@vib-rato/ai 0.16.0

package/src/model-pricing.d.ts

@@ -0,0 +1,3 @@
+import type { Api, Model, ModelCost } from "./types";
+export declare function getOpenAIModelCost<TApi extends Api>(model: Model<TApi>, inputTokens: number): ModelCost | undefined;
+export declare function applyOpenAIModelPricing<TApi extends Api>(model: Model<TApi>): void;

package/src/model-pricing.ts

@@ -0,0 +1,68 @@
+import type { Api, LongContextPricing, Model, ModelCost } from "./types";
+
+interface TieredPricing {
+	cost: ModelCost;
+	longContextPricing: LongContextPricing;
+}
+
+const LONG_CONTEXT_THRESHOLD = 272_000;
+
+const GPT_5_6_SOL_PRICING: TieredPricing = {
+	cost: { input: 5, output: 30, cacheRead: 0.5, cacheWrite: 6.25 },
+	longContextPricing: {
+		threshold: LONG_CONTEXT_THRESHOLD,
+		cost: { input: 10, output: 45, cacheRead: 1, cacheWrite: 12.5 },
+	},
+};
+
+// OpenAI Standard pricing: https://developers.openai.com/api/docs/pricing
+const OPENAI_GPT_5_6_PRICING: ReadonlyMap<string, TieredPricing> = new Map([
+	["gpt-5.6", GPT_5_6_SOL_PRICING],
+	["gpt-5.6-sol", GPT_5_6_SOL_PRICING],
+	[
+		"gpt-5.6-terra",
+		{
+			cost: { input: 2, output: 12, cacheRead: 0.2, cacheWrite: 2.5 },
+			longContextPricing: {
+				threshold: LONG_CONTEXT_THRESHOLD,
+				cost: { input: 4, output: 18, cacheRead: 0.4, cacheWrite: 5 },
+			},
+		},
+	],
+	[
+		"gpt-5.6-luna",
+		{
+			cost: { input: 0.2, output: 1.2, cacheRead: 0.02, cacheWrite: 0.25 },
+			longContextPricing: {
+				threshold: LONG_CONTEXT_THRESHOLD,
+				cost: { input: 0.4, output: 1.8, cacheRead: 0.04, cacheWrite: 0.5 },
+			},
+		},
+	],
+]);
+
+export function getOpenAIModelCost<TApi extends Api>(model: Model<TApi>, inputTokens: number): ModelCost | undefined {
+	if (model.provider !== "openai" && model.provider !== "openai-codex") {
+		return undefined;
+	}
+	const pricing = OPENAI_GPT_5_6_PRICING.get(model.id);
+	if (!pricing) {
+		return undefined;
+	}
+	return inputTokens > pricing.longContextPricing.threshold ? pricing.longContextPricing.cost : pricing.cost;
+}
+
+export function applyOpenAIModelPricing<TApi extends Api>(model: Model<TApi>): void {
+	if (model.provider !== "openai" && model.provider !== "openai-codex") {
+		return;
+	}
+	const pricing = OPENAI_GPT_5_6_PRICING.get(model.id);
+	if (!pricing) {
+		return;
+	}
+	model.cost = { ...pricing.cost };
+	model.longContextPricing = {
+		threshold: pricing.longContextPricing.threshold,
+		cost: { ...pricing.longContextPricing.cost },
+	};
+}

package/src/model-retirements.d.ts

@@ -0,0 +1,6 @@
+export declare const RETIRED_MODEL_KEYS: readonly ["google-antigravity/gemini-3.1-pro-high", "google-antigravity/gemini-3.7-flash-high", "google-antigravity/gemini-3.7-flash-low", "google-antigravity/gemini-3.7-flash-medium"];
+export declare function isRetiredModelKey(provider: string, modelId: string): boolean;
+export declare function isRetiredModel(model: {
+    provider: string;
+    id: string;
+}): boolean;

package/src/model-retirements.ts

@@ -0,0 +1,19 @@
+// Retired from advertised catalogs because Cloud Code Assist rejects live calls.
+// Keep the callable alternatives visible: gemini-3.1-pro-low:high and
+// gemini-3.7-flash-tiered.
+export const RETIRED_MODEL_KEYS = [
+	"google-antigravity/gemini-3.1-pro-high",
+	"google-antigravity/gemini-3.7-flash-high",
+	"google-antigravity/gemini-3.7-flash-low",
+	"google-antigravity/gemini-3.7-flash-medium",
+] as const;
+
+const RETIRED_MODEL_KEY_SET = new Set<string>(RETIRED_MODEL_KEYS);
+
+export function isRetiredModelKey(provider: string, modelId: string): boolean {
+	return RETIRED_MODEL_KEY_SET.has(`${provider}/${modelId}`);
+}
+
+export function isRetiredModel(model: { provider: string; id: string }): boolean {
+	return isRetiredModelKey(model.provider, model.id);
+}

package/src/model-thinking.d.ts

@@ -0,0 +1,100 @@
+import type { Api, Model as ApiModel } from "./types";
+/** User-facing thinking levels, ordered least to most intensive. */
+export declare const enum Effort {
+    Minimal = "minimal",
+    Low = "low",
+    Medium = "medium",
+    High = "high",
+    XHigh = "xhigh",
+    Max = "max"
+}
+export declare const THINKING_EFFORTS: readonly Effort[];
+/**
+ * Static fallback model injected when Cloudflare AI Gateway discovery
+ * returns no results. Ensures the provider always has at least one usable
+ * model entry in the catalog.
+ */
+export declare const CLOUDFLARE_FALLBACK_MODEL: ApiModel<"anthropic-messages">;
+export declare function isGroqCompoundReasoningUnsupported(model: Pick<ApiModel<Api>, "provider" | "id">): boolean;
+/**
+ * Returns a copy of the model with canonical thinking metadata attached.
+ *
+ * This helper belongs to catalog enrichment only. Runtime consumers should
+ * trust `model.thinking` and avoid inferring capabilities on demand.
+ */
+export declare function enrichModelThinking<TApi extends Api>(model: ApiModel<TApi>): ApiModel<TApi>;
+/**
+ * Returns a copy of the model with thinking metadata recomputed from the
+ * canonical rules, replacing any existing `thinking`.
+ */
+export declare function refreshModelThinking<TApi extends Api>(model: ApiModel<TApi>): ApiModel<TApi>;
+/**
+ * Returns whether the configured transport has an audited user-facing reasoning control.
+ *
+ * Custom OpenAI-compatible endpoints fail closed: declaring a model as reasoning-capable
+ * is not enough to prove that the proxy accepts OpenAI reasoning parameters. Unknown
+ * endpoints must opt in with `compat.supportsReasoningEffort: true`; providers using a
+ * non-OpenAI request shape must also declare `compat.thinkingFormat`. Bundled providers
+ * remain governed by their catalog and compatibility metadata.
+ */
+export declare function modelSupportsReasoningControl<TApi extends Api>(model: ApiModel<TApi>, resolvedBaseUrl?: string): boolean;
+/**
+ * Apply upstream metadata corrections to a mutable array of models.
+ *
+ * Each model is first normalized through `refreshModelThinking()` so generated
+ * catalogs keep canonical thinking metadata and policy fixes in one pass.
+ */
+export declare function applyGeneratedModelPolicies(models: ApiModel<Api>[]): void;
+/**
+ * Link OpenAI model variants to their context promotion targets.
+ *
+ * When a model's context is exhausted, the agent can promote to a sibling
+ * model with a larger context window on the same provider:
+ * - `OpenAI code backend-spark` variants promote to `gpt-5.5`.
+ *
+ * `gpt-5.5` itself is a 1M-context model and is not demoted to `gpt-5.4`
+ * (which has a smaller window), so it has no promotion target.
+ */
+export declare function linkOpenAIPromotionTargets(models: ApiModel<Api>[]): void;
+/**
+ * Returns the supported thinking efforts declared on the model metadata.
+ *
+ * Catalog enrichment is responsible for normalizing bundled model metadata up front.
+ * Runtime callers must treat explicit `model.thinking` on custom models as authoritative
+ * so proxy-specific overrides from `models.yml` survive request construction.
+ *
+ * @throws Error when a reasoning-capable model is missing thinking metadata
+ */
+export declare function getSupportedEfforts<TApi extends Api>(model: ApiModel<TApi>): readonly Effort[];
+/**
+ * Clamps a requested thinking level against explicit model metadata.
+ *
+ * Non-reasoning models always resolve to `undefined`.
+ */
+export declare function clampThinkingLevelForModel<TApi extends Api>(model: ApiModel<TApi> | undefined, requested: Effort | undefined): Effort | undefined;
+export declare function requireSupportedEffort<TApi extends Api>(model: ApiModel<TApi>, effort: Effort): Effort;
+/** Maps a normalized thinking effort to Google's `thinkingLevel` enum values. */
+export declare function mapEffortToGoogleThinkingLevel<TApi extends Api>(model: ApiModel<TApi>, effort: Effort): "MINIMAL" | "LOW" | "MEDIUM" | "HIGH";
+/** Maps a normalized thinking effort to Anthropic adaptive effort values. */
+export declare function mapEffortToAnthropicAdaptiveEffort<TApi extends Api>(model: ApiModel<TApi>, effort: Effort): "low" | "medium" | "high" | "xhigh" | "max";
+/**
+ * Returns true for Anthropic models with Opus 4.7 API restrictions:
+ * - Sampling parameters (temperature/top_p/top_k) return 400 error
+ * - Thinking content is omitted by default (needs display: "summarized")
+ */
+export declare function hasOpus47ApiRestrictions(modelId: string): boolean;
+/**
+ * Adaptive thinking `display` is supported starting with Anthropic Opus 4.7.
+ * Older adaptive-thinking models (Opus 4.6, Sonnet 4.6+) reject the field.
+ * Fable (5+) postdates Opus 4.7, accepts `display`, and defaults it to
+ * "omitted" — thinking tokens are billed but no content streams back — so it
+ * must opt in like Opus 4.7+ (issue #2791).
+ *
+ * Shares `hasOpus47ApiRestrictions` version parsing on purpose: the two
+ * predicates describe the same API generation, and a private `claude-opus-(\d+)-(\d+)`
+ * regex silently disagreed with it for single-component aliases (`claude-opus-5`
+ * matched nothing while `claude-opus-5-20260101` matched), so the same model sent
+ * a different thinking shape and beta set depending on which id string was used.
+ * Bedrock region/inference-profile prefixes are handled by the canonical parser.
+ */
+export declare function supportsAnthropicAdaptiveThinkingDisplay(modelId: string): boolean;