@hilbras/omnicode 1.2.2 → 1.2.3

package/CHANGELOG.md

@@ -1,5 +1,13 @@
 # @hilbras/omnicode
 
+## 1.2.3
+
+### Patch Changes
+
+- [`b9e3bbd`](https://github.com/Hilbras/OmniCode/commit/b9e3bbd570cde32e613db2f0e6997c52279e7698) Thanks [@lacrous](https://github.com/lacrous)! - Show each model's real context window from its provider catalog instead of a flat default; a model with no declared window now says so until you set one with /context.
+
+- Requests to providers that reject the prompt_cache_key field now retry without it instead of failing.
+
 ## 1.2.2
 
 ### Patch Changes

package/dist/main.mjs

@@ -55760,12 +55760,19 @@ function userExtras(existing, remoteOwnedFields) {
 	}
 	return out;
 }
+function preservedMaxContextSize(current, remote) {
+	if (remote.maxContextSize !== void 0) return void 0;
+	const existing = current["maxContextSize"];
+	return typeof existing === "number" && Number.isFinite(existing) && existing > 0 ? existing : void 0;
+}
 function mergeRefreshedModelAlias(existing, remote, remoteOwnedFields) {
 	const current = isRecord$25(existing) ? existing : {};
 	const overrides = cloneOverrides(isRecord$25(current["overrides"]) ? current["overrides"] : void 0);
+	const carried = preservedMaxContextSize(current, remote);
 	return {
 		...userExtras(current, remoteOwnedFields),
 		...remote,
+		maxContextSize: remote.maxContextSize ?? carried,
 		...overrides !== void 0 ? { overrides } : {}
 	};
 }
@@ -57170,7 +57177,6 @@ function resolveMaxContextSize(model) {
 	const output = model.limit?.output;
 	if (typeof context === "number" && Number.isInteger(context) && context > 0) return context;
 	if (typeof output === "number" && Number.isInteger(output) && output > 0) return output;
-	return CUSTOM_REGISTRY_DEFAULT_MAX_CONTEXT;
 }
 function resolveCapabilities(model) {
 	if (hasRichCapabilityHints(model)) return capabilitiesFromCustomEntry(model);
@@ -57349,7 +57355,7 @@ function applyCustomRegistryEntries(config, entries, source, removal) {
 	config["defaultProvider"] = r.previousDefaultProvider !== void 0 && config.providers[r.previousDefaultProvider] !== void 0 ? r.previousDefaultProvider : void 0;
 	if (r.previousDefault !== void 0 && config.defaultModel === void 0) config.thinking = void 0;
 }
-var CUSTOM_REGISTRY_DEFAULT_MAX_CONTEXT, CUSTOM_REGISTRY_DEFAULT_CAPABILITIES, ALLOWED_PROVIDER_TYPES, RESERVED_PROVIDER_IDS, CustomRegistryApiError;
+var CUSTOM_REGISTRY_DEFAULT_CAPABILITIES, ALLOWED_PROVIDER_TYPES, RESERVED_PROVIDER_IDS, CustomRegistryApiError;
 var init_custom_registry = __esmMin((() => {
 	init_api_error();
 	init_managed_omnicode();
@@ -57357,7 +57363,6 @@ var init_custom_registry = __esmMin((() => {
 	init_open_platform();
 	init_provider_credential();
 	init_utils$4();
-	CUSTOM_REGISTRY_DEFAULT_MAX_CONTEXT = 131072;
 	CUSTOM_REGISTRY_DEFAULT_CAPABILITIES = ["tool_use"];
 	ALLOWED_PROVIDER_TYPES = new Set([
 		"anthropic",
@@ -57481,7 +57486,7 @@ function toDiscoveredModel(row) {
 	const rawId = row["id"] ?? row["name"];
 	if (typeof rawId !== "string" || rawId.trim().length === 0) return void 0;
 	const id = stripResourcePrefix(rawId.trim());
-	const maxContextSize = positiveInt$1(row["context_length"] ?? row["context_window"]) ?? positiveInt$1(row["max_output_tokens"]) ?? positiveInt$1(row["max_context_size"]);
+	const maxContextSize = positiveInt$1(row["context_length"] ?? row["context_window"]) ?? positiveInt$1(row["max_context_size"]);
 	const rawName = row["name"] ?? row["display_name"];
 	const capabilities = [];
 	const object = row["object"];
@@ -57595,7 +57600,7 @@ function applyDiscoveredModels(config, providerId, models, aliasPrefix = `${prov
 		const alias = {
 			provider: providerId,
 			model: model.id,
-			maxContextSize: model.maxContextSize ?? 131072,
+			maxContextSize: model.maxContextSize,
 			capabilities: [...model.capabilities ?? CUSTOM_REGISTRY_DEFAULT_CAPABILITIES],
 			displayName: model.displayName ?? model.id
 		};
@@ -62933,7 +62938,7 @@ var init_schema$4 = __esmMin((() => {
 	ModelAliasBaseSchema = object({
 		provider: string(),
 		model: string(),
-		maxContextSize: number$1().int().min(1),
+		maxContextSize: number$1().int().min(1).optional(),
 		maxInputSize: number$1().int().min(1).optional(),
 		maxOutputSize: number$1().int().min(1).optional(),
 		capabilities: array(string()).optional(),
@@ -64194,7 +64199,7 @@ function applyEnvModelConfig(config, env = process.env) {
 	const apiKey = trimmed$2(env["OMNICODE_MODEL_API_KEY"]);
 	if (apiKey === void 0) fail$1("OMNICODE_MODEL_NAME is set but OMNICODE_MODEL_API_KEY is missing.");
 	const maxContextRaw = trimmed$2(env["OMNICODE_MODEL_MAX_CONTEXT_SIZE"]);
-	const maxContextSize = maxContextRaw === void 0 ? DEFAULT_MAX_CONTEXT_SIZE$1 : parsePositiveInt$4(maxContextRaw, "OMNICODE_MODEL_MAX_CONTEXT_SIZE");
+	const maxContextSize = maxContextRaw === void 0 ? DEFAULT_MAX_CONTEXT_SIZE : parsePositiveInt$4(maxContextRaw, "OMNICODE_MODEL_MAX_CONTEXT_SIZE");
 	const type = parseProviderType(trimmed$2(env["OMNICODE_MODEL_PROVIDER_TYPE"]));
 	const baseUrl = trimmed$2(env["OMNICODE_MODEL_BASE_URL"]) ?? DEFAULT_BASE_URL$2[type];
 	const provider = {
@@ -64265,7 +64270,7 @@ function rawThinking(config) {
 	const raw = config.raw?.["thinking"];
 	return typeof raw === "object" && raw !== null && !Array.isArray(raw) ? raw : void 0;
 }
-var ENV_MODEL_PROVIDER_KEY$1, ENV_MODEL_ALIAS_KEY$1, ALLOWED_TYPES, DEFAULT_BASE_URL$2, DEFAULT_MAX_CONTEXT_SIZE$1, DEFAULT_CAPABILITIES$1;
+var ENV_MODEL_PROVIDER_KEY$1, ENV_MODEL_ALIAS_KEY$1, ALLOWED_TYPES, DEFAULT_BASE_URL$2, DEFAULT_MAX_CONTEXT_SIZE, DEFAULT_CAPABILITIES$1;
 var init_env_model = __esmMin((() => {
 	init_errors$46();
 	init_resolve$1();
@@ -64281,7 +64286,7 @@ var init_env_model = __esmMin((() => {
 		omnicode: "https://api.hilbras.ai/v1",
 		openai: "https://api.openai.com/v1"
 	};
-	DEFAULT_MAX_CONTEXT_SIZE$1 = 262144;
+	DEFAULT_MAX_CONTEXT_SIZE = 262144;
 	DEFAULT_CAPABILITIES$1 = ["image_in", "thinking"];
 }));
 //#endregion
@@ -64830,7 +64835,7 @@ function effectiveModelAlias(alias, providerType) {
 		...overrides
 	};
 	if (overrides?.supportEfforts !== void 0 && overrides.defaultEffort === void 0 && effective.defaultEffort !== void 0 && !overrides.supportEfforts.includes(effective.defaultEffort)) delete effective.defaultEffort;
-	return withAnthropicProfile$1(effective.maxInputSize !== void 0 && effective.maxInputSize > effective.maxContextSize ? {
+	return withAnthropicProfile$1(effective.maxInputSize !== void 0 && effective.maxContextSize !== void 0 && effective.maxInputSize > effective.maxContextSize ? {
 		...effective,
 		maxInputSize: effective.maxContextSize
 	} : effective, providerType);
@@ -186227,20 +186232,6 @@ function parseAnthropicModelVersion(model, requireClaudeMarker = false) {
 	};
 	return null;
 }
-function lookupClaudeCeiling(version) {
-	const { family, major, minor } = version;
-	if (minor !== null) for (let candidate = minor; candidate >= 0; candidate--) {
-		const ceiling = CEILING_BY_FAMILY_VERSION[`${family}-${major}-${candidate}`];
-		if (ceiling !== void 0) return ceiling;
-	}
-	return CEILING_BY_FAMILY_VERSION[`${family}-${major}`];
-}
-function resolveDefaultMaxTokens(model, override) {
-	const parsed = parseAnthropicModelVersion(model, true);
-	const ceiling = parsed === null ? void 0 : lookupClaudeCeiling(parsed);
-	if (ceiling === void 0) return override ?? FALLBACK_MAX_TOKENS;
-	return override === void 0 ? ceiling : Math.min(override, ceiling);
-}
 function matchKnownAnthropicModelProfile(model) {
 	const normalized = model.toLowerCase();
 	if (/mythos[-._]preview/.test(normalized)) return ALWAYS_ADAPTIVE_MAX_PROFILE;
@@ -186334,7 +186325,7 @@ function encodeThinking(thinking, model) {
 	if ((profile.supportsEffortParam || budgetTokens === void 0) && effort !== "on") patch["output_config"] = { effort };
 	return patch;
 }
-var INTERLEAVED_THINKING_BETA, BUDGET_THINKING_EFFORTS, ADAPTIVE_MAX_EFFORTS, LATEST_OPUS_THINKING_EFFORTS, BUDGET_PROFILE, OPUS_45_PROFILE, ADAPTIVE_MAX_PROFILE, LATEST_OPUS_PROFILE, ALWAYS_ADAPTIVE_PROFILE, ALWAYS_ADAPTIVE_MAX_PROFILE, FAMILY_FIRST_RE, VERSION_FIRST_RE, BARE_FAMILY_RE, CEILING_BY_FAMILY_VERSION, FALLBACK_MAX_TOKENS, CLAUDE_FAMILY_WORD_RE;
+var INTERLEAVED_THINKING_BETA, BUDGET_THINKING_EFFORTS, ADAPTIVE_MAX_EFFORTS, LATEST_OPUS_THINKING_EFFORTS, BUDGET_PROFILE, OPUS_45_PROFILE, ADAPTIVE_MAX_PROFILE, LATEST_OPUS_PROFILE, ALWAYS_ADAPTIVE_PROFILE, ALWAYS_ADAPTIVE_MAX_PROFILE, FAMILY_FIRST_RE, VERSION_FIRST_RE, BARE_FAMILY_RE, CLAUDE_FAMILY_WORD_RE;
 var init_profile$1 = __esmMin((() => {
 	init_thinking$2();
 	INTERLEAVED_THINKING_BETA = "interleaved-thinking-2025-05-14";
@@ -186389,32 +186380,6 @@ var init_profile$1 = __esmMin((() => {
 	FAMILY_FIRST_RE = /(opus|sonnet|haiku|fable|mythos)[-._](\d{1,2})(?!\d)(?:[-._](\d{1,2})(?!\d))?/;
 	VERSION_FIRST_RE = /(\d{1,2})[-._](\d{1,2})[-._](opus|sonnet|haiku)/;
 	BARE_FAMILY_RE = /(\d{1,2})[-._](opus|sonnet|haiku)/;
-	CEILING_BY_FAMILY_VERSION = {
-		"fable-5": 128e3,
-		"mythos-5": 128e3,
-		"opus-4-8": 128e3,
-		"opus-4-7": 128e3,
-		"opus-4-6": 128e3,
-		"opus-4-5": 64e3,
-		"opus-4-1": 32e3,
-		"opus-4-0": 32e3,
-		"opus-4": 32e3,
-		"sonnet-5": 128e3,
-		"sonnet-4-6": 128e3,
-		"sonnet-4-5": 64e3,
-		"sonnet-4-0": 64e3,
-		"sonnet-4": 64e3,
-		"haiku-4-5": 64e3,
-		"haiku-4": 64e3,
-		"opus-3-5": 8192,
-		"sonnet-3-5": 8192,
-		"sonnet-3-7": 8192,
-		"haiku-3-5": 8192,
-		"opus-3": 4096,
-		"sonnet-3": 4096,
-		"haiku-3": 4096
-	};
-	FALLBACK_MAX_TOKENS = 64e3;
 	CLAUDE_FAMILY_WORD_RE = /\b(?:opus|sonnet|haiku|fable|mythos)\b/;
 }));
 //#endregion
@@ -186521,7 +186486,7 @@ function assembleAnthropicRequest(input, parts) {
 	return {
 		params: {
 			model: input.model.model,
-			max_tokens: resolveDefaultMaxTokens(input.model.model, input.model.defaultMaxTokens),
+			max_tokens: input.model.defaultMaxTokens,
 			metadata: input.cacheKey === void 0 ? void 0 : { user_id: input.cacheKey },
 			...restKwargs,
 			system: input.systemPrompt ? [{
@@ -186730,13 +186695,10 @@ function prepareAnthropicRequest(input, options) {
 	if (input.thinking !== void 0) kwargs = applyThinking(kwargs, input.thinking, trait?.thinking, ctx, (t, c) => encodeThinking(t, c.model)).kwargs;
 	if (input.responseFormat !== void 0) kwargs = applyAnthropicResponseFormat(kwargs, input.responseFormat);
 	const cap = resolveMaxCompletionCap(input);
-	if (cap !== void 0) {
-		const capped = resolveDefaultMaxTokens(ctx.model.model, cap);
-		kwargs = {
-			...kwargs,
-			...trait?.encodeMaxCompletionTokens?.(capped, ctx) ?? encodeAnthropicMaxTokens(capped)
-		};
-	}
+	if (cap !== void 0) kwargs = {
+		...kwargs,
+		...trait?.encodeMaxCompletionTokens?.(cap, ctx) ?? encodeAnthropicMaxTokens(cap)
+	};
 	kwargs = assign$1(kwargs, input.extraParams?.anthropic ?? {});
 	if (input.thinking?.keep !== void 0) kwargs = applyAnthropicThinkingKeep(kwargs, input.thinking.keep);
 	kwargs = shake(kwargs);
@@ -188283,7 +188245,9 @@ var init_prompt_cache_key_fallback = __esmMin((() => {
 		/unsupported parameter/,
 		/unknown parameter/,
 		/unrecognized request argument/,
-		/extra inputs are not permitted/
+		/extra inputs are not permitted/,
+		/is unsupported/,
+		/is not supported/
 	];
 	PromptCacheKeyRejections = class {
 		rejected = /* @__PURE__ */ new Set();
@@ -189523,7 +189487,7 @@ var init_protocolAdapterRegistry = __esmMin((() => {
 				provider: route.providerId,
 				model: model.name,
 				capability: toLlmCapability(model.capabilities),
-				maxContextSize: model.maxContextSize > 0 ? model.maxContextSize : void 0,
+				maxContextSize: typeof model.maxContextSize === "number" && model.maxContextSize > 0 ? model.maxContextSize : void 0,
 				maxInputSize: model.maxInputSize,
 				defaultMaxTokens: model.providerOptions?.defaultMaxTokens,
 				baseUrl: model.baseUrl,
@@ -189687,7 +189651,7 @@ function mergeModelOverrides(existing, fromEnv) {
 		...fromEnv
 	};
 }
-var ENV_MODEL_ALIAS_KEY, DEFAULT_MAX_CONTEXT_SIZE, DEFAULT_CAPABILITIES, omnicodeModelEnvOverlay;
+var ENV_MODEL_ALIAS_KEY, DEFAULT_CAPABILITIES, omnicodeModelEnvOverlay;
 var init_envOverlay = __esmMin((() => {
 	init_errors$39();
 	init_env$1();
@@ -189697,7 +189661,6 @@ var init_envOverlay = __esmMin((() => {
 	init_provider_definition();
 	init_configSection$17();
 	ENV_MODEL_ALIAS_KEY = "__omnicode_env_model__";
-	DEFAULT_MAX_CONTEXT_SIZE = 262144;
 	DEFAULT_CAPABILITIES = ["image_in", "thinking"];
 	omnicodeModelEnvOverlay = {
 		apply(effective, getEnv, validate) {
@@ -189721,7 +189684,7 @@ var init_envOverlay = __esmMin((() => {
 				return changed;
 			}
 			const maxContextRaw = trimmed(getEnv("OMNICODE_MODEL_MAX_CONTEXT_SIZE"));
-			const maxContextSize = maxContextRaw === void 0 ? DEFAULT_MAX_CONTEXT_SIZE : parsePositiveInt$2(maxContextRaw, "OMNICODE_MODEL_MAX_CONTEXT_SIZE");
+			const maxContextSize = maxContextRaw === void 0 ? void 0 : parsePositiveInt$2(maxContextRaw, "OMNICODE_MODEL_MAX_CONTEXT_SIZE");
 			const maxOutputRaw = trimmed(getEnv("OMNICODE_MODEL_MAX_OUTPUT_SIZE"));
 			const maxOutputSize = maxOutputRaw === void 0 ? void 0 : parsePositiveInt$2(maxOutputRaw, "OMNICODE_MODEL_MAX_OUTPUT_SIZE");
 			const capabilities = parseCapabilities(getEnv("OMNICODE_MODEL_CAPABILITIES")) ?? DEFAULT_CAPABILITIES;
@@ -190173,11 +190136,6 @@ function resolveModelForReady(modelId, models, providers, defaultProvider) {
 		resolved: false,
 		reason: "unresolvable"
 	};
-	const maxContextSize = model.maxContextSize;
-	if (maxContextSize === void 0 || maxContextSize <= 0) return {
-		resolved: false,
-		reason: "unresolvable"
-	};
 	if (resolveModelProtocol(model, provider) === void 0) return {
 		resolved: false,
 		reason: "unresolvable"
@@ -190217,7 +190175,7 @@ function toProtocolModelFallback(modelId, record, providerType) {
 		provider: effective.provider ?? "",
 		model: modelId,
 		display_name: effective.displayName ?? effective.model ?? modelId,
-		max_context_size: effective.maxContextSize ?? 0,
+		max_context_size: effective.maxContextSize,
 		capabilities: effective.capabilities,
 		support_efforts: effective.supportEfforts,
 		default_effort: effective.defaultEffort
@@ -190253,7 +190211,7 @@ var init_catalog$2 = __esmMin((() => {
 		provider: string().min(1),
 		model: string().min(1),
 		display_name: string().min(1).optional(),
-		max_context_size: number$1().int().min(1),
+		max_context_size: number$1().int().min(1).optional(),
 		capabilities: array(string()).optional(),
 		support_efforts: array(string()).optional(),
 		default_effort: string().optional()
@@ -191414,7 +191372,7 @@ function resolveModelCapabilities(declaredCapabilities, detected, maxContextSize
 		audio_in: declared.has("audio_in") || detected.audio_in,
 		thinking: declared.has("thinking") || declared.has("always_thinking") || detected.thinking,
 		tool_use: declared.has("tool_use") || detected.tool_use,
-		max_context_tokens: maxContextSize,
+		max_context_tokens: maxContextSize ?? 0,
 		max_input_tokens: maxInputSize,
 		dynamically_loaded_tools: declared.has("dynamically_loaded_tools") || detected.dynamically_loaded_tools === true
 	};
@@ -191735,7 +191693,6 @@ var init_catalog_service = __esmMin((() => {
 			const providerType = providerConfig?.type ?? protocol;
 			const resolvedBaseUrl = protocol === "anthropic" && rawBaseUrl !== void 0 ? stripTrailingV1(rawBaseUrl) : rawBaseUrl;
 			if (wireName === void 0) throw new Error2(CONFIG_INVALID_ERROR_CODE, `Model "${id}" must define a wire-facing name in config.toml.`);
-			if (model.maxContextSize === void 0) throw new Error2(CONFIG_INVALID_ERROR_CODE, `Model "${id}" must define a positive max_context_size in config.toml.`);
 			const detectedCapability = this.protocolRegistry.resolveCapability(protocol, wireName, providerType);
 			const capabilities = resolveModelCapabilities(model.capabilities, detectedCapability, model.maxContextSize, model.maxInputSize);
 			const providerOptions = buildProtocolProviderOptions(model, protocol, providerConfig, resolvedBaseUrl);
@@ -284072,7 +284029,7 @@ function catalogModelToAlias(providerId, model) {
 	return {
 		provider: providerId,
 		model: model.id,
-		maxContextSize: model.capability.max_context_tokens,
+		maxContextSize: model.capability.max_context_tokens > 0 ? model.capability.max_context_tokens : void 0,
 		maxInputSize: model.capability.max_input_tokens,
 		maxOutputSize: model.maxOutputSize,
 		capabilities: model.alwaysThinking === true ? caps?.map((cap) => cap === "thinking" ? "always_thinking" : cap) : caps,
@@ -334266,6 +334223,195 @@ var init_catalog_fetch = __esmMin((() => {
 	init_built_in_catalog();
 }));
 //#endregion
+//#region src/utils/built-in-providers.ts
+function getBuiltInProvider(id) {
+	const needle = id.trim().toLowerCase();
+	return BUILT_IN_PROVIDERS.find((p) => p.id === needle);
+}
+var OPENCODE_FREE_MODEL_PATTERNS, OPENCODE_FREE_TIER, BUILT_IN_PROVIDERS;
+var init_built_in_providers = __esmMin((() => {
+	OPENCODE_FREE_MODEL_PATTERNS = [
+		"*-free",
+		"*-contributor-free",
+		"big-pickle"
+	];
+	OPENCODE_FREE_TIER = {
+		baseUrl: "https://opencode.ai/inference/v1",
+		match: OPENCODE_FREE_MODEL_PATTERNS,
+		apiKey: "public"
+	};
+	BUILT_IN_PROVIDERS = [
+		{
+			id: "cline",
+			name: "Cline",
+			wire: "openai",
+			baseUrl: "https://api.cline.bot/api/v1",
+			description: "One key for Anthropic, OpenAI, Google and more",
+			consoleUrl: "https://app.cline.bot"
+		},
+		{
+			id: "openrouter",
+			name: "OpenRouter",
+			wire: "openai",
+			baseUrl: "https://openrouter.ai/api/v1",
+			description: "One key for 400+ models across many providers",
+			consoleUrl: "https://openrouter.ai/keys",
+			catalogProvider: "openrouter"
+		},
+		{
+			id: "opencode-zen",
+			name: "OpenCode Zen",
+			wire: "openai",
+			baseUrl: "https://opencode.ai/zen/v1",
+			description: "Curated models tested by the OpenCode team",
+			consoleUrl: "https://opencode.ai/zen",
+			protocolOverrides: { "claude-*": "anthropic" },
+			excludedModels: OPENCODE_FREE_MODEL_PATTERNS,
+			freeTier: [OPENCODE_FREE_TIER],
+			catalogProvider: "opencode"
+		},
+		{
+			id: "opencode-go",
+			name: "OpenCode Go",
+			wire: "openai",
+			baseUrl: "https://opencode.ai/zen/go/v1",
+			description: "OpenCode Zen models on the Go plan",
+			consoleUrl: "https://opencode.ai/zen",
+			excludedModels: OPENCODE_FREE_MODEL_PATTERNS,
+			freeTier: [OPENCODE_FREE_TIER],
+			catalogProvider: "opencode"
+		},
+		{
+			id: "nvidia",
+			name: "NVIDIA",
+			wire: "openai",
+			baseUrl: "https://integrate.api.nvidia.com/v1",
+			description: "NVIDIA-hosted open models, free developer tier",
+			consoleUrl: "https://build.nvidia.com",
+			keyHint: "nvapi-",
+			catalogProvider: "nvidia"
+		},
+		{
+			id: "nara",
+			name: "NaraRouter",
+			wire: "openai",
+			baseUrl: "https://router.bynara.id/v1",
+			description: "Affordable multi-model gateway",
+			consoleUrl: "https://router.bynara.id",
+			keyHint: "sk-nry-"
+		},
+		{
+			id: "tokenharbor",
+			name: "Token Harbor",
+			wire: "openai",
+			baseUrl: "https://tokenharbor.ai/v1",
+			description: "One universal key for many AI providers",
+			consoleUrl: "https://tokenharbor.ai/dashboard/api-keys",
+			keyHint: "thk_live_"
+		},
+		{
+			id: "openai",
+			name: "OpenAI",
+			wire: "openai",
+			baseUrl: "https://api.openai.com/v1",
+			description: "GPT, o-series and Codex models",
+			consoleUrl: "https://platform.openai.com/api-keys",
+			keyHint: "sk-",
+			catalogProvider: "openai"
+		},
+		{
+			id: "anthropic",
+			name: "Anthropic",
+			wire: "anthropic",
+			baseUrl: "https://api.anthropic.com",
+			description: "Claude models",
+			consoleUrl: "https://console.anthropic.com/settings/keys",
+			keyHint: "sk-ant-",
+			authStyle: "x-api-key",
+			catalogProvider: "anthropic"
+		},
+		{
+			id: "gemini",
+			name: "Google Gemini",
+			wire: "google-genai",
+			baseUrl: "https://generativelanguage.googleapis.com/v1beta",
+			description: "Gemini models",
+			consoleUrl: "https://aistudio.google.com/apikey",
+			keyHint: "AIza",
+			authStyle: "x-goog-api-key",
+			catalogProvider: "google"
+		},
+		{
+			id: "grok",
+			name: "xAI Grok",
+			wire: "openai",
+			baseUrl: "https://api.x.ai/v1",
+			description: "Grok models",
+			consoleUrl: "https://console.x.ai",
+			keyHint: "xai-",
+			catalogProvider: "xai"
+		},
+		{
+			id: "groq",
+			name: "Groq",
+			wire: "openai",
+			baseUrl: "https://api.groq.com/openai/v1",
+			description: "Fast open models on Groq hardware",
+			consoleUrl: "https://console.groq.com/keys",
+			keyHint: "gsk_",
+			catalogProvider: "groq"
+		},
+		{
+			id: "qwen",
+			name: "Qwen",
+			wire: "openai",
+			baseUrl: "https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
+			description: "Alibaba Qwen models",
+			consoleUrl: "https://bailian.console.alibabacloud.com/",
+			keyHint: "sk-",
+			catalogProvider: "alibaba"
+		},
+		{
+			id: "minimax",
+			name: "MiniMax",
+			wire: "openai",
+			baseUrl: "https://api.minimax.io/v1",
+			description: "MiniMax models",
+			consoleUrl: "https://platform.minimax.io/user-center/basic-information/interface-key",
+			catalogProvider: "minimax"
+		},
+		{
+			id: "deepseek",
+			name: "DeepSeek",
+			wire: "openai",
+			baseUrl: "https://api.deepseek.com",
+			description: "DeepSeek chat and reasoning models",
+			consoleUrl: "https://platform.deepseek.com/api_keys",
+			keyHint: "sk-",
+			catalogProvider: "deepseek"
+		},
+		{
+			id: "mistral",
+			name: "Mistral",
+			wire: "openai",
+			baseUrl: "https://api.mistral.ai/v1",
+			description: "Mistral and Magistral models",
+			consoleUrl: "https://console.mistral.ai/api-keys",
+			catalogProvider: "mistral"
+		},
+		{
+			id: "huggingface",
+			name: "Hugging Face",
+			wire: "openai",
+			baseUrl: "https://router.huggingface.co/v1",
+			description: "Open models via the Hugging Face router",
+			consoleUrl: "https://huggingface.co/settings/tokens",
+			keyHint: "hf_",
+			catalogProvider: "huggingface"
+		}
+	];
+}));
+//#endregion
 //#region src/utils/builtin-model-import.ts
 /**
 * Whether `id` is named by a pattern: `claude-*` is a prefix, `*-free` a suffix,
@@ -334410,7 +334556,8 @@ function backfillCatalogLimits(input) {
 		if (id === void 0) continue;
 		const limit = input.limits.get(id);
 		if (limit === void 0) continue;
-		if (record["maxContextSize"] !== 131072) continue;
+		const current = record["maxContextSize"];
+		if (current !== void 0 && current !== 131072) continue;
 		input.models[key] = {
 			...record,
 			maxContextSize: limit.context
@@ -334420,6 +334567,32 @@ function backfillCatalogLimits(input) {
 	return touched;
 }
 /**
+* Fills real context windows onto every configured built-in provider's aliases.
+*
+* Runs after a provider refresh: `/models` answers with ids and no limits, so a
+* refresh that preserves an alias keeps whatever it already held — including a
+* flat default written by an older release. Without this pass a config written
+* before real values were resolved never converges on them, and the user has to
+* re-add the provider to see a real context window.
+*
+* Providers with no `catalogProvider`, and an unreachable catalog, are no-ops.
+*/
+async function backfillBuiltInContextWindows(input) {
+	let touched = false;
+	for (const providerId of input.providerIds) {
+		const catalogProvider = getBuiltInProvider(providerId)?.catalogProvider;
+		if (catalogProvider === void 0) continue;
+		const limits = await fetchCatalogLimits(catalogProvider, input.fetchImpl ?? fetch);
+		if (limits.size === 0) continue;
+		if (backfillCatalogLimits({
+			models: input.models,
+			providerId,
+			limits
+		})) touched = true;
+	}
+	return touched;
+}
+/**
 * Drops free-tier aliases for a provider that the free endpoint no longer
 * offers.
 *
@@ -334445,183 +334618,7 @@ function pruneRetiredFreeTierAliases(input) {
 var init_builtin_model_import = __esmMin((() => {
 	init_src$5();
 	init_catalog_fetch();
-}));
-//#endregion
-//#region src/utils/built-in-providers.ts
-function getBuiltInProvider(id) {
-	const needle = id.trim().toLowerCase();
-	return BUILT_IN_PROVIDERS.find((p) => p.id === needle);
-}
-var OPENCODE_FREE_MODEL_PATTERNS, OPENCODE_FREE_TIER, BUILT_IN_PROVIDERS;
-var init_built_in_providers = __esmMin((() => {
-	OPENCODE_FREE_MODEL_PATTERNS = [
-		"*-free",
-		"*-contributor-free",
-		"big-pickle"
-	];
-	OPENCODE_FREE_TIER = {
-		baseUrl: "https://opencode.ai/inference/v1",
-		match: OPENCODE_FREE_MODEL_PATTERNS,
-		apiKey: "public"
-	};
-	BUILT_IN_PROVIDERS = [
-		{
-			id: "cline",
-			name: "Cline",
-			wire: "openai",
-			baseUrl: "https://api.cline.bot/api/v1",
-			description: "One key for Anthropic, OpenAI, Google and more",
-			consoleUrl: "https://app.cline.bot"
-		},
-		{
-			id: "openrouter",
-			name: "OpenRouter",
-			wire: "openai",
-			baseUrl: "https://openrouter.ai/api/v1",
-			description: "One key for 400+ models across many providers",
-			consoleUrl: "https://openrouter.ai/keys"
-		},
-		{
-			id: "opencode-zen",
-			name: "OpenCode Zen",
-			wire: "openai",
-			baseUrl: "https://opencode.ai/zen/v1",
-			description: "Curated models tested by the OpenCode team",
-			consoleUrl: "https://opencode.ai/zen",
-			protocolOverrides: { "claude-*": "anthropic" },
-			excludedModels: OPENCODE_FREE_MODEL_PATTERNS,
-			freeTier: [OPENCODE_FREE_TIER],
-			catalogProvider: "opencode"
-		},
-		{
-			id: "opencode-go",
-			name: "OpenCode Go",
-			wire: "openai",
-			baseUrl: "https://opencode.ai/zen/go/v1",
-			description: "OpenCode Zen models on the Go plan",
-			consoleUrl: "https://opencode.ai/zen",
-			excludedModels: OPENCODE_FREE_MODEL_PATTERNS,
-			freeTier: [OPENCODE_FREE_TIER],
-			catalogProvider: "opencode"
-		},
-		{
-			id: "nvidia",
-			name: "NVIDIA",
-			wire: "openai",
-			baseUrl: "https://integrate.api.nvidia.com/v1",
-			description: "NVIDIA-hosted open models, free developer tier",
-			consoleUrl: "https://build.nvidia.com",
-			keyHint: "nvapi-"
-		},
-		{
-			id: "nara",
-			name: "NaraRouter",
-			wire: "openai",
-			baseUrl: "https://router.bynara.id/v1",
-			description: "Affordable multi-model gateway",
-			consoleUrl: "https://router.bynara.id",
-			keyHint: "sk-nry-"
-		},
-		{
-			id: "tokenharbor",
-			name: "Token Harbor",
-			wire: "openai",
-			baseUrl: "https://tokenharbor.ai/v1",
-			description: "One universal key for many AI providers",
-			consoleUrl: "https://tokenharbor.ai/dashboard/api-keys",
-			keyHint: "thk_live_"
-		},
-		{
-			id: "openai",
-			name: "OpenAI",
-			wire: "openai",
-			baseUrl: "https://api.openai.com/v1",
-			description: "GPT, o-series and Codex models",
-			consoleUrl: "https://platform.openai.com/api-keys",
-			keyHint: "sk-"
-		},
-		{
-			id: "anthropic",
-			name: "Anthropic",
-			wire: "anthropic",
-			baseUrl: "https://api.anthropic.com",
-			description: "Claude models",
-			consoleUrl: "https://console.anthropic.com/settings/keys",
-			keyHint: "sk-ant-",
-			authStyle: "x-api-key"
-		},
-		{
-			id: "gemini",
-			name: "Google Gemini",
-			wire: "google-genai",
-			baseUrl: "https://generativelanguage.googleapis.com/v1beta",
-			description: "Gemini models",
-			consoleUrl: "https://aistudio.google.com/apikey",
-			keyHint: "AIza",
-			authStyle: "x-goog-api-key"
-		},
-		{
-			id: "grok",
-			name: "xAI Grok",
-			wire: "openai",
-			baseUrl: "https://api.x.ai/v1",
-			description: "Grok models",
-			consoleUrl: "https://console.x.ai",
-			keyHint: "xai-"
-		},
-		{
-			id: "groq",
-			name: "Groq",
-			wire: "openai",
-			baseUrl: "https://api.groq.com/openai/v1",
-			description: "Fast open models on Groq hardware",
-			consoleUrl: "https://console.groq.com/keys",
-			keyHint: "gsk_"
-		},
-		{
-			id: "qwen",
-			name: "Qwen",
-			wire: "openai",
-			baseUrl: "https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
-			description: "Alibaba Qwen models",
-			consoleUrl: "https://bailian.console.alibabacloud.com/",
-			keyHint: "sk-"
-		},
-		{
-			id: "minimax",
-			name: "MiniMax",
-			wire: "openai",
-			baseUrl: "https://api.minimax.io/v1",
-			description: "MiniMax models",
-			consoleUrl: "https://platform.minimax.io/user-center/basic-information/interface-key"
-		},
-		{
-			id: "deepseek",
-			name: "DeepSeek",
-			wire: "openai",
-			baseUrl: "https://api.deepseek.com",
-			description: "DeepSeek chat and reasoning models",
-			consoleUrl: "https://platform.deepseek.com/api_keys",
-			keyHint: "sk-"
-		},
-		{
-			id: "mistral",
-			name: "Mistral",
-			wire: "openai",
-			baseUrl: "https://api.mistral.ai/v1",
-			description: "Mistral and Magistral models",
-			consoleUrl: "https://console.mistral.ai/api-keys"
-		},
-		{
-			id: "huggingface",
-			name: "Hugging Face",
-			wire: "openai",
-			baseUrl: "https://router.huggingface.co/v1",
-			description: "Open models via the Hugging Face router",
-			consoleUrl: "https://huggingface.co/settings/tokens",
-			keyHint: "hf_"
-		}
-	];
+	init_built_in_providers();
 }));
 //#endregion
 //#region src/utils/free-tier-sync.ts
@@ -334702,14 +334699,34 @@ async function refreshAllProviderModels(host, options = {}) {
 		resolveOAuthToken: (providerName, oauthRef) => host.resolveOAuthToken(providerName, oauthRef),
 		userAgent: host.userAgent
 	}, options);
+	await backfillContextWindows(host, options.providerId);
 	await syncConfiguredFreeTiers({
 		getConfig: () => host.getConfig(),
 		setConfig: (patch) => host.setConfig(patch)
 	}, options.providerId);
 	return result;
 }
+/**
+* Upgrades aliases that still carry a flat default, or no declared window at
+* all, to the real context window from the models.dev catalog.
+*
+* A refresh merges the endpoint's own fields over each alias, and `/models`
+* declares no limits — so a value written by an older release survives the
+* refresh unchanged forever. Only re-running the catalog read converges it,
+* which is why this rides along with every refresh rather than only at add.
+*/
+async function backfillContextWindows(host, providerId) {
+	const config = await host.getConfig();
+	const models = { ...config.models };
+	const ids = Object.keys(config.providers ?? {});
+	if (await backfillBuiltInContextWindows({
+		models,
+		providerIds: providerId === void 0 ? ids : ids.filter((id) => id === providerId)
+	})) await host.setConfig({ models });
+}
 var init_refresh_providers = __esmMin((() => {
 	init_src$14();
+	init_builtin_model_import();
 	init_free_tier_sync();
 }));
 //#endregion
@@ -408749,15 +408766,15 @@ var init_thinking_config = __esmMin((() => {}));
 //#endregion
 //#region src/utils/usage/usage-format.ts
 /**
-* Format a token count in 1024-based units: context sizes are powers of
-* two, so 262144 reads as "256k", not "262.1k". k values at or above
-* 100 are rounded to whole numbers ("977k").
+* Format a token count in the 1000-based units vendors declare contexts with,
+* so a 1,000,000-token window reads "1M" rather than "977k". A value that
+* rounds up into the next unit is promoted to it, so 999,999 never reads "1000k".
 */
 function formatTokenCount(n) {
 	if (!Number.isFinite(n) || n < 0) return "0";
-	if (n >= 1024 * 1024) return `${trimDecimal(n / (1024 * 1024))}M`;
-	if (n >= 1024) {
-		const k = n / 1024;
+	if (n >= 1e3 && Math.round(n / 1e3) >= 1e3) return `${trimDecimal(n / 1e6)}M`;
+	if (n >= 1e3) {
+		const k = n / 1e3;
 		return `${k >= 100 ? Math.round(k) : trimDecimal(k)}k`;
 	}
 	return String(n);
@@ -409639,15 +409656,15 @@ function buildUsageReportLines(options) {
 	const value = (text) => currentTheme.fg("text", text);
 	const muted = (text) => currentTheme.fg("textDim", text);
 	const lines = [accent("Session usage"), ...buildSessionUsageSection(options.sessionUsage, options.sessionUsageError, value, muted)];
+	lines.push("");
+	lines.push(accent("Context window"));
 	if (options.maxContextTokens > 0) {
 		const ratio = safeUsageRatio(options.contextUsage);
 		const bar = renderProgressBar(ratio, 20);
 		const pct = `${String(usagePercent(options.contextTokens, options.maxContextTokens))}%`;
 		const barColoured = currentTheme.fg(severityColor(ratioSeverity(ratio)), bar);
-		lines.push("");
-		lines.push(accent("Context window"));
 		lines.push(`  ${barColoured}  ${value(pct.padStart(6, " "))}  ` + muted(`(${formatTokenCount(options.contextTokens)} / ${formatTokenCount(options.maxContextTokens)})`));
-	}
+	} else lines.push(`  ${value(formatTokenCount(options.contextTokens))} used  ` + muted("(no context window declared — set one with /context)"));
 	const managedSection = buildManagedUsageReportLines({
 		managedUsage: options.managedUsage,
 		managedUsageError: options.managedUsageError
@@ -409795,7 +409812,7 @@ function buildStatusReportLines(options) {
 		const bar = renderProgressBar(safeRatio, 20);
 		const barColoured = currentTheme.fg(severityToken(ratioSeverity(safeRatio)), bar);
 		lines.push(`  ${barColoured}  ${value(`${String(usagePercent(tokens, maxTokens))}%`.padStart(6, " "))}  ` + muted(`(${formatTokenCount(tokens)} / ${formatTokenCount(maxTokens)})`));
-	} else lines.push(`  ${muted("No context window data available.")}`);
+	} else lines.push(`  ${value(formatTokenCount(tokens))} used  ` + muted("(no context window declared — set one with /context)"));
 	const managedSection = buildManagedUsageReportLines({
 		managedUsage: options.managedUsage,
 		managedUsageError: options.managedUsageError
@@ -410717,10 +410734,13 @@ function parseContextSize(raw) {
 	if (!Number.isSafeInteger(value) || value < 1) return void 0;
 	return value;
 }
+/**
+* The same 1000-based reading the footer and `/usage` use, so one window is
+* written identically wherever it appears — and a value that is not a round
+* thousand (1,048,576) reads as a unit instead of "1048.576K".
+*/
 function formatContextSize(tokens) {
-	if (tokens >= 1e6 && tokens % 1e5 === 0) return `${String(tokens / 1e6)}M tokens`;
-	if (tokens >= 1e3) return `${String(tokens / 1e3)}K tokens`;
-	return `${String(tokens)} tokens`;
+	return `${formatTokenCount(tokens)} tokens`;
 }
 async function handleContextCommand(host, args) {
 	const trimmed = args.trim();
@@ -410823,6 +410843,7 @@ async function applyContextSize(host, alias, raw) {
 var CONTEXT_SIZE_PATTERN, SIZE_MULTIPLIERS;
 var init_context = __esmMin((() => {
 	init_src$5();
+	init_usage_format();
 	init_api_key_input_dialog();
 	init_choice_picker();
 	init_model_selector();
@@ -483927,20 +483948,19 @@ async function snapshotFreeTierAliases(host, providerId) {
 * `importBuiltInExtras`, so both add paths leave identical config.
 */
 async function routeFreeTierAliases(host, target) {
-	for (const source of target.freeTier ?? []) {
-		const limits = target.catalogProvider === void 0 ? /* @__PURE__ */ new Map() : await fetchCatalogLimits(target.catalogProvider);
-		let config = await host.harness.getConfig({ reload: true });
-		const models = { ...config.models };
-		const changed = backfillCatalogLimits({
+	const limits = target.catalogProvider === void 0 ? /* @__PURE__ */ new Map() : await fetchCatalogLimits(target.catalogProvider);
+	if (limits.size > 0) {
+		const models = { ...(await host.harness.getConfig({ reload: true })).models };
+		if (backfillCatalogLimits({
 			models,
 			providerId: target.providerId,
 			limits
-		});
+		})) await host.harness.setConfig({ models });
+	}
+	for (const source of target.freeTier ?? []) {
+		let config = await host.harness.getConfig({ reload: true });
+		const models = { ...config.models };
 		const { ids, failure } = await fetchFreeTierModelIds(source);
-		if (changed) {
-			await host.harness.setConfig({ models });
-			config = await host.harness.getConfig({ reload: true });
-		}
 		if (ids.length === 0) {
 			host.showError(describeFreeTierFailure(source.baseUrl, failure));
 			const preserved = target.previousFreeAliases ?? {};
@@ -498195,7 +498215,7 @@ var AuthFlowController = class {
 			availableModels,
 			availableProviders,
 			model: defaultModel,
-			maxContextTokens: selected.maxContextSize
+			maxContextTokens: selected.overrides?.maxContextSize ?? selected.maxContextSize ?? 0
 		};
 		host.setAppState(appStatePatch);
 		return activated;
@@ -514530,7 +514550,8 @@ var OmnicodeTUI = class {
 		if (startupModel !== void 0) {
 			patch.model = startupModel;
 			const selected = config.models?.[startupModel];
-			if (selected?.maxContextSize !== void 0) patch.maxContextTokens = selected.maxContextSize;
+			const declaredWindow = selected?.overrides?.maxContextSize ?? selected?.maxContextSize;
+			if (declaredWindow !== void 0) patch.maxContextTokens = declaredWindow;
 		} else {
 			patch.model = "";
 			patch.maxContextTokens = 0;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hilbras/omnicode",
-  "version": "1.2.2",
+  "version": "1.2.3",
   "description": "The Starting Point for Next-Gen Agents",
   "license": "MIT",
   "author": "Hilbras",