return;
 		}
 	}
@@ -1131,8 +1135,9 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 		headers: flatHeaders,
 		body
 	};
-	const effectiveStrict = resolveStrictMode(defaults.strict, req.headers);
-	const missWouldProxy = wouldProxyMiss(effectiveStrict, defaults.record, providerKey);
+	const missDecisions = resolveRequestDecisions(defaults.strict, req.headers);
+	const effectiveStrict = missDecisions.fixtureMiss;
+	const missWouldProxy = wouldProxyMiss(!missDecisions.upstream, defaults.record, providerKey);
 	const noFixtureSource = missWouldProxy ? "proxy" : "internal";
 	const recordBypass = (reason) => {
 		const entry = getRequestEntry();
@@ -1179,12 +1184,13 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 					...strictOverrideField(defaults.strict, req.headers)
 				}
 			});
-			writeErrorResponse(res, strictStatus, openRouter ? serializeOpenRouterError(strictStatus, strictMessage) : JSON.stringify({ error: {
+			writeErrorResponse(res, strictStatus, openRouter ? serializeOpenRouterError(strictStatus, strictMessage) : renderError("openai", "strict-miss", {
 				message: strictMessage,
 				type: "invalid_request_error",
 				param: null,
-				code: "no_fixture_match"
-			} }));
+				code: "no_fixture_match",
+				includeParam: true
+			}).body);
 			return;
 		}
 		if (defaults.record && providerKey) {
@@ -1254,12 +1260,13 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 				...strictOverrideField(defaults.strict, req.headers)
 			}
 		});
-		writeErrorResponse(res, 404, openRouter ? serializeOpenRouterError(404, "No fixture matched") : JSON.stringify({ error: {
+		writeErrorResponse(res, 404, openRouter ? serializeOpenRouterError(404, "No fixture matched") : renderError("openai", "fixture-miss", {
 			message: "No fixture matched",
 			type: "invalid_request_error",
 			param: null,
-			code: "no_fixture_match"
-		} }));
+			code: "no_fixture_match",
+			includeParam: true
+		}).body);
 		return;
 	}
 	const resolvedResponse = preResolvedResponse ?? await resolveResponse(fixture, body);
@@ -1290,12 +1297,13 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 			testId
 		});
 		if (!evaluation.summary?.evaluations.some((row) => row.outcome === "error")) defaults.logger.error(`${evaluation.code}: ${evaluation.message}`);
-		writeErrorResponse(res, evaluation.status, openRouter ? serializeOpenRouterError(evaluation.status, `${evaluation.code}: ${evaluation.message}`) : JSON.stringify({ error: {
+		writeErrorResponse(res, evaluation.status, openRouter ? serializeOpenRouterError(evaluation.status, `${evaluation.code}: ${evaluation.message}`) : renderError("openai", "misbehavior", {
 			message: evaluation.message,
 			type: "invalid_request_error",
 			param: null,
-			code: evaluation.code
-		} }));
+			code: evaluation.code,
+			includeParam: true
+		}).body);
 		return;
 	}
 	const appliedPlan = evaluation.kind === "applied" ? prepareOpenAIChatMisbehavior(evaluation) : void 0;
@@ -1309,22 +1317,28 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 			testId
 		});
 	};
-	const appliedReasoning = appliedPlan && appliedPlan.reasoning === void 0 && (isTextResponse(response) || isToolCallResponse(response) || isCombinedFixtureResponse(response)) ? resolveReasoningForModel(response.reasoning, responseModel, resolveStrictMode(defaults.strict, req.headers), defaults.logger) : void 0;
+	const appliedReasoning = appliedPlan && appliedPlan.reasoning === void 0 && (isTextResponse(response) || isToolCallResponse(response) || isCombinedFixtureResponse(response)) ? resolveReasoningForModel(response.reasoning, responseModel, resolveRequestDecisions(defaults.strict, req.headers).reasoningSuppression, defaults.logger) : void 0;
 	const appliedUsage = appliedPlan ? resolveOpenAIChatMisbehaviorUsage(appliedPlan, body, appliedPlan.reasoning === void 0 ? appliedReasoning !== void 0 : openRouter) : void 0;
 	const latency = fixture.latency ?? defaults.latency;
 	const chunkSize = Math.max(1, fixture.chunkSize ?? defaults.chunkSize);
 	const emitStreamingUsage = body.stream === true && body.stream_options?.include_usage === true || openRouter;
 	const streamingPromptText = body.messages.map((m) => typeof m.content === "string" ? m.content : Array.isArray(m.content) ? m.content.map((p) => p.text ?? "").join("") : "").join("");
 	const resolveStreamingUsageTokens = (overrides, completionText) => appliedUsage ?? resolveUsage(overrides, streamingPromptText, completionText);
-	const resolveRecordedTimings = (chunks) => {
+	const resolveRecordedTimings = (chunkCount) => {
 		const timings = fixture.recordedTimings;
 		if (!appliedPlan?.duplicateId || !timings) return timings;
 		const gaps = timings.interChunkDelaysMs;
 		return {
 			...timings,
-			interChunkDelaysMs: Array.from({ length: Math.max(gaps.length, chunks.length - 1) }, (_, index) => gaps[index] ?? gaps.at(-1) ?? 0)
+			interChunkDelaysMs: Array.from({ length: Math.max(gaps.length, chunkCount - 1) }, (_, index) => gaps[index] ?? gaps.at(-1) ?? 0)
 		};
 	};
+	const awaitNonStreamingTiming = (contentLength) => {
+		if (fixture.nonStreamingTiming !== true) return Promise.resolve();
+		if (!fixture.streamingProfile && !fixture.recordedTimings) return Promise.resolve();
+		const chunkCount = Math.max(1, Math.ceil(contentLength / chunkSize));
+		return delay(calculateTotalDelay(contentLength, chunkSize, fixture.streamingProfile, resolveRecordedTimings(chunkCount), fixture.replaySpeed ?? defaults.replaySpeed));
+	};
 	const shapeORCompletion = (completion, overrides) => {
 		if (appliedUsage) completion.usage = appliedUsage;
 		if (!openRouter) return completion;
@@ -1363,10 +1377,10 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 				fixture
 			}
 		}));
-		writeErrorResponse(res, 422, openRouter ? serializeOpenRouterError(422, "Audio responses are not supported on the chat completions endpoint. Use Gemini generateContent or a dedicated audio endpoint.") : JSON.stringify({ error: {
+		writeErrorResponse(res, 422, openRouter ? serializeOpenRouterError(422, "Audio responses are not supported on the chat completions endpoint. Use Gemini generateContent or a dedicated audio endpoint.") : renderError("openai", "validation", {
 			message: "Audio responses are not supported on the chat completions endpoint. Use Gemini generateContent or a dedicated audio endpoint.",
 			type: "invalid_request_error"
-		} }));
+		}).body);
 		return;
 	}
 	if (appliedPlan && isTextResponse(response) && (appliedPlan.stop === "refusal" || appliedPlan.stop === "content_filter" || appliedPlan.reasoning !== void 0)) {
@@ -1384,6 +1398,7 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 		recordOutcome(entry);
 		if (body.stream !== true) {
 			const completion = appliedPlan.stop === "refusal" ? buildOpenAIRefusalCompletion(appliedPlan.refusal ?? "", responseModel, overrides, body.messages) : appliedPlan.stop === "content_filter" ? buildOpenAIContentFilterCompletion(responseModel, overrides, body.messages) : buildOpenAIReasoningCompletion(appliedPlan.reasoning ?? "", responseModel, openRouter, overrides, body.messages);
+			await awaitNonStreamingTiming((appliedPlan.stop === "refusal" ? appliedPlan.refusal ?? "" : appliedPlan.stop === "content_filter" ? "" : appliedPlan.reasoning ?? "").length);
 			res.writeHead(200, { "Content-Type": "application/json" });
 			res.end(JSON.stringify(shapeORCompletion(completion, overrides)));
 		} else {
@@ -1397,7 +1412,7 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 				signal: interruption?.signal,
 				onChunkSent: interruption?.tick,
 				usageChunk,
-				recordedTimings: resolveRecordedTimings(chunks),
+				recordedTimings: resolveRecordedTimings(chunks.length),
 				replaySpeed: fixture.replaySpeed ?? defaults.replaySpeed,
 				openRouterProcessing
 			})) {
@@ -1412,7 +1427,7 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 	if (isCombinedFixtureResponse(response)) {
 		if (response.webSearches?.length) defaults.logger.warn("webSearches in fixture response are not supported for Chat Completions API — ignoring");
 		const overrides = extractOverrides(response);
-		const effectiveStrict = resolveStrictMode(defaults.strict, req.headers);
+		const effectiveStrict = resolveRequestDecisions(defaults.strict, req.headers).reasoningSuppression;
 		const effReasoning = appliedPlan ? appliedReasoning : resolveReasoningForModel(response.reasoning, responseModel, effectiveStrict, defaults.logger);
 		const streaming = body.stream === true;
 		const blockOutcome = response.blocks && response.blocks.length > 0 ? resolveFixtureBlockOutcome(response.blocks) : void 0;
@@ -1428,7 +1443,9 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 		});
 		recordOutcome(journalEntry);
 		const toolCalls = requireEmittedFunctionToolCalls(response, wire);
+		const completionText = blockOutcome ? blockOutcome.content + blockOutcome.toolCalls.map((tc) => tc.name + tc.arguments).join("") : (response.content ?? "") + toolCalls.map((tc) => tc.name + tc.arguments).join("");
 		if (!streaming) {
+			await awaitNonStreamingTiming((effReasoning?.length ?? 0) + completionText.length);
 			const completion = buildContentWithToolCallsCompletion(blockOutcome?.content ?? response.content ?? "", blockOutcome?.toolCalls ?? toolCalls, responseModel, effReasoning, blockOutcome ? {
 				...overrides,
 				finishReason: overrides?.finishReason ?? (blockOutcome.hasToolCalls ? "tool_calls" : "stop")
@@ -1437,7 +1454,6 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 			res.end(JSON.stringify(shapeORCompletion(completion, overrides)));
 		} else {
 			const chunks = buildContentWithToolCallsChunks(blockOutcome?.content ?? response.content ?? "", blockOutcome?.toolCalls ?? toolCalls, responseModel, chunkSize, effReasoning, overrides, blockOutcome?.ordered);
-			const completionText = blockOutcome ? blockOutcome.content + blockOutcome.toolCalls.map((tc) => tc.name + tc.arguments).join("") : (response.content ?? "") + toolCalls.map((tc) => tc.name + tc.arguments).join("");
 			const usageChunk = emitStreamingUsage ? buildUsageChunk(chunks[0]?.id ?? "chatcmpl-unknown", overrides?.model ?? responseModel, chunks[0]?.created ?? Math.floor(Date.now() / 1e3), resolveStreamingUsageTokens(overrides, completionText), overrides?.systemFingerprint) : void 0;
 			shapeORChunks(chunks, usageChunk, overrides);
 			const interruption = createInterruptionSignal(fixture);
@@ -1447,7 +1463,7 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 				signal: interruption?.signal,
 				onChunkSent: interruption?.tick,
 				usageChunk,
-				recordedTimings: resolveRecordedTimings(chunks),
+				recordedTimings: resolveRecordedTimings(chunks.length),
 				replaySpeed: fixture.replaySpeed ?? defaults.replaySpeed,
 				openRouterProcessing
 			})) {
@@ -1462,7 +1478,7 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 	if (isTextResponse(response)) {
 		if (response.webSearches?.length) defaults.logger.warn("webSearches in fixture response are not supported for Chat Completions API — ignoring");
 		const overrides = extractOverrides(response);
-		const effectiveStrict = resolveStrictMode(defaults.strict, req.headers);
+		const effectiveStrict = resolveRequestDecisions(defaults.strict, req.headers).reasoningSuppression;
 		const effReasoning = appliedPlan ? appliedReasoning : resolveReasoningForModel(response.reasoning, responseModel, effectiveStrict, defaults.logger);
 		const journalEntry = journal.add({
 			method: req.method ?? "POST",
@@ -1476,6 +1492,7 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 		});
 		recordOutcome(journalEntry);
 		if (body.stream !== true) {
+			await awaitNonStreamingTiming((effReasoning?.length ?? 0) + response.content.length);
 			const completion = buildTextCompletion(response.content, responseModel, effReasoning, overrides, body.messages);
 			res.writeHead(200, { "Content-Type": "application/json" });
 			res.end(JSON.stringify(shapeORCompletion(completion, overrides)));
@@ -1490,7 +1507,7 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 				signal: interruption?.signal,
 				onChunkSent: interruption?.tick,
 				usageChunk,
-				recordedTimings: resolveRecordedTimings(chunks),
+				recordedTimings: resolveRecordedTimings(chunks.length),
 				replaySpeed: fixture.replaySpeed ?? defaults.replaySpeed,
 				openRouterProcessing
 			})) {
@@ -1505,7 +1522,7 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 	if (isToolCallResponse(response)) {
 		if (response.webSearches?.length) defaults.logger.warn("webSearches in fixture response are not supported for Chat Completions API — ignoring");
 		const overrides = extractOverrides(response);
-		const effectiveStrict = resolveStrictMode(defaults.strict, req.headers);
+		const effectiveStrict = resolveRequestDecisions(defaults.strict, req.headers).reasoningSuppression;
 		const effReasoning = appliedPlan ? appliedReasoning : resolveReasoningForModel(response.reasoning, responseModel, effectiveStrict, defaults.logger);
 		const journalEntry = journal.add({
 			method: req.method ?? "POST",
@@ -1520,6 +1537,8 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 		recordOutcome(journalEntry);
 		const toolCalls = requireServedFunctionToolCalls(response, wire);
 		if (body.stream !== true) {
+			const toolCallText = toolCalls.map((tc) => tc.name + tc.arguments).join("");
+			await awaitNonStreamingTiming((effReasoning?.length ?? 0) + toolCallText.length);
 			const completion = buildToolCallCompletion(toolCalls, responseModel, effReasoning, overrides, body.messages);
 			res.writeHead(200, { "Content-Type": "application/json" });
 			res.end(JSON.stringify(shapeORCompletion(completion, overrides)));
@@ -1535,7 +1554,7 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 				signal: interruption?.signal,
 				onChunkSent: interruption?.tick,
 				usageChunk,
-				recordedTimings: resolveRecordedTimings(chunks),
+				recordedTimings: resolveRecordedTimings(chunks.length),
 				replaySpeed: fixture.replaySpeed ?? defaults.replaySpeed,
 				openRouterProcessing
 			})) {
@@ -1557,10 +1576,10 @@ async function handleCompletions(req, res, fixtures, journal, defaults, getReque
 			fixture
 		}
 	}));
-	writeErrorResponse(res, 500, openRouter ? serializeOpenRouterError(500, "Fixture response did not match any known type") : JSON.stringify({ error: {
+	writeErrorResponse(res, 500, openRouter ? serializeOpenRouterError(500, "Fixture response did not match any known type") : renderError("openai", "unknown-response", {
 		message: "Fixture response did not match any known type",
 		type: "server_error"
-	} }));
+	}).body);
 }
 async function createServer(fixtures, options, mounts, serviceFixtures) {
 	return createServerWithResolvedAuth(fixtures, options, resolveInboundAuth({
@@ -1629,6 +1648,7 @@ async function startServer(fixtures, options, resolvedAuth, mountList, logger, s
 	const port = options?.port ?? 0;
 	const registry = options?.metrics ? createMetricsRegistry() : void 0;
 	const serverOptions = options ?? {};
+	parsePolicy(serverOptions.policy, (message) => new Logger("warn").warn(message));
 	const validResponsesTools = (mode) => mode === "legacy" || mode === "extended";
 	if (serverOptions.responsesTools !== void 0 && !validResponsesTools(serverOptions.responsesTools)) new Logger("warn").warn(`Ignoring responsesTools because it must be "legacy" or "extended", got ${JSON.stringify(serverOptions.responsesTools)}. Using "legacy".`);
 	fixtures = createServerFixtureList(callerFixtures, () => serverOptions.responsesTools, responsesToolsExtendedView);
@@ -1687,6 +1707,9 @@ async function startServer(fixtures, options, resolvedAuth, mountList, logger, s
 		get strict() {
 			return serverOptions.strict;
 		},
+		get policy() {
+			return serverOptions.policy;
+		},
 		get strictToolArguments() {
 			return serverOptions.strictToolArguments;
 		},
@@ -1812,70 +1835,63 @@ async function startServer(fixtures, options, resolvedAuth, mountList, logger, s
 	* `toolCallEnvelope` instead of using this one.
 	*/
 	function invalidToolArgumentsEnvelope(pathname, message, code) {
-		if (pathname === OLLAMA_CHAT_PATH) return JSON.stringify({
-			error: message,
+		if (pathname === OLLAMA_CHAT_PATH) return renderError("ollama", "tool-argument", {
+			message,
 			code
-		});
-		if (pathname === GEMINI_INTERACTIONS_PATH) return JSON.stringify({ error: {
+		}).body;
+		if (pathname === GEMINI_INTERACTIONS_PATH) return renderError("interactions", "tool-argument", {
+			message,
+			code
+		}).body;
+		if (GEMINI_PATH_RE.test(pathname) || VERTEX_AI_RE.test(pathname)) return renderError("gemini", "tool-argument", {
+			message,
 			code,
-			message
-		} });
-		if (GEMINI_PATH_RE.test(pathname) || VERTEX_AI_RE.test(pathname)) return JSON.stringify({ error: {
-			code: 500,
+			googleDetails: googleFixtureToolCallErrorDetails(code)
+		}).body;
+		if (pathname === MESSAGES_PATH) return renderError("messages", "tool-argument", {
 			message,
-			status: "INTERNAL",
-			details: googleFixtureToolCallErrorDetails(code)
-		} });
-		if (pathname === MESSAGES_PATH) return JSON.stringify({
-			type: "error",
-			error: {
-				type: "api_error",
-				code,
-				message
-			}
-		});
-		if (BEDROCK_INVOKE_RE.test(pathname) || BEDROCK_STREAM_RE.test(pathname) || BEDROCK_CONVERSE_RE.test(pathname) || BEDROCK_CONVERSE_STREAM_RE.test(pathname)) return JSON.stringify({
-			__type: "InternalServerException",
+			code
+		}).body;
+		if (BEDROCK_INVOKE_RE.test(pathname) || BEDROCK_STREAM_RE.test(pathname) || BEDROCK_CONVERSE_RE.test(pathname) || BEDROCK_CONVERSE_STREAM_RE.test(pathname)) return renderError("bedrock", "tool-argument", {
 			message,
-			reason: code
-		});
-		return JSON.stringify({ error: {
+			code
+		}).body;
+		return renderError("openai", "tool-argument", {
 			message,
-			type: "server_error",
 			code
-		} });
+		}).body;
 	}
 	function invalidMisbehaviorHeaderEnvelope(pathname, detail) {
 		const code = "aimock_misbehavior_invalid";
 		const message = `${code}: ${detail}`;
-		if (pathname === OLLAMA_CHAT_PATH || pathname === OLLAMA_GENERATE_PATH) return JSON.stringify({ error: message });
-		if (pathname === COHERE_CHAT_PATH) return JSON.stringify({ message });
-		if (pathname === GEMINI_INTERACTIONS_PATH) return JSON.stringify({ error: {
-			code,
-			message
-		} });
-		if (GEMINI_PATH_RE.test(pathname) || VERTEX_AI_RE.test(pathname)) return JSON.stringify({ error: {
-			code: 400,
+		if (pathname === OLLAMA_CHAT_PATH || pathname === OLLAMA_GENERATE_PATH) return renderError("ollama", "invalid-misbehavior-header", {
 			message,
-			status: "INVALID_ARGUMENT"
-		} });
-		if (pathname === MESSAGES_PATH) return JSON.stringify({
-			type: "error",
-			error: {
-				type: "invalid_request_error",
-				code,
-				message
-			}
-		});
-		if (BEDROCK_INVOKE_RE.test(pathname) || BEDROCK_STREAM_RE.test(pathname) || BEDROCK_CONVERSE_RE.test(pathname) || BEDROCK_CONVERSE_STREAM_RE.test(pathname)) return JSON.stringify({
-			__type: "ValidationException",
-			message
-		});
-		return JSON.stringify({ error: {
+			code
+		}).body;
+		if (pathname === COHERE_CHAT_PATH) return renderError("cohere", "invalid-misbehavior-header", {
+			message,
+			code
+		}).body;
+		if (pathname === GEMINI_INTERACTIONS_PATH) return renderError("interactions", "invalid-misbehavior-header", {
+			message,
+			code
+		}).body;
+		if (GEMINI_PATH_RE.test(pathname) || VERTEX_AI_RE.test(pathname)) return renderError("gemini", "invalid-misbehavior-header", {
+			message,
+			code
+		}).body;
+		if (pathname === MESSAGES_PATH) return renderError("messages", "invalid-misbehavior-header", {
 			message,
-			type: "invalid_request_error",
 			code
-		} });
+		}).body;
+		if (BEDROCK_INVOKE_RE.test(pathname) || BEDROCK_STREAM_RE.test(pathname) || BEDROCK_CONVERSE_RE.test(pathname) || BEDROCK_CONVERSE_STREAM_RE.test(pathname)) return renderError("bedrock", "invalid-misbehavior-header", {
+			message,
+			code
+		}).body;
+		return renderError("openai", "invalid-misbehavior-header", {
+			message,
+			code
+		}).body;
 	}
 	/**
 	* The newest journal entry this request already produced, or null.
@@ -1990,10 +2006,10 @@ async function startServer(fixtures, options, resolvedAuth, mountList, logger, s
 				logger.warn(`${route}: journal write failed after handler error: ${jErr instanceof Error ? jErr.message : String(jErr)}`);
 			}
 			setCorsHeaders(res);
-			writeErrorResponse(res, status, invalidToolArguments ? (opts?.toolCallEnvelope ?? ((m, c) => invalidToolArgumentsEnvelope(pathname, m, c)))(msg, fixtureToolCallErrorCode(err)) : opts?.envelope ? opts.envelope(msg, status) : JSON.stringify({ error: {
+			writeErrorResponse(res, status, invalidToolArguments ? (opts?.toolCallEnvelope ?? ((m, c) => invalidToolArgumentsEnvelope(pathname, m, c)))(msg, fixtureToolCallErrorCode(err)) : opts?.envelope ? opts.envelope(msg, status) : renderError("openai", "route-failure", {
 				message: msg,
 				type: errorTypeForStatus(status)
-			} }));
+			}).body);
 		} else {
 			const midStreamReason = clientAbort ? "client aborted" : "handler crashed mid-stream";
 			if (existing) {
@@ -2110,10 +2126,10 @@ async function startServer(fixtures, options, resolvedAuth, mountList, logger, s
 				logger.warn(`${route}: journal write failed after handler error: ${jErr instanceof Error ? jErr.message : String(jErr)}`);
 			}
 			setCorsHeaders(res);
-			writeErrorResponse(res, 400, JSON.stringify({ error: {
+			writeErrorResponse(res, 400, renderError("openai", "route-failure", {
 				message: "Invalid request target",
 				type: "invalid_request_error"
-			} }));
+			}).body);
 			return;
 		}
 		pathname = parsedUrl.pathname;
@@ -2126,7 +2142,7 @@ async function startServer(fixtures, options, resolvedAuth, mountList, logger, s
 		}
 		if (!(req.method === "GET" && (pathname === HEALTH_PATH || pathname === READY_PATH || pathname === METRICS_PATH))) {
 			if (!validateRequestApiKey(req, resolvedAuth.policy).ok) {
-				writeApiKeyHttpRejection(res, setCorsHeaders);
+				writeApiKeyHttpRejection(res, setCorsHeaders, pathname, resolvePolicy(defaults, req.headers).authErrors);
 				return;
 			}
 			markAuthenticatedRequest(req, resolvedAuth.policy);
@@ -2193,18 +2209,9 @@ async function startServer(fixtures, options, resolvedAuth, mountList, logger, s
 		}
 		if (pathname === OLLAMA_TAGS_PATH && req.method === "GET") {
 			setCorsHeaders(res);
-			const modelIds = /* @__PURE__ */ new Set();
-			for (const f of fixtures) if (f.match.model && typeof f.match.model === "string") modelIds.add(f.match.model);
-			const models = (modelIds.size > 0 ? [...modelIds] : DEFAULT_MODELS).map((name) => ({
-				name,
-				model: name,
-				modified_at: (/* @__PURE__ */ new Date()).toISOString(),
-				size: 0,
-				digest: "",
-				details: {}
-			}));
+			const body = renderOllamaTags(collect(fixtures, "all").ids, () => (/* @__PURE__ */ new Date()).toISOString());
 			res.writeHead(200, { "Content-Type": "application/json" });
-			res.end(JSON.stringify({ models }));
+			res.end(JSON.stringify(body));
 			return;
 		}
 		const openRouterVideoContentMatch = pathname.match(OPENROUTER_VIDEO_CONTENT_RE);
@@ -2325,21 +2332,70 @@ async function startServer(fixtures, options, resolvedAuth, mountList, logger, s
 			res.end(registry.serialize());
 			return;
 		}
-		if (pathname === MODELS_PATH && req.method === "GET") {
+		const modelDetailMatch = pathname.match(MODELS_DETAIL_RE);
+		const geminiModelDetailMatch = pathname.match(GEMINI_MODEL_DETAIL_RE);
+		if (req.method === "GET" && (pathname === MODELS_PATH || pathname === GEMINI_MODELS_PATH || modelDetailMatch || geminiModelDetailMatch)) {
 			setCorsHeaders(res);
-			const modelIds = /* @__PURE__ */ new Set();
-			for (const f of fixtures) if (f.match.model && typeof f.match.model === "string") modelIds.add(f.match.model);
-			const data = (modelIds.size > 0 ? [...modelIds] : DEFAULT_MODELS).map((id) => ({
-				id,
-				object: "model",
-				created: 1686935002,
-				owned_by: "aimock"
-			}));
+			const google = pathname === GEMINI_MODELS_PATH || geminiModelDetailMatch !== null;
+			const policy = resolvePolicy(defaults, req.headers);
+			const anthropic = !google && (modelDetailMatch !== null || policy.modelCatalog === "native") && isAnthropicModelRequest(req.headers);
+			const missingHeader = anthropic && policy.requiredHeaders === "enforce" ? missingRequiredHeader("messages", req.headers) : void 0;
+			if (missingHeader) {
+				writeErrorResponse(res, 400, renderError("messages", "thinking", {
+					message: `${missingHeader}: header is required`,
+					requestId
+				}).body);
+				return;
+			}
+			const ids = coreModelIds(collect(fixtures, "all").ids);
+			const visible = google ? idsFor(ids, "google") : anthropic ? idsFor(ids, "anthropic") : ids;
+			const catalogError = (status, message) => {
+				const rendered = google ? renderError("gemini", "fixture", {
+					bodyCode: status,
+					status: status === 404 ? "NOT_FOUND" : "INVALID_ARGUMENT",
+					message
+				}) : anthropic ? renderError("messages", "thinking", {
+					type: status === 404 ? "not_found_error" : "invalid_request_error",
+					message
+				}) : renderError("openai", "validation", { message });
+				res.writeHead(status, {
+					"Content-Type": "application/json",
+					...rendered.headers
+				});
+				res.end(rendered.body);
+			};
+			let body;
+			const detail = modelDetailMatch ?? geminiModelDetailMatch;
+			if (detail) {
+				let id;
+				try {
+					id = decodeURIComponent(detail[1]);
+				} catch {
+					catalogError(404, "Model not found");
+					return;
+				}
+				if (!visible.includes(id)) {
+					catalogError(404, `Model not found: ${id}`);
+					return;
+				}
+				body = google ? renderGeminiModel(id) : anthropic ? renderAnthropicModel(id) : renderOpenAIModel(id);
+			} else if (google) {
+				const result = geminiPage(visible, parsedUrl.searchParams);
+				if (result.error !== void 0) {
+					catalogError(400, result.error);
+					return;
+				}
+				body = renderGeminiModels(result.page);
+			} else if (anthropic) {
+				const result = anthropicPage(visible, parsedUrl.searchParams);
+				if (result.error !== void 0) {
+					catalogError(400, result.error);
+					return;
+				}
+				body = renderAnthropicModels(result.page);
+			} else body = renderOpenAIModels(ids);
 			res.writeHead(200, { "Content-Type": "application/json" });
-			res.end(JSON.stringify({
-				object: "list",
-				data
-			}));
+			res.end(JSON.stringify(body));
 			return;
 		}
 		const batchesCancelMatch = pathname.match(BATCHES_CANCEL_RE);
@@ -2597,10 +2653,10 @@ async function startServer(fixtures, options, resolvedAuth, mountList, logger, s
 				if (limitParam) {
 					const limit = Number(limitParam);
 					if (!/^\+?\d+$/.test(limitParam.trim()) || !Number.isSafeInteger(limit) || limit <= 0) {
-						writeErrorResponse(res, 400, JSON.stringify({ error: {
+						writeErrorResponse(res, 400, renderError("openai", "validation", {
 							message: `Invalid limit parameter: "${limitParam}"`,
 							type: "invalid_request_error"
-						} }));
+						}).body);
 						return;
 					}
 					opts = { limit };
@@ -2625,7 +2681,27 @@ async function startServer(fixtures, options, resolvedAuth, mountList, logger, s
 				ensureRawFixturePositions();
 				await handleResponses(req, res, raw, fixtures, journal, defaults, setCorsHeaders);
 			} catch (err) {
-				routeError(req, res, err, pathname, { streamEvent: (m) => `event: error\ndata: ${JSON.stringify({ error: { message: m } })}\n\n` });
+				routeError(req, res, err, pathname, { streamEvent: (m) => `event: error\ndata: ${renderError("openai", "stream-failure", { message: m }).body}\n\n` });
+			}
+			return;
+		}
+		const geminiCountMatch = pathname.match(GEMINI_COUNT_RE);
+		if (req.method === "POST" && (pathname === MESSAGES_COUNT_PATH || geminiCountMatch)) {
+			const google = geminiCountMatch !== null;
+			try {
+				handleCountTokens(req, res, await readBody(req), google, geminiCountMatch?.[1] ?? "", journal, defaults, setCorsHeaders);
+			} catch (err) {
+				routeError(req, res, err, pathname, {
+					service: google ? "gemini" : "anthropic",
+					envelope: (message, status) => google ? renderError("gemini", "fixture", {
+						message,
+						bodyCode: status,
+						status: status === 400 ? "INVALID_ARGUMENT" : "INTERNAL"
+					}).body : renderError("messages", "thinking", {
+						message,
+						type: status === 400 ? "invalid_request_error" : "api_error"
+					}).body
+				});
 			}
 			return;
 		}
@@ -2635,7 +2711,7 @@ async function startServer(fixtures, options, resolvedAuth, mountList, logger, s
 				ensureRawFixturePositions();
 				await handleMessages(req, res, raw, fixtures, journal, defaults, setCorsHeaders);
 			} catch (err) {
-				routeError(req, res, err, pathname, { streamEvent: (m) => `event: error\ndata: ${JSON.stringify({ error: { message: m } })}\n\n` });
+				routeError(req, res, err, pathname, { streamEvent: (m) => `event: error\ndata: ${renderError("openai", "stream-failure", { message: m }).body}\n\n` });
 			}
 			return;
 		}
@@ -2645,7 +2721,7 @@ async function startServer(fixtures, options, resolvedAuth, mountList, logger, s
 				ensureRawFixturePositions();
 				await handleCohere(req, res, raw, fixtures, journal, defaults, setCorsHeaders);
 			} catch (err) {
-				routeError(req, res, err, pathname, { streamEvent: (m) => `event: error\ndata: ${JSON.stringify({ error: { message: m } })}\n\n` });
+				routeError(req, res, err, pathname, { streamEvent: (m) => `event: error\ndata: ${renderError("openai", "stream-failure", { message: m }).body}\n\n` });
 			}
 			return;
 		}
@@ -2811,7 +2887,7 @@ async function startServer(fixtures, options, resolvedAuth, mountList, logger, s
 				ensureRawFixturePositions();
 				await handleGeminiInteractions(req, res, raw, fixtures, journal, defaults, setCorsHeaders);
 			} catch (err) {
-				routeError(req, res, err, pathname, { streamEvent: (m) => `data: ${JSON.stringify({ error: { message: m } })}\n\n` });
+				routeError(req, res, err, pathname, { streamEvent: (m) => `data: ${renderError("openai", "stream-failure", { message: m }).body}\n\n` });
 			}
 			return;
 		}
@@ -2836,7 +2912,7 @@ async function startServer(fixtures, options, resolvedAuth, mountList, logger, s
 				ensureRawFixturePositions();
 				await handleGemini(req, res, raw, geminiModel, streaming, fixtures, journal, defaults, setCorsHeaders);
 			} catch (err) {
-				routeError(req, res, err, pathname, { streamEvent: (m) => `data: ${JSON.stringify({ error: { message: m } })}\n\n` });
+				routeError(req, res, err, pathname, { streamEvent: (m) => `data: ${renderError("openai", "stream-failure", { message: m }).body}\n\n` });
 			}
 			return;
 		}
@@ -2849,7 +2925,7 @@ async function startServer(fixtures, options, resolvedAuth, mountList, logger, s
 				ensureRawFixturePositions();
 				await handleGemini(req, res, raw, vertexModel, streaming, fixtures, journal, defaults, setCorsHeaders, "vertexai");
 			} catch (err) {
-				routeError(req, res, err, pathname, { streamEvent: (m) => `data: ${JSON.stringify({ error: { message: m } })}\n\n` });
+				routeError(req, res, err, pathname, { streamEvent: (m) => `data: ${renderError("openai", "stream-failure", { message: m }).body}\n\n` });
 			}
 			return;
 		}
@@ -2965,10 +3041,10 @@ async function startServer(fixtures, options, resolvedAuth, mountList, logger, s
 			try {
 				voiceId = decodeURIComponent(elevenLabsVoiceMatch[1]);
 			} catch {
-				writeErrorResponse(res, 400, JSON.stringify({ error: {
+				writeErrorResponse(res, 400, renderError("openai", "validation", {
 					message: `Invalid voice id '${elevenLabsVoiceMatch[1]}': malformed percent-encoding`,
 					type: "invalid_request_error"
-				} }));
+				}).body);
 				journal.add({
 					method: req.method ?? "GET",
 					path: req.url ?? pathname,
@@ -3118,15 +3194,15 @@ async function startServer(fixtures, options, resolvedAuth, mountList, logger, s
 			await handleCompletions(req, res, fixtures, journal, defaults, () => lastJournalEntryFor(req), azureDeploymentId, completionsProvider, isOpenRouter, ensureRawFixturePositions, originalPathname.startsWith("/api/v3/"));
 		} catch (err) {
 			routeError(req, res, err, pathname, {
-				envelope: (m, status) => isOpenRouter ? serializeOpenRouterError(status, m) : JSON.stringify({ error: {
+				envelope: (m, status) => isOpenRouter ? serializeOpenRouterError(status, m) : renderError("openai", "route-failure", {
 					message: m,
 					type: errorTypeForStatus(status)
-				} }),
+				}).body,
 				...isOpenRouter ? { toolCallEnvelope: (m, code) => serializeOpenRouterError(500, m, { reason: code }) } : {},
-				streamEvent: (m, status) => `data: ${isOpenRouter ? serializeOpenRouterError(status, m) : JSON.stringify({ error: {
+				streamEvent: (m, status) => `data: ${isOpenRouter ? serializeOpenRouterError(status, m) : renderError("openai", "route-failure", {
 					message: m,
 					type: errorTypeForStatus(status)
-				} })}\n\n`
+				}).body}\n\n`
 			});
 		}
 	}