@broberg/ai-sdk 0.28.0 → 0.30.0

package/dist/{chunk-LUPUAAHZ.js → chunk-LKVCPMVI.js}

@@ -73,31 +73,34 @@ var PRICING = {
   // $0, which is worse than no tracking (a confident wrong number).
   "vertex:gemini-2.5-flash": { inputPer1M: 0.3, outputPer1M: 2.5, version: V },
   "vertex:gemini-2.5-flash-lite": { inputPer1M: 0.1, outputPer1M: 0.4, version: "2026-06-04-or-xref" },
+  // Cached prompt tokens bill at 10% of the input rate (F039, measured 2026-08-27:
+  // an 8,810-token prefix reported 8,784 cached on the second call WITH a
+  // prompt_cache_key, and 0 without one at every size up to 57k).
   // Mistral (direct, La Plateforme). Official prices from mistral.ai/pricing
   // (2026-06-04, per Christian's CD report). EU/Paris-hosted — the designated
   // GDPR-safe provider for client/personal-data workloads (see F015). NB:
   // medium-3.5 is the premium "Vibe" coding tier ($1.5/$7.5); Large 3 ($0.5/$1.5)
   // is the cheaper frontier general-purpose model despite the higher number.
-  "mistral:mistral-large-latest": { inputPer1M: 0.5, outputPer1M: 1.5, version: MS },
-  "mistral:mistral-large-2512": { inputPer1M: 0.5, outputPer1M: 1.5, version: MS },
-  "mistral:mistral-medium-latest": { inputPer1M: 1.5, outputPer1M: 7.5, version: MS },
-  "mistral:mistral-medium-3.5": { inputPer1M: 1.5, outputPer1M: 7.5, version: MS },
-  "mistral:mistral-medium-3": { inputPer1M: 0.4, outputPer1M: 2, version: "2026-06-04-or-xref" },
-  "mistral:mistral-small-latest": { inputPer1M: 0.1, outputPer1M: 0.3, version: MS },
-  "mistral:mistral-small-2603": { inputPer1M: 0.1, outputPer1M: 0.3, version: MS },
-  "mistral:ministral-3b-latest": { inputPer1M: 0.1, outputPer1M: 0.1, version: MS },
-  "mistral:ministral-8b-latest": { inputPer1M: 0.15, outputPer1M: 0.15, version: MS },
-  "mistral:ministral-14b-latest": { inputPer1M: 0.2, outputPer1M: 0.2, version: MS },
-  "mistral:magistral-medium-latest": { inputPer1M: 2, outputPer1M: 5, version: MS },
-  "mistral:magistral-small-latest": { inputPer1M: 0.5, outputPer1M: 1.5, version: MS },
-  "mistral:devstral-latest": { inputPer1M: 0.4, outputPer1M: 2, version: MS },
-  "mistral:codestral-latest": { inputPer1M: 0.3, outputPer1M: 0.9, version: MS },
-  "mistral:open-mistral-nemo": { inputPer1M: 0.15, outputPer1M: 0.15, version: MS },
+  "mistral:mistral-large-latest": { inputPer1M: 0.5, cacheReadPer1M: 0.05, outputPer1M: 1.5, version: MS },
+  "mistral:mistral-large-2512": { inputPer1M: 0.5, cacheReadPer1M: 0.05, outputPer1M: 1.5, version: MS },
+  "mistral:mistral-medium-latest": { inputPer1M: 1.5, cacheReadPer1M: 0.15, outputPer1M: 7.5, version: MS },
+  "mistral:mistral-medium-3.5": { inputPer1M: 1.5, cacheReadPer1M: 0.15, outputPer1M: 7.5, version: MS },
+  "mistral:mistral-medium-3": { inputPer1M: 0.4, cacheReadPer1M: 0.04, outputPer1M: 2, version: "2026-06-04-or-xref" },
+  "mistral:mistral-small-latest": { inputPer1M: 0.1, cacheReadPer1M: 0.01, outputPer1M: 0.3, version: MS },
+  "mistral:mistral-small-2603": { inputPer1M: 0.1, cacheReadPer1M: 0.01, outputPer1M: 0.3, version: MS },
+  "mistral:ministral-3b-latest": { inputPer1M: 0.1, cacheReadPer1M: 0.01, outputPer1M: 0.1, version: MS },
+  "mistral:ministral-8b-latest": { inputPer1M: 0.15, cacheReadPer1M: 0.015, outputPer1M: 0.15, version: MS },
+  "mistral:ministral-14b-latest": { inputPer1M: 0.2, cacheReadPer1M: 0.02, outputPer1M: 0.2, version: MS },
+  "mistral:magistral-medium-latest": { inputPer1M: 2, cacheReadPer1M: 0.2, outputPer1M: 5, version: MS },
+  "mistral:magistral-small-latest": { inputPer1M: 0.5, cacheReadPer1M: 0.05, outputPer1M: 1.5, version: MS },
+  "mistral:devstral-latest": { inputPer1M: 0.4, cacheReadPer1M: 0.04, outputPer1M: 2, version: MS },
+  "mistral:codestral-latest": { inputPer1M: 0.3, cacheReadPer1M: 0.03, outputPer1M: 0.9, version: MS },
+  "mistral:open-mistral-nemo": { inputPer1M: 0.15, cacheReadPer1M: 0.015, outputPer1M: 0.15, version: MS },
   // Moderation (F016.4) — per input token; output 0. (OCR is per-page in the adapter.)
-  "mistral:mistral-moderation-latest": { inputPer1M: 0.1, outputPer1M: 0, version: MS },
+  "mistral:mistral-moderation-latest": { inputPer1M: 0.1, cacheReadPer1M: 0.01, outputPer1M: 0, version: MS },
   // Embeddings (F016.5) — per input token.
-  "mistral:mistral-embed": { inputPer1M: 0.1, outputPer1M: 0, version: MS },
-  "mistral:codestral-embed": { inputPer1M: 0.15, outputPer1M: 0, version: MS }
+  "mistral:mistral-embed": { inputPer1M: 0.1, cacheReadPer1M: 0.01, outputPer1M: 0, version: MS },
+  "mistral:codestral-embed": { inputPer1M: 0.15, cacheReadPer1M: 0.015, outputPer1M: 0, version: MS }
 };
 function getPrice(provider, model) {
   const exact = PRICING[`${provider}:${model}`];
@@ -111,4 +114,4 @@ export {
   PRICING,
   getPrice
 };
-//# sourceMappingURL=chunk-LUPUAAHZ.js.map
+//# sourceMappingURL=chunk-LKVCPMVI.js.map

package/dist/chunk-LKVCPMVI.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/cost/pricing.ts"],"sourcesContent":["// Versioned per-(provider, model) pricing. F3.6 populates the table + adds tests\n// + MiniMax coverage. F3.1 ships the type + lookup with an empty table, so\n// computeCost returns 0 for every model until F3.6 lands (calls still complete).\nexport interface PricingEntry {\n  /** USD per 1M input tokens. */\n  inputPer1M: number;\n  /** USD per 1M output tokens. */\n  outputPer1M: number;\n  /** USD per 1M cache-read tokens (falls back to input rate if unset). */\n  cacheReadPer1M?: number;\n  /** USD per 1M cache-write/creation tokens (falls back to input rate if unset). */\n  cacheWritePer1M?: number;\n  /** Pricing snapshot version (date or tag) so stale entries are detectable. */\n  version: string;\n}\n\n// USD per 1M tokens. Anthropic cache multipliers follow the standard model:\n// cache-read ≈ 0.1× input, cache-write ≈ 1.25× input. Verified against the\n// pricing tables in cms (packages/cms-ai/src/providers) + trail (model-lab).\n// MiniMax M2.7 is an estimate pending confirmation against OpenRouter's live\n// price page — flagged in its version string.\nconst V = \"2026-06-02\";\n// Mistral prices come straight from mistral.ai/pricing (per Christian's CD report).\nconst MS = \"2026-06-04-mistral.ai\";\n\n/** Keyed `${provider}:${model}`. Exported so the catalogue-research job (F014)\n *  can enumerate every priced entry and diff it against the live provider lists. */\nexport const PRICING: Record<string, PricingEntry> = {\n  // Anthropic (direct API). DEFAULT_TIER_MAP: fast/cheap=haiku, smart/vision=sonnet, powerful=opus.\n  \"anthropic:claude-haiku-4-5\": {\n    inputPer1M: 0.8,\n    outputPer1M: 4.0,\n    cacheReadPer1M: 0.08,\n    cacheWritePer1M: 1.0,\n    version: V,\n  },\n  \"anthropic:claude-sonnet-4-6\": {\n    inputPer1M: 3.0,\n    outputPer1M: 15.0,\n    cacheReadPer1M: 0.3,\n    cacheWritePer1M: 3.75,\n    version: V,\n  },\n  \"anthropic:claude-opus-4-8\": {\n    inputPer1M: 15.0,\n    outputPer1M: 75.0,\n    cacheReadPer1M: 1.5,\n    cacheWritePer1M: 18.75,\n    version: V,\n  },\n\n  // OpenAI. embedding default tier = text-embedding-3-small (no output tokens).\n  \"openai:text-embedding-3-small\": { inputPer1M: 0.02, outputPer1M: 0, version: V },\n  \"openai:text-embedding-3-large\": { inputPer1M: 0.13, outputPer1M: 0, version: V },\n  \"openai:gpt-4o\": { inputPer1M: 2.5, outputPer1M: 10.0, version: V },\n  \"openai:gpt-4o-mini\": { inputPer1M: 0.15, outputPer1M: 0.6, version: V },\n  // Whisper is priced per minute, not per token — not representable here; transcribe\n  // (F5.6) computes its own cost. Listed as 0 so token-based compute never charges it.\n  \"openai:whisper-1\": { inputPer1M: 0, outputPer1M: 0, version: V },\n\n  // OpenRouter (meta-router — model slugs include the upstream vendor). Slugs use\n  // dots (claude-sonnet-4.6) to match OpenRouter's live ids; the dashed forms\n  // never matched a real call. Caught by the F014 catalogue research.\n  \"openrouter:anthropic/claude-sonnet-4.6\": { inputPer1M: 3.0, outputPer1M: 15.0, version: V },\n  // OpenRouter ground-truth $1/$5 — a markup over Anthropic-direct's $0.8/$4\n  // (the `anthropic:` entry above). Was masked while the slug used dashes.\n  \"openrouter:anthropic/claude-haiku-4.5\": { inputPer1M: 1.0, outputPer1M: 5.0, version: \"2026-06-04\" },\n  \"openrouter:google/gemini-2.5-flash\": { inputPer1M: 0.3, outputPer1M: 2.5, version: V },\n  // Ground-truth from OpenRouter /api/v1/models (was a 0.3 estimate; now 0.279).\n  \"openrouter:minimax/minimax-m2.7\": {\n    inputPer1M: 0.279,\n    outputPer1M: 1.2,\n    version: \"2026-06-04\",\n  },\n  // DeepSeek V4 (CN-hosted — NOT GDPR-safe; non-personal-data workloads only).\n  // On 2026-05-22 DeepSeek made the \"75% off\" promo the permanent official price.\n  // V4-Pro $0.435/$0.87 is ~34x cheaper than GPT-5.5 on output; flash is cheaper\n  // still. Numbers match OpenRouter /api/v1/models 1:1 (no router markup). A strong\n  // cheap route for fleet background work once `claude -p` is API-billed (15 Jun).\n  \"openrouter:deepseek/deepseek-v4-pro\": { inputPer1M: 0.435, outputPer1M: 0.87, version: \"2026-05-22-deepseek-official\" },\n  \"openrouter:deepseek/deepseek-v4-flash\": { inputPer1M: 0.0983, outputPer1M: 0.1966, version: \"2026-05-22-deepseek-official\" },\n  // DeepSeek DIRECT API (provider \"deepseek\", F030 non-PII secondary). Rates from\n  // api-docs.deepseek.com 2026-06-30 ($0.14/$0.28 per 1M; both map to deepseek-v4-flash).\n  // `deepseek-chat` (non-thinking) + `deepseek-reasoner` (thinking) DEPRECATE 2026-07-24.\n  // (The bare `deepseek-v4-flash` basename is already priced via the openrouter entry\n  // above — kept distinct here to avoid a basename collision in the F027 pricing-API.)\n  // Verify against a real key when it lands.\n  \"deepseek:deepseek-chat\": { inputPer1M: 0.14, outputPer1M: 0.28, version: \"2026-06-30-deepseek-direct\" },\n  \"deepseek:deepseek-reasoner\": { inputPer1M: 0.14, outputPer1M: 0.28, version: \"2026-06-30-deepseek-direct\" },\n\n  // Google Gemini (direct). Provider key is \"gemini\" — matches the adapter's\n  // usage.provider + the override.provider callers pass. (Image-gen models are\n  // priced per-image in the adapter, not here.)\n  \"gemini:gemini-2.5-flash\": { inputPer1M: 0.3, outputPer1M: 2.5, version: V },\n  // flash-lite is the default `video` tier (F019) — cheap native video understanding.\n  \"gemini:gemini-2.5-flash-lite\": { inputPer1M: 0.1, outputPer1M: 0.4, version: \"2026-06-04-or-xref\" },\n\n  // Vertex AI (F038) — the EU-resident route to the SAME Gemini models, so Google's\n  // published Gemini token prices apply. Listed separately because cost lookups key on\n  // `provider:model`: without these rows an EU vision/video call would silently log\n  // $0, which is worse than no tracking (a confident wrong number).\n  \"vertex:gemini-2.5-flash\": { inputPer1M: 0.3, outputPer1M: 2.5, version: V },\n  \"vertex:gemini-2.5-flash-lite\": { inputPer1M: 0.1, outputPer1M: 0.4, version: \"2026-06-04-or-xref\" },\n\n  // Cached prompt tokens bill at 10% of the input rate (F039, measured 2026-08-27:\n  // an 8,810-token prefix reported 8,784 cached on the second call WITH a\n  // prompt_cache_key, and 0 without one at every size up to 57k).\n  // Mistral (direct, La Plateforme). Official prices from mistral.ai/pricing\n  // (2026-06-04, per Christian's CD report). EU/Paris-hosted — the designated\n  // GDPR-safe provider for client/personal-data workloads (see F015). NB:\n  // medium-3.5 is the premium \"Vibe\" coding tier ($1.5/$7.5); Large 3 ($0.5/$1.5)\n  // is the cheaper frontier general-purpose model despite the higher number.\n  \"mistral:mistral-large-latest\": { inputPer1M: 0.5, cacheReadPer1M: 0.05, outputPer1M: 1.5, version: MS },\n  \"mistral:mistral-large-2512\": { inputPer1M: 0.5, cacheReadPer1M: 0.05, outputPer1M: 1.5, version: MS },\n  \"mistral:mistral-medium-latest\": { inputPer1M: 1.5, cacheReadPer1M: 0.15, outputPer1M: 7.5, version: MS },\n  \"mistral:mistral-medium-3.5\": { inputPer1M: 1.5, cacheReadPer1M: 0.15, outputPer1M: 7.5, version: MS },\n  \"mistral:mistral-medium-3\": { inputPer1M: 0.4, cacheReadPer1M: 0.04, outputPer1M: 2.0, version: \"2026-06-04-or-xref\" },\n  \"mistral:mistral-small-latest\": { inputPer1M: 0.1, cacheReadPer1M: 0.01, outputPer1M: 0.3, version: MS },\n  \"mistral:mistral-small-2603\": { inputPer1M: 0.1, cacheReadPer1M: 0.01, outputPer1M: 0.3, version: MS },\n  \"mistral:ministral-3b-latest\": { inputPer1M: 0.1, cacheReadPer1M: 0.01, outputPer1M: 0.1, version: MS },\n  \"mistral:ministral-8b-latest\": { inputPer1M: 0.15, cacheReadPer1M: 0.015, outputPer1M: 0.15, version: MS },\n  \"mistral:ministral-14b-latest\": { inputPer1M: 0.2, cacheReadPer1M: 0.02, outputPer1M: 0.2, version: MS },\n  \"mistral:magistral-medium-latest\": { inputPer1M: 2.0, cacheReadPer1M: 0.2, outputPer1M: 5.0, version: MS },\n  \"mistral:magistral-small-latest\": { inputPer1M: 0.5, cacheReadPer1M: 0.05, outputPer1M: 1.5, version: MS },\n  \"mistral:devstral-latest\": { inputPer1M: 0.4, cacheReadPer1M: 0.04, outputPer1M: 2.0, version: MS },\n  \"mistral:codestral-latest\": { inputPer1M: 0.3, cacheReadPer1M: 0.03, outputPer1M: 0.9, version: MS },\n  \"mistral:open-mistral-nemo\": { inputPer1M: 0.15, cacheReadPer1M: 0.015, outputPer1M: 0.15, version: MS },\n  // Moderation (F016.4) — per input token; output 0. (OCR is per-page in the adapter.)\n  \"mistral:mistral-moderation-latest\": { inputPer1M: 0.1, cacheReadPer1M: 0.01, outputPer1M: 0, version: MS },\n  // Embeddings (F016.5) — per input token.\n  \"mistral:mistral-embed\": { inputPer1M: 0.1, cacheReadPer1M: 0.01, outputPer1M: 0, version: MS },\n  \"mistral:codestral-embed\": { inputPer1M: 0.15, cacheReadPer1M: 0.015, outputPer1M: 0, version: MS },\n};\n\nexport function getPrice(provider: string, model: string): PricingEntry | undefined {\n  const exact = PRICING[`${provider}:${model}`];\n  if (exact) return exact;\n  // Providers ship dated model snapshots, e.g. \"claude-haiku-4-5-20251001\".\n  // Strip a trailing -YYYYMMDD and retry the base lookup so a dated variant\n  // prices the same as its base model instead of falling through to 0 — a real\n  // paid call must never be logged as $0 (F012). Covers openrouter slugs too.\n  const base = model.replace(/-\\d{8}$/, \"\");\n  if (base !== model) return PRICING[`${provider}:${base}`];\n  return undefined;\n}\n"],"mappings":";AAqBA,IAAM,IAAI;AAEV,IAAM,KAAK;AAIJ,IAAM,UAAwC;AAAA;AAAA,EAEnD,8BAA8B;AAAA,IAC5B,YAAY;AAAA,IACZ,aAAa;AAAA,IACb,gBAAgB;AAAA,IAChB,iBAAiB;AAAA,IACjB,SAAS;AAAA,EACX;AAAA,EACA,+BAA+B;AAAA,IAC7B,YAAY;AAAA,IACZ,aAAa;AAAA,IACb,gBAAgB;AAAA,IAChB,iBAAiB;AAAA,IACjB,SAAS;AAAA,EACX;AAAA,EACA,6BAA6B;AAAA,IAC3B,YAAY;AAAA,IACZ,aAAa;AAAA,IACb,gBAAgB;AAAA,IAChB,iBAAiB;AAAA,IACjB,SAAS;AAAA,EACX;AAAA;AAAA,EAGA,iCAAiC,EAAE,YAAY,MAAM,aAAa,GAAG,SAAS,EAAE;AAAA,EAChF,iCAAiC,EAAE,YAAY,MAAM,aAAa,GAAG,SAAS,EAAE;AAAA,EAChF,iBAAiB,EAAE,YAAY,KAAK,aAAa,IAAM,SAAS,EAAE;AAAA,EAClE,sBAAsB,EAAE,YAAY,MAAM,aAAa,KAAK,SAAS,EAAE;AAAA;AAAA;AAAA,EAGvE,oBAAoB,EAAE,YAAY,GAAG,aAAa,GAAG,SAAS,EAAE;AAAA;AAAA;AAAA;AAAA,EAKhE,0CAA0C,EAAE,YAAY,GAAK,aAAa,IAAM,SAAS,EAAE;AAAA;AAAA;AAAA,EAG3F,yCAAyC,EAAE,YAAY,GAAK,aAAa,GAAK,SAAS,aAAa;AAAA,EACpG,sCAAsC,EAAE,YAAY,KAAK,aAAa,KAAK,SAAS,EAAE;AAAA;AAAA,EAEtF,mCAAmC;AAAA,IACjC,YAAY;AAAA,IACZ,aAAa;AAAA,IACb,SAAS;AAAA,EACX;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAMA,uCAAuC,EAAE,YAAY,OAAO,aAAa,MAAM,SAAS,+BAA+B;AAAA,EACvH,yCAAyC,EAAE,YAAY,QAAQ,aAAa,QAAQ,SAAS,+BAA+B;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAO5H,0BAA0B,EAAE,YAAY,MAAM,aAAa,MAAM,SAAS,6BAA6B;AAAA,EACvG,8BAA8B,EAAE,YAAY,MAAM,aAAa,MAAM,SAAS,6BAA6B;AAAA;AAAA;AAAA;AAAA,EAK3G,2BAA2B,EAAE,YAAY,KAAK,aAAa,KAAK,SAAS,EAAE;AAAA;AAAA,EAE3E,gCAAgC,EAAE,YAAY,KAAK,aAAa,KAAK,SAAS,qBAAqB;AAAA;AAAA;AAAA;AAAA;AAAA,EAMnG,2BAA2B,EAAE,YAAY,KAAK,aAAa,KAAK,SAAS,EAAE;AAAA,EAC3E,gCAAgC,EAAE,YAAY,KAAK,aAAa,KAAK,SAAS,qBAAqB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAUnG,gCAAgC,EAAE,YAAY,KAAK,gBAAgB,MAAM,aAAa,KAAK,SAAS,GAAG;AAAA,EACvG,8BAA8B,EAAE,YAAY,KAAK,gBAAgB,MAAM,aAAa,KAAK,SAAS,GAAG;AAAA,EACrG,iCAAiC,EAAE,YAAY,KAAK,gBAAgB,MAAM,aAAa,KAAK,SAAS,GAAG;AAAA,EACxG,8BAA8B,EAAE,YAAY,KAAK,gBAAgB,MAAM,aAAa,KAAK,SAAS,GAAG;AAAA,EACrG,4BAA4B,EAAE,YAAY,KAAK,gBAAgB,MAAM,aAAa,GAAK,SAAS,qBAAqB;AAAA,EACrH,gCAAgC,EAAE,YAAY,KAAK,gBAAgB,MAAM,aAAa,KAAK,SAAS,GAAG;AAAA,EACvG,8BAA8B,EAAE,YAAY,KAAK,gBAAgB,MAAM,aAAa,KAAK,SAAS,GAAG;AAAA,EACrG,+BAA+B,EAAE,YAAY,KAAK,gBAAgB,MAAM,aAAa,KAAK,SAAS,GAAG;AAAA,EACtG,+BAA+B,EAAE,YAAY,MAAM,gBAAgB,OAAO,aAAa,MAAM,SAAS,GAAG;AAAA,EACzG,gCAAgC,EAAE,YAAY,KAAK,gBAAgB,MAAM,aAAa,KAAK,SAAS,GAAG;AAAA,EACvG,mCAAmC,EAAE,YAAY,GAAK,gBAAgB,KAAK,aAAa,GAAK,SAAS,GAAG;AAAA,EACzG,kCAAkC,EAAE,YAAY,KAAK,gBAAgB,MAAM,aAAa,KAAK,SAAS,GAAG;AAAA,EACzG,2BAA2B,EAAE,YAAY,KAAK,gBAAgB,MAAM,aAAa,GAAK,SAAS,GAAG;AAAA,EAClG,4BAA4B,EAAE,YAAY,KAAK,gBAAgB,MAAM,aAAa,KAAK,SAAS,GAAG;AAAA,EACnG,6BAA6B,EAAE,YAAY,MAAM,gBAAgB,OAAO,aAAa,MAAM,SAAS,GAAG;AAAA;AAAA,EAEvG,qCAAqC,EAAE,YAAY,KAAK,gBAAgB,MAAM,aAAa,GAAG,SAAS,GAAG;AAAA;AAAA,EAE1G,yBAAyB,EAAE,YAAY,KAAK,gBAAgB,MAAM,aAAa,GAAG,SAAS,GAAG;AAAA,EAC9F,2BAA2B,EAAE,YAAY,MAAM,gBAAgB,OAAO,aAAa,GAAG,SAAS,GAAG;AACpG;AAEO,SAAS,SAAS,UAAkB,OAAyC;AAClF,QAAM,QAAQ,QAAQ,GAAG,QAAQ,IAAI,KAAK,EAAE;AAC5C,MAAI,MAAO,QAAO;AAKlB,QAAM,OAAO,MAAM,QAAQ,WAAW,EAAE;AACxC,MAAI,SAAS,MAAO,QAAO,QAAQ,GAAG,QAAQ,IAAI,IAAI,EAAE;AACxD,SAAO;AACT;","names":[]}

package/dist/{chunk-IT7HNKLY.js → chunk-V2PD522L.js}

@@ -1,21 +1,56 @@
+// src/routing/tier-map.ts
+var DEFAULT_TIER_MAP = {
+  fast: { provider: "mistral", model: "mistral-small-latest", transport: "http" },
+  smart: { provider: "mistral", model: "mistral-large-latest", transport: "http" },
+  powerful: { provider: "mistral", model: "mistral-large-latest", transport: "http" },
+  cheap: { provider: "mistral", model: "mistral-small-latest", transport: "http" },
+  // Vision: small-latest (vision-capable, cheap EU) is the default; override to
+  // mistral-large-latest for demanding image/spatial/composition work.
+  vision: { provider: "mistral", model: "mistral-small-latest", transport: "http" },
+  // Native video understanding — Gemini leads; flash-lite is the cheap default (F019).
+  // NOT Anthropic → out of the F030 phase-out (its own EU epic if/when needed).
+  video: { provider: "gemini", model: "gemini-2.5-flash-lite", transport: "http" },
+  // NOT Anthropic → out of F030 (EU-embedding migration is its own future epic).
+  embedding: { provider: "openai", model: "text-embedding-3-small", transport: "http" }
+};
+function resolveTier(tier, override, configMap) {
+  const base = configMap?.[tier] ?? DEFAULT_TIER_MAP[tier];
+  return { ...base, ...override };
+}
+
 // src/availability/registry.ts
 var SUSPENDED_FABLE_MYTHOS = "suspended \u2014 US export-control directive (2026-06-12)";
 var DEFAULTS = [
   // ── Anthropic ────────────────────────────────────────────────────────────
-  { id: "claude-haiku-4-5", aliases: ["haiku", "fast"], provider: "anthropic", available: true, status: "available", source: "default" },
-  { id: "claude-sonnet-4-6", aliases: ["sonnet", "smart"], provider: "anthropic", available: true, status: "available", source: "default" },
-  { id: "claude-opus-4-8", aliases: ["opus", "powerful"], provider: "anthropic", available: true, status: "available", source: "default" },
+  { id: "claude-haiku-4-5", aliases: ["haiku"], provider: "anthropic", available: true, status: "available", source: "default" },
+  { id: "claude-sonnet-4-6", aliases: ["sonnet"], provider: "anthropic", available: true, status: "available", source: "default" },
+  { id: "claude-opus-4-8", aliases: ["opus"], provider: "anthropic", available: true, status: "available", source: "default" },
   { id: "claude-fable-5", aliases: ["fable"], provider: "anthropic", available: true, status: "available", source: "default" },
   { id: "claude-mythos-5", aliases: ["mythos"], provider: "anthropic", available: false, status: "suspended", note: SUSPENDED_FABLE_MYTHOS, source: "default" },
   // ── Gemini ───────────────────────────────────────────────────────────────
   { id: "gemini-2.5-flash", aliases: ["gemini-flash"], provider: "gemini", available: true, status: "available", source: "default" },
-  { id: "gemini-2.5-flash-lite", aliases: ["gemini-flash-lite", "video"], provider: "gemini", available: true, status: "available", source: "default" },
+  { id: "gemini-2.5-flash-lite", aliases: ["gemini-flash-lite"], provider: "gemini", available: true, status: "available", source: "default" },
   // ── OpenAI ───────────────────────────────────────────────────────────────
-  { id: "text-embedding-3-small", aliases: ["embedding"], provider: "openai", available: true, status: "available", source: "default" },
+  { id: "text-embedding-3-small", aliases: [], provider: "openai", available: true, status: "available", source: "default" },
   // ── Mistral (EU / GDPR) ──────────────────────────────────────────────────
   { id: "mistral-large-latest", aliases: ["mistral-large"], provider: "mistral", available: true, status: "available", source: "default" },
   { id: "mistral-small-latest", aliases: ["mistral-small"], provider: "mistral", available: true, status: "available", source: "default" }
 ];
+var TIER_ALIAS_CONFLICTS = [];
+function applyTierAliases(entries) {
+  for (const [tier, spec] of Object.entries(DEFAULT_TIER_MAP)) {
+    const owner = entries.find((e) => e.id === spec.model && e.provider === spec.provider);
+    for (const e of entries) {
+      if (e !== owner && e.aliases.includes(tier)) {
+        TIER_ALIAS_CONFLICTS.push(`${tier} is hand-declared on ${e.id} (${e.provider}) but tier ${tier} calls ${spec.model} (${spec.provider})`);
+        e.aliases = e.aliases.filter((a) => a !== tier);
+      }
+    }
+    if (owner && !owner.aliases.includes(tier)) owner.aliases.push(tier);
+  }
+  return entries;
+}
+applyTierAliases(DEFAULTS);
 var OVERLAY = /* @__PURE__ */ new Map();
 var ALIAS_INDEX = /* @__PURE__ */ new Map();
 function seed() {
@@ -83,15 +118,16 @@ var ModelUnavailableError = class extends Error {
 function listModels(opts = {}) {
   return allEntries(opts.provider);
 }
-function isAvailable(requested) {
+function isAvailable(requested, requireKnown = false) {
   const e = getEntry(requested);
-  return e ? e.available : true;
+  if (!e) return !requireKnown;
+  return e.available;
 }
 function resolveModel(requested, opts = {}) {
   const id = canonicalId(requested) ?? requested;
   const entry = getEntry(requested);
   const provider = opts.provider ?? entry?.provider;
-  if (isAvailable(requested)) {
+  if (isAvailable(requested, opts.requireKnown)) {
     return {
       ok: true,
       model: id,
@@ -103,7 +139,7 @@ function resolveModel(requested, opts = {}) {
   }
   const chain = opts.fallback === void 0 ? [] : Array.isArray(opts.fallback) ? opts.fallback : [opts.fallback];
   for (const fb of chain) {
-    if (isAvailable(fb)) {
+    if (isAvailable(fb, opts.requireKnown)) {
       const fbId = canonicalId(fb) ?? fb;
       return {
         ok: false,
@@ -116,8 +152,10 @@ function resolveModel(requested, opts = {}) {
       };
     }
   }
+  const unknownAndGating = !entry && opts.requireKnown;
+  const reason = unknownAndGating ? `${id} is not a model this registry knows \u2014 requireKnown was set, so it is not assumed usable` : entry?.note ?? `${id} is unavailable`;
   if (opts.throwIfUnavailable) {
-    throw new ModelUnavailableError(id, entry?.note, provider);
+    throw new ModelUnavailableError(id, reason, provider);
   }
   return {
     ok: false,
@@ -125,12 +163,14 @@ function resolveModel(requested, opts = {}) {
     requested: id,
     provider,
     fellBack: false,
-    status: entry?.status ?? "suspended",
-    reason: entry?.note ?? `${id} is unavailable`
+    status: entry?.status ?? "unknown",
+    reason
   };
 }
 
 export {
+  DEFAULT_TIER_MAP,
+  resolveTier,
   resetRegistry,
   providerIds,
   setAvailability,
@@ -138,4 +178,4 @@ export {
   listModels,
   resolveModel
 };
-//# sourceMappingURL=chunk-IT7HNKLY.js.map
+//# sourceMappingURL=chunk-V2PD522L.js.map

package/dist/chunk-V2PD522L.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/routing/tier-map.ts","../src/availability/registry.ts","../src/availability/types.ts","../src/availability/resolve.ts"],"sourcesContent":["// Tier routing: a named Tier resolves to a concrete (provider, model, transport).\n// Precedence is per-call override > client config map > built-in defaults.\nimport type { Tier, TierSpec } from \"../types.js\";\n\n/** Built-in defaults. Every entry is overridable via AiConfig.defaults or a\n *  per-call override.\n *\n *  F030 — Anthropic API phase-out: `ANTHROPIC_API_KEY` was globally removed, so the\n *  default cloud route may NOT hit Anthropic Console. `fast`/`smart`/`powerful`/\n *  `vision` now default to **Mistral EU** (Paris-hosted, Schrems II-safe — so every\n *  default text/vision call is GDPR-safe by default). Claude stays reachable as a\n *  NON-default quality fallback for non-PII via `override:{provider:\"openrouter\",\n *  model:\"anthropic/claude-…\"}`. DeepSeek (CN) is the opt-in non-PII secondary\n *  (`provider:\"deepseek\"`), never a default. Magistral (reasoning) / mistral-large\n *  for vision are per-call overrides, not defaults (don't pay the premium on all). */\nexport const DEFAULT_TIER_MAP: Record<Tier, TierSpec> = {\n  fast: { provider: \"mistral\", model: \"mistral-small-latest\", transport: \"http\" },\n  smart: { provider: \"mistral\", model: \"mistral-large-latest\", transport: \"http\" },\n  powerful: { provider: \"mistral\", model: \"mistral-large-latest\", transport: \"http\" },\n  cheap: { provider: \"mistral\", model: \"mistral-small-latest\", transport: \"http\" },\n  // Vision: small-latest (vision-capable, cheap EU) is the default; override to\n  // mistral-large-latest for demanding image/spatial/composition work.\n  vision: { provider: \"mistral\", model: \"mistral-small-latest\", transport: \"http\" },\n  // Native video understanding — Gemini leads; flash-lite is the cheap default (F019).\n  // NOT Anthropic → out of the F030 phase-out (its own EU epic if/when needed).\n  video: { provider: \"gemini\", model: \"gemini-2.5-flash-lite\", transport: \"http\" },\n  // NOT Anthropic → out of F030 (EU-embedding migration is its own future epic).\n  embedding: { provider: \"openai\", model: \"text-embedding-3-small\", transport: \"http\" },\n};\n\n/**\n * Resolve a Tier to a concrete TierSpec.\n *\n * Merge order (later wins): DEFAULT_TIER_MAP < configMap < override.\n * - `configMap` is the client-level AiConfig.defaults (per-tier full specs).\n * - `override` is a per-call Partial<TierSpec> — only the fields it sets win.\n */\nexport function resolveTier(\n  tier: Tier,\n  override?: Partial<TierSpec>,\n  configMap?: Partial<Record<Tier, TierSpec>>,\n): TierSpec {\n  const base = configMap?.[tier] ?? DEFAULT_TIER_MAP[tier];\n  return { ...base, ...override };\n}\n","// F022 — the model-availability registry: the ONE source both resolveModel()\n// (spawn / call path) and listModels() (UI picker) read. A curated default seed\n// (works offline — the durable floor) plus a mutable overlay that\n// refreshAvailability() updates from the live provider list.\n//\n// Scope note: this is a LIVENESS view (is this id alive right now?), not the\n// rich capability/price inventory (that is F017 src/catalogue). We only track\n// ids we want to assert status on; anything not here is fail-open (treated\n// available) so we never block a model we simply do not track.\nimport { DEFAULT_TIER_MAP } from \"../routing/tier-map.js\";\nimport type { AvailabilityStatus, AvailabilitySource, ModelStatus } from \"./types.js\";\n\n/** Internal registry row. `aliases[0]` surfaces as ModelStatus.alias. */\nexport interface RegistryEntry {\n  id: string;\n  aliases: string[];\n  provider: string;\n  available: boolean;\n  status: AvailabilityStatus;\n  note?: string;\n  source: AvailabilitySource;\n}\n\nconst SUSPENDED_FABLE_MYTHOS = \"suspended — US export-control directive (2026-06-12)\";\n\n/** Curated defaults. Aliases here are MODEL-IDENTITY names only (short names for\n *  the model itself). TIER aliases — fast/smart/powerful/cheap/vision/video/\n *  embedding — are NOT written here: they are derived from DEFAULT_TIER_MAP below,\n *  because hand-maintaining them in two places is exactly how they drifted.\n *\n *  They HAD drifted, and it crossed the EU border: this list said `smart` was\n *  claude-sonnet-4-6 (Anthropic, US) for the ~3 months after F030 pointed the\n *  default tiers at Mistral EU, so resolveModel('smart') named a US model while\n *  ai.chat({tier:'smart'}) called an EU one. Anyone using resolveModel to SHOW or\n *  DECIDE where data goes — the obvious use — got the wrong answer. Measured by\n *  fd-sundhed in 0.21.1, still true in 0.28.0, fixed here by deleting the second\n *  list rather than correcting it. */\nconst DEFAULTS: RegistryEntry[] = [\n  // ── Anthropic ────────────────────────────────────────────────────────────\n  { id: \"claude-haiku-4-5\", aliases: [\"haiku\"], provider: \"anthropic\", available: true, status: \"available\", source: \"default\" },\n  { id: \"claude-sonnet-4-6\", aliases: [\"sonnet\"], provider: \"anthropic\", available: true, status: \"available\", source: \"default\" },\n  { id: \"claude-opus-4-8\", aliases: [\"opus\"], provider: \"anthropic\", available: true, status: \"available\", source: \"default\" },\n  { id: \"claude-fable-5\", aliases: [\"fable\"], provider: \"anthropic\", available: true, status: \"available\", source: \"default\" },\n  { id: \"claude-mythos-5\", aliases: [\"mythos\"], provider: \"anthropic\", available: false, status: \"suspended\", note: SUSPENDED_FABLE_MYTHOS, source: \"default\" },\n  // ── Gemini ───────────────────────────────────────────────────────────────\n  { id: \"gemini-2.5-flash\", aliases: [\"gemini-flash\"], provider: \"gemini\", available: true, status: \"available\", source: \"default\" },\n  { id: \"gemini-2.5-flash-lite\", aliases: [\"gemini-flash-lite\"], provider: \"gemini\", available: true, status: \"available\", source: \"default\" },\n  // ── OpenAI ───────────────────────────────────────────────────────────────\n  { id: \"text-embedding-3-small\", aliases: [], provider: \"openai\", available: true, status: \"available\", source: \"default\" },\n  // ── Mistral (EU / GDPR) ──────────────────────────────────────────────────\n  { id: \"mistral-large-latest\", aliases: [\"mistral-large\"], provider: \"mistral\", available: true, status: \"available\", source: \"default\" },\n  { id: \"mistral-small-latest\", aliases: [\"mistral-small\"], provider: \"mistral\", available: true, status: \"available\", source: \"default\" },\n];\n\n/** Attach every tier name as an alias of the model that tier ACTUALLY calls.\n *  One source: DEFAULT_TIER_MAP decides, the registry follows. A tier whose model\n *  is not in the registry is left unaliased on purpose — resolveModel then reports\n *  status \"unknown\" (fail-open) instead of inventing a row, and the drift test\n *  catches it. */\nexport const TIER_ALIAS_CONFLICTS: string[] = [];\n\nfunction applyTierAliases(entries: RegistryEntry[]): RegistryEntry[] {\n  for (const [tier, spec] of Object.entries(DEFAULT_TIER_MAP)) {\n    const owner = entries.find((e) => e.id === spec.model && e.provider === spec.provider);\n    // A tier name hand-written on any OTHER row is the drift bug returning. It\n    // would resolve correctly TODAY only because seed() lets the last write win\n    // and the rows happen to be ordered favourably — reorder the array and the\n    // tier silently points at the wrong provider again. Record it so a test can\n    // fail on it, then strip it so runtime is right regardless.\n    for (const e of entries) {\n      if (e !== owner && e.aliases.includes(tier)) {\n        TIER_ALIAS_CONFLICTS.push(`${tier} is hand-declared on ${e.id} (${e.provider}) but tier ${tier} calls ${spec.model} (${spec.provider})`);\n        e.aliases = e.aliases.filter((a) => a !== tier);\n      }\n    }\n    if (owner && !owner.aliases.includes(tier)) owner.aliases.push(tier);\n  }\n  return entries;\n}\napplyTierAliases(DEFAULTS);\n\n/** The live overlay, keyed by canonical id. Seeded from DEFAULTS (deep-copied so\n *  resetting is clean). refreshAvailability() mutates this; resolve/listModels\n *  read it synchronously. */\nlet OVERLAY = new Map<string, RegistryEntry>();\n/** alias → canonical id, rebuilt whenever the overlay is seeded. */\nlet ALIAS_INDEX = new Map<string, string>();\n\nfunction seed(): void {\n  OVERLAY = new Map(DEFAULTS.map((e) => [e.id, { ...e, aliases: [...e.aliases] }]));\n  ALIAS_INDEX = new Map();\n  for (const e of DEFAULTS) for (const a of e.aliases) ALIAS_INDEX.set(a, e.id);\n}\nseed();\n\n/** Reset the overlay back to the curated defaults. For tests. */\nexport function resetRegistry(): void {\n  seed();\n}\n\n/** Canonical id for a model id OR alias; null when we track neither. */\nexport function canonicalId(requested: string): string | null {\n  if (OVERLAY.has(requested)) return requested;\n  return ALIAS_INDEX.get(requested) ?? null;\n}\n\n/** The current entry for an id/alias, or undefined when untracked (fail-open). */\nexport function getEntry(requested: string): RegistryEntry | undefined {\n  const id = canonicalId(requested);\n  return id ? OVERLAY.get(id) : undefined;\n}\n\n/** All tracked entries (optionally provider-scoped), as a public ModelStatus[]. */\nexport function allEntries(provider?: string): ModelStatus[] {\n  const rows: ModelStatus[] = [];\n  for (const e of OVERLAY.values()) {\n    if (provider && e.provider !== provider) continue;\n    rows.push({\n      id: e.id,\n      alias: e.aliases[0],\n      provider: e.provider,\n      available: e.available,\n      status: e.status,\n      note: e.note,\n      source: e.source,\n    });\n  }\n  return rows;\n}\n\n/** Provider-scoped canonical ids (for refresh reconciliation). */\nexport function providerIds(provider: string): string[] {\n  return [...OVERLAY.values()].filter((e) => e.provider === provider).map((e) => e.id);\n}\n\n/** Mark a tracked id available/suspended from a live refresh. No-op if untracked. */\nexport function setAvailability(id: string, available: boolean, note?: string): void {\n  const e = OVERLAY.get(id);\n  if (!e) return;\n  e.available = available;\n  e.status = available ? \"available\" : \"suspended\";\n  e.source = \"refresh\";\n  if (note !== undefined) e.note = note;\n  else if (available) e.note = undefined;\n}\n","// F022 — Model Availability Harness. Public types for the availability layer:\n// the shared status read (ModelStatus), the resolve result, and the structured\n// error a caller can flag on. The registry is the one source both the spawn /\n// call path (resolveModel) and UI pickers (listModels) read.\n\nexport type AvailabilityStatus = \"available\" | \"suspended\" | \"unknown\";\n\n/** Where a model's current availability came from: the curated default seed,\n *  or a live provider refresh (Anthropic GET /v1/models). */\nexport type AvailabilitySource = \"default\" | \"refresh\";\n\n/** One row of the shared status read — what a UI model-picker renders. */\nexport interface ModelStatus {\n  /** Canonical provider model id, e.g. \"claude-fable-5\". */\n  id: string;\n  /** Short/tier alias, e.g. \"fable\" (the first registered alias). */\n  alias?: string;\n  /** \"anthropic\" | \"openai\" | \"gemini\" | \"mistral\" | … */\n  provider: string;\n  available: boolean;\n  status: AvailabilityStatus;\n  /** Friendly reason, e.g. \"suspended — US export-control directive (2026-06-12)\". */\n  note?: string;\n  source: AvailabilitySource;\n}\n\n/** Result of resolveModel — the spawn / call path consumes this synchronously. */\nexport interface ResolveResult {\n  /** True when the requested model itself is available. */\n  ok: boolean;\n  /** The id to actually use: `requested` when ok, else the first available fallback. */\n  model: string;\n  /** What the caller asked for (id or alias, normalized to the canonical id). */\n  requested: string;\n  provider?: string;\n  /** True when `model` differs from `requested` because we fell back. */\n  fellBack: boolean;\n  status: AvailabilityStatus;\n  /** Why it degraded / why it is unavailable. */\n  reason?: string;\n}\n\n/** Thrown by resolveModel when the requested model is unavailable, no usable\n *  fallback exists, and the caller passed `throwIfUnavailable`. Callers flag on\n *  `.code === \"model_unavailable\"`. */\nexport class ModelUnavailableError extends Error {\n  readonly code = \"model_unavailable\";\n  readonly requested: string;\n  readonly provider?: string;\n  readonly note?: string;\n  constructor(requested: string, note?: string, provider?: string) {\n    super(`model \"${requested}\" is unavailable${note ? ` (${note})` : \"\"}`);\n    this.name = \"ModelUnavailableError\";\n    this.requested = requested;\n    this.note = note;\n    this.provider = provider;\n  }\n}\n","// F022 — the synchronous, zero-I/O resolve + status read. This is the spawn /\n// call hot path (buddy's launcher calls resolveModel per spawn, cardmem #4842):\n// it MUST never await and never touch the network. Freshness comes only from a\n// prior async refreshAvailability(); resolve just reads the in-memory registry.\nimport { allEntries, canonicalId, getEntry } from \"./registry.js\";\nimport { ModelUnavailableError } from \"./types.js\";\nimport type { ModelStatus, ResolveResult } from \"./types.js\";\n\nexport interface ResolveOptions {\n  /** One id/alias or an ordered chain to try when `requested` is unavailable. */\n  fallback?: string | string[];\n  /** Scope hint (passed through to the result); does not gate lookup. */\n  provider?: string;\n  /** Throw ModelUnavailableError instead of returning ok:false when there is no\n   *  usable fallback. For callers that want to flag rather than degrade. */\n  throwIfUnavailable?: boolean;\n  /** Treat an id we do not track as UNUSABLE instead of fail-open.\n   *\n   *  Default is fail-open: we never block a model we simply do not track. That is\n   *  right for liveness, and wrong for a caller who is GATING — cms measured the\n   *  consequence: resolveModel(\"cheap\") answered {ok:true, model:\"cheap\"}, so a\n   *  consumer following our own instruction to gate on `ok` passed the gate and\n   *  then sent the literal string \"cheap\" to a provider as a model id. A\n   *  success-shaped non-answer is worse than an error, because an error gets\n   *  handled and a shape does not.\n   *\n   *  Set this when you need \"a model you actually know about\". Off by default so\n   *  no existing caller changes behaviour. */\n  requireKnown?: boolean;\n}\n\n/** The shared status read — UI pickers grey out `available:false` rows. */\nexport function listModels(opts: { provider?: string } = {}): ModelStatus[] {\n  return allEntries(opts.provider);\n}\n\n/** Is this id/alias currently usable? Untracked ids are fail-open (true) unless\n *  the caller asked for requireKnown. */\nfunction isAvailable(requested: string, requireKnown = false): boolean {\n  const e = getEntry(requested);\n  if (!e) return !requireKnown; // fail-open on unknown, unless gating\n  return e.available;\n}\n\n/**\n * Resolve a requested model (id or alias) to one that is actually usable.\n * Synchronous + offline by contract (cardmem #4842) — reads the registry only.\n *\n * - Available → pass through ({ ok:true, fellBack:false }).\n * - Unavailable + a fallback that IS available → swap ({ ok:false, fellBack:true }).\n * - Unavailable + no usable fallback → throw (throwIfUnavailable) or return ok:false.\n * - Unknown id → treated available (never block a model we do not track).\n */\nexport function resolveModel(requested: string, opts: ResolveOptions = {}): ResolveResult {\n  const id = canonicalId(requested) ?? requested;\n  const entry = getEntry(requested);\n  const provider = opts.provider ?? entry?.provider;\n\n  if (isAvailable(requested, opts.requireKnown)) {\n    return {\n      ok: true,\n      model: id,\n      requested: id,\n      provider,\n      fellBack: false,\n      status: entry?.status ?? \"unknown\",\n    };\n  }\n\n  // Requested is suspended — walk the fallback chain for the first available one.\n  const chain = opts.fallback === undefined ? [] : Array.isArray(opts.fallback) ? opts.fallback : [opts.fallback];\n  for (const fb of chain) {\n    if (isAvailable(fb, opts.requireKnown)) {\n      const fbId = canonicalId(fb) ?? fb;\n      return {\n        ok: false,\n        model: fbId,\n        requested: id,\n        provider,\n        fellBack: true,\n        status: entry?.status ?? \"suspended\",\n        reason: entry?.note ?? `${id} is unavailable`,\n      };\n    }\n  }\n\n  // No usable fallback.\n  const unknownAndGating = !entry && opts.requireKnown;\n  const reason = unknownAndGating\n    ? `${id} is not a model this registry knows — requireKnown was set, so it is not assumed usable`\n    : (entry?.note ?? `${id} is unavailable`);\n  if (opts.throwIfUnavailable) {\n    throw new ModelUnavailableError(id, reason, provider);\n  }\n  return {\n    ok: false,\n    model: id,\n    requested: id,\n    provider,\n    fellBack: false,\n    status: entry?.status ?? \"unknown\",\n    reason,\n  };\n}\n"],"mappings":";AAeO,IAAM,mBAA2C;AAAA,EACtD,MAAM,EAAE,UAAU,WAAW,OAAO,wBAAwB,WAAW,OAAO;AAAA,EAC9E,OAAO,EAAE,UAAU,WAAW,OAAO,wBAAwB,WAAW,OAAO;AAAA,EAC/E,UAAU,EAAE,UAAU,WAAW,OAAO,wBAAwB,WAAW,OAAO;AAAA,EAClF,OAAO,EAAE,UAAU,WAAW,OAAO,wBAAwB,WAAW,OAAO;AAAA;AAAA;AAAA,EAG/E,QAAQ,EAAE,UAAU,WAAW,OAAO,wBAAwB,WAAW,OAAO;AAAA;AAAA;AAAA,EAGhF,OAAO,EAAE,UAAU,UAAU,OAAO,yBAAyB,WAAW,OAAO;AAAA;AAAA,EAE/E,WAAW,EAAE,UAAU,UAAU,OAAO,0BAA0B,WAAW,OAAO;AACtF;AASO,SAAS,YACd,MACA,UACA,WACU;AACV,QAAM,OAAO,YAAY,IAAI,KAAK,iBAAiB,IAAI;AACvD,SAAO,EAAE,GAAG,MAAM,GAAG,SAAS;AAChC;;;ACrBA,IAAM,yBAAyB;AAc/B,IAAM,WAA4B;AAAA;AAAA,EAEhC,EAAE,IAAI,oBAAoB,SAAS,CAAC,OAAO,GAAG,UAAU,aAAa,WAAW,MAAM,QAAQ,aAAa,QAAQ,UAAU;AAAA,EAC7H,EAAE,IAAI,qBAAqB,SAAS,CAAC,QAAQ,GAAG,UAAU,aAAa,WAAW,MAAM,QAAQ,aAAa,QAAQ,UAAU;AAAA,EAC/H,EAAE,IAAI,mBAAmB,SAAS,CAAC,MAAM,GAAG,UAAU,aAAa,WAAW,MAAM,QAAQ,aAAa,QAAQ,UAAU;AAAA,EAC3H,EAAE,IAAI,kBAAkB,SAAS,CAAC,OAAO,GAAG,UAAU,aAAa,WAAW,MAAM,QAAQ,aAAa,QAAQ,UAAU;AAAA,EAC3H,EAAE,IAAI,mBAAmB,SAAS,CAAC,QAAQ,GAAG,UAAU,aAAa,WAAW,OAAO,QAAQ,aAAa,MAAM,wBAAwB,QAAQ,UAAU;AAAA;AAAA,EAE5J,EAAE,IAAI,oBAAoB,SAAS,CAAC,cAAc,GAAG,UAAU,UAAU,WAAW,MAAM,QAAQ,aAAa,QAAQ,UAAU;AAAA,EACjI,EAAE,IAAI,yBAAyB,SAAS,CAAC,mBAAmB,GAAG,UAAU,UAAU,WAAW,MAAM,QAAQ,aAAa,QAAQ,UAAU;AAAA;AAAA,EAE3I,EAAE,IAAI,0BAA0B,SAAS,CAAC,GAAG,UAAU,UAAU,WAAW,MAAM,QAAQ,aAAa,QAAQ,UAAU;AAAA;AAAA,EAEzH,EAAE,IAAI,wBAAwB,SAAS,CAAC,eAAe,GAAG,UAAU,WAAW,WAAW,MAAM,QAAQ,aAAa,QAAQ,UAAU;AAAA,EACvI,EAAE,IAAI,wBAAwB,SAAS,CAAC,eAAe,GAAG,UAAU,WAAW,WAAW,MAAM,QAAQ,aAAa,QAAQ,UAAU;AACzI;AAOO,IAAM,uBAAiC,CAAC;AAE/C,SAAS,iBAAiB,SAA2C;AACnE,aAAW,CAAC,MAAM,IAAI,KAAK,OAAO,QAAQ,gBAAgB,GAAG;AAC3D,UAAM,QAAQ,QAAQ,KAAK,CAAC,MAAM,EAAE,OAAO,KAAK,SAAS,EAAE,aAAa,KAAK,QAAQ;AAMrF,eAAW,KAAK,SAAS;AACvB,UAAI,MAAM,SAAS,EAAE,QAAQ,SAAS,IAAI,GAAG;AAC3C,6BAAqB,KAAK,GAAG,IAAI,wBAAwB,EAAE,EAAE,KAAK,EAAE,QAAQ,cAAc,IAAI,UAAU,KAAK,KAAK,KAAK,KAAK,QAAQ,GAAG;AACvI,UAAE,UAAU,EAAE,QAAQ,OAAO,CAAC,MAAM,MAAM,IAAI;AAAA,MAChD;AAAA,IACF;AACA,QAAI,SAAS,CAAC,MAAM,QAAQ,SAAS,IAAI,EAAG,OAAM,QAAQ,KAAK,IAAI;AAAA,EACrE;AACA,SAAO;AACT;AACA,iBAAiB,QAAQ;AAKzB,IAAI,UAAU,oBAAI,IAA2B;AAE7C,IAAI,cAAc,oBAAI,IAAoB;AAE1C,SAAS,OAAa;AACpB,YAAU,IAAI,IAAI,SAAS,IAAI,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,GAAG,GAAG,SAAS,CAAC,GAAG,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC;AAChF,gBAAc,oBAAI,IAAI;AACtB,aAAW,KAAK,SAAU,YAAW,KAAK,EAAE,QAAS,aAAY,IAAI,GAAG,EAAE,EAAE;AAC9E;AACA,KAAK;AAGE,SAAS,gBAAsB;AACpC,OAAK;AACP;AAGO,SAAS,YAAY,WAAkC;AAC5D,MAAI,QAAQ,IAAI,SAAS,EAAG,QAAO;AACnC,SAAO,YAAY,IAAI,SAAS,KAAK;AACvC;AAGO,SAAS,SAAS,WAA8C;AACrE,QAAM,KAAK,YAAY,SAAS;AAChC,SAAO,KAAK,QAAQ,IAAI,EAAE,IAAI;AAChC;AAGO,SAAS,WAAW,UAAkC;AAC3D,QAAM,OAAsB,CAAC;AAC7B,aAAW,KAAK,QAAQ,OAAO,GAAG;AAChC,QAAI,YAAY,EAAE,aAAa,SAAU;AACzC,SAAK,KAAK;AAAA,MACR,IAAI,EAAE;AAAA,MACN,OAAO,EAAE,QAAQ,CAAC;AAAA,MAClB,UAAU,EAAE;AAAA,MACZ,WAAW,EAAE;AAAA,MACb,QAAQ,EAAE;AAAA,MACV,MAAM,EAAE;AAAA,MACR,QAAQ,EAAE;AAAA,IACZ,CAAC;AAAA,EACH;AACA,SAAO;AACT;AAGO,SAAS,YAAY,UAA4B;AACtD,SAAO,CAAC,GAAG,QAAQ,OAAO,CAAC,EAAE,OAAO,CAAC,MAAM,EAAE,aAAa,QAAQ,EAAE,IAAI,CAAC,MAAM,EAAE,EAAE;AACrF;AAGO,SAAS,gBAAgB,IAAY,WAAoB,MAAqB;AACnF,QAAM,IAAI,QAAQ,IAAI,EAAE;AACxB,MAAI,CAAC,EAAG;AACR,IAAE,YAAY;AACd,IAAE,SAAS,YAAY,cAAc;AACrC,IAAE,SAAS;AACX,MAAI,SAAS,OAAW,GAAE,OAAO;AAAA,WACxB,UAAW,GAAE,OAAO;AAC/B;;;ACnGO,IAAM,wBAAN,cAAoC,MAAM;AAAA,EACtC,OAAO;AAAA,EACP;AAAA,EACA;AAAA,EACA;AAAA,EACT,YAAY,WAAmB,MAAe,UAAmB;AAC/D,UAAM,UAAU,SAAS,mBAAmB,OAAO,KAAK,IAAI,MAAM,EAAE,EAAE;AACtE,SAAK,OAAO;AACZ,SAAK,YAAY;AACjB,SAAK,OAAO;AACZ,SAAK,WAAW;AAAA,EAClB;AACF;;;ACzBO,SAAS,WAAW,OAA8B,CAAC,GAAkB;AAC1E,SAAO,WAAW,KAAK,QAAQ;AACjC;AAIA,SAAS,YAAY,WAAmB,eAAe,OAAgB;AACrE,QAAM,IAAI,SAAS,SAAS;AAC5B,MAAI,CAAC,EAAG,QAAO,CAAC;AAChB,SAAO,EAAE;AACX;AAWO,SAAS,aAAa,WAAmB,OAAuB,CAAC,GAAkB;AACxF,QAAM,KAAK,YAAY,SAAS,KAAK;AACrC,QAAM,QAAQ,SAAS,SAAS;AAChC,QAAM,WAAW,KAAK,YAAY,OAAO;AAEzC,MAAI,YAAY,WAAW,KAAK,YAAY,GAAG;AAC7C,WAAO;AAAA,MACL,IAAI;AAAA,MACJ,OAAO;AAAA,MACP,WAAW;AAAA,MACX;AAAA,MACA,UAAU;AAAA,MACV,QAAQ,OAAO,UAAU;AAAA,IAC3B;AAAA,EACF;AAGA,QAAM,QAAQ,KAAK,aAAa,SAAY,CAAC,IAAI,MAAM,QAAQ,KAAK,QAAQ,IAAI,KAAK,WAAW,CAAC,KAAK,QAAQ;AAC9G,aAAW,MAAM,OAAO;AACtB,QAAI,YAAY,IAAI,KAAK,YAAY,GAAG;AACtC,YAAM,OAAO,YAAY,EAAE,KAAK;AAChC,aAAO;AAAA,QACL,IAAI;AAAA,QACJ,OAAO;AAAA,QACP,WAAW;AAAA,QACX;AAAA,QACA,UAAU;AAAA,QACV,QAAQ,OAAO,UAAU;AAAA,QACzB,QAAQ,OAAO,QAAQ,GAAG,EAAE;AAAA,MAC9B;AAAA,IACF;AAAA,EACF;AAGA,QAAM,mBAAmB,CAAC,SAAS,KAAK;AACxC,QAAM,SAAS,mBACX,GAAG,EAAE,iGACJ,OAAO,QAAQ,GAAG,EAAE;AACzB,MAAI,KAAK,oBAAoB;AAC3B,UAAM,IAAI,sBAAsB,IAAI,QAAQ,QAAQ;AAAA,EACtD;AACA,SAAO;AAAA,IACL,IAAI;AAAA,IACJ,OAAO;AAAA,IACP,WAAW;AAAA,IACX;AAAA,IACA,UAAU;AAAA,IACV,QAAQ,OAAO,UAAU;AAAA,IACzB;AAAA,EACF;AACF;","names":[]}

package/dist/index.d.ts

@@ -117,6 +117,12 @@ interface ChatRequest {
     spec: TierSpec;
     tools?: Tool[];
     maxTokens?: number;
+    /** F039 — Mistral prompt caching. A stable application-level id for "the same
+     *  conversation" (a conversation id, a session id). Requests sharing a prefix AND
+     *  this key reuse the cached prefix at 10% of the input rate. Not auto-generated:
+     *  only the caller knows what "the same conversation" means, and an SDK-invented
+     *  key would either collide across tenants or never hit. */
+    promptCacheKey?: string;
     temperature?: number;
     /** "json" → request JSON-object output where the provider supports it (F009). */
     responseFormat?: "json" | "text";
@@ -595,6 +601,9 @@ declare const chatInputSchema: z.ZodObject<{
     }>]>, "many">>;
     purpose: z.ZodOptional<z.ZodString>;
     labels: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodString>>;
+    /** F039 — reuse a cached prompt prefix on Mistral at 10% of the input rate.
+     *  A stable id for "the same conversation" (conversation id, session id). */
+    promptCacheKey: z.ZodOptional<z.ZodString>;
     prompt: z.ZodOptional<z.ZodString>;
     messages: z.ZodOptional<z.ZodArray<z.ZodObject<{
         role: z.ZodEnum<["system", "user", "assistant", "tool"]>;
@@ -687,6 +696,7 @@ declare const chatInputSchema: z.ZodObject<{
     responseFormat: z.ZodOptional<z.ZodEnum<["json", "text"]>>;
 }, "strip", z.ZodTypeAny, {
     system?: string | undefined;
+    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     prompt?: string | undefined;
     messages?: {
         role: "system" | "user" | "assistant" | "tool";
@@ -712,9 +722,9 @@ declare const chatInputSchema: z.ZodObject<{
     }[] | undefined;
     temperature?: number | undefined;
     purpose?: string | undefined;
+    promptCacheKey?: string | undefined;
     maxTokens?: number | undefined;
     responseFormat?: "text" | "json" | undefined;
-    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -728,6 +738,7 @@ declare const chatInputSchema: z.ZodObject<{
     labels?: Record<string, string> | undefined;
 }, {
     system?: string | undefined;
+    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     prompt?: string | undefined;
     messages?: {
         role: "system" | "user" | "assistant" | "tool";
@@ -753,9 +764,9 @@ declare const chatInputSchema: z.ZodObject<{
     }[] | undefined;
     temperature?: number | undefined;
     purpose?: string | undefined;
+    promptCacheKey?: string | undefined;
     maxTokens?: number | undefined;
     responseFormat?: "text" | "json" | undefined;
-    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -808,9 +819,9 @@ declare const visionInputSchema: z.ZodObject<{
     image: string | Uint8Array<ArrayBuffer>;
     prompt: string;
     system?: string | undefined;
+    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     mimeType?: string | undefined;
     purpose?: string | undefined;
-    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -826,9 +837,9 @@ declare const visionInputSchema: z.ZodObject<{
     image: string | Uint8Array<ArrayBuffer>;
     prompt: string;
     system?: string | undefined;
+    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     mimeType?: string | undefined;
     purpose?: string | undefined;
-    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -880,9 +891,9 @@ declare const videoInputSchema: z.ZodObject<{
     video: string | Uint8Array<ArrayBuffer>;
     prompt: string;
     system?: string | undefined;
+    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     mimeType?: string | undefined;
     purpose?: string | undefined;
-    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -898,9 +909,9 @@ declare const videoInputSchema: z.ZodObject<{
     video: string | Uint8Array<ArrayBuffer>;
     prompt: string;
     system?: string | undefined;
+    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     mimeType?: string | undefined;
     purpose?: string | undefined;
-    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -949,8 +960,8 @@ declare const translateInputSchema: z.ZodObject<{
 }, "strip", z.ZodTypeAny, {
     text: string;
     to: string;
-    purpose?: string | undefined;
     tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
+    purpose?: string | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -966,8 +977,8 @@ declare const translateInputSchema: z.ZodObject<{
 }, {
     text: string;
     to: string;
-    purpose?: string | undefined;
     tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
+    purpose?: string | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1045,6 +1056,7 @@ declare const imageInputSchema: z.ZodObject<{
     retryOnBlack: z.ZodOptional<z.ZodBoolean>;
 }, "strip", z.ZodTypeAny, {
     prompt: string;
+    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     seed?: number | undefined;
     purpose?: string | undefined;
     loras?: {
@@ -1054,7 +1066,6 @@ declare const imageInputSchema: z.ZodObject<{
     lora?: string | undefined;
     width?: number | undefined;
     height?: number | undefined;
-    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1074,6 +1085,7 @@ declare const imageInputSchema: z.ZodObject<{
     retryOnBlack?: boolean | undefined;
 }, {
     prompt: string;
+    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     seed?: number | undefined;
     purpose?: string | undefined;
     loras?: {
@@ -1083,7 +1095,6 @@ declare const imageInputSchema: z.ZodObject<{
     lora?: string | undefined;
     width?: number | undefined;
     height?: number | undefined;
-    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1142,10 +1153,10 @@ declare const animateInputSchema: z.ZodObject<{
     resolution: z.ZodOptional<z.ZodString>;
 }, "strip", z.ZodTypeAny, {
     image: string | Uint8Array<ArrayBuffer>;
+    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     prompt?: string | undefined;
     resolution?: string | undefined;
     purpose?: string | undefined;
-    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1160,10 +1171,10 @@ declare const animateInputSchema: z.ZodObject<{
     durationSec?: number | undefined;
 }, {
     image: string | Uint8Array<ArrayBuffer>;
+    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     prompt?: string | undefined;
     resolution?: string | undefined;
     purpose?: string | undefined;
-    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1216,9 +1227,9 @@ declare const trainStyleInputSchema: z.ZodObject<{
     createMasks: z.ZodOptional<z.ZodBoolean>;
 }, "strip", z.ZodTypeAny, {
     images: string | string[];
+    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     purpose?: string | undefined;
     steps?: number | undefined;
-    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1235,9 +1246,9 @@ declare const trainStyleInputSchema: z.ZodObject<{
     createMasks?: boolean | undefined;
 }, {
     images: string | string[];
+    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     purpose?: string | undefined;
     steps?: number | undefined;
-    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1286,8 +1297,8 @@ declare const embeddingInputSchema: z.ZodObject<{
     text: z.ZodUnion<[z.ZodString, z.ZodArray<z.ZodString, "many">]>;
 }, "strip", z.ZodTypeAny, {
     text: string | string[];
-    purpose?: string | undefined;
     tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
+    purpose?: string | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1301,8 +1312,8 @@ declare const embeddingInputSchema: z.ZodObject<{
     labels?: Record<string, string> | undefined;
 }, {
     text: string | string[];
-    purpose?: string | undefined;
     tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
+    purpose?: string | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1359,9 +1370,9 @@ declare const transcribeInputSchema: z.ZodObject<{
     timestamps: z.ZodOptional<z.ZodUnion<[z.ZodEnum<["word", "segment"]>, z.ZodArray<z.ZodEnum<["word", "segment"]>, "many">]>>;
 }, "strip", z.ZodTypeAny, {
     audio: string | Uint8Array<ArrayBuffer>;
+    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     language?: string | undefined;
     purpose?: string | undefined;
-    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1378,9 +1389,9 @@ declare const transcribeInputSchema: z.ZodObject<{
     timestamps?: "word" | "segment" | ("word" | "segment")[] | undefined;
 }, {
     audio: string | Uint8Array<ArrayBuffer>;
+    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     language?: string | undefined;
     purpose?: string | undefined;
-    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1432,9 +1443,9 @@ declare const ocrInputSchema: z.ZodObject<{
     mimeType: z.ZodOptional<z.ZodString>;
 }, "strip", z.ZodTypeAny, {
     document: string | Uint8Array<ArrayBuffer>;
+    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     mimeType?: string | undefined;
     purpose?: string | undefined;
-    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1448,9 +1459,9 @@ declare const ocrInputSchema: z.ZodObject<{
     labels?: Record<string, string> | undefined;
 }, {
     document: string | Uint8Array<ArrayBuffer>;
+    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     mimeType?: string | undefined;
     purpose?: string | undefined;
-    tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1496,8 +1507,8 @@ declare const moderationInputSchema: z.ZodObject<{
     input: z.ZodUnion<[z.ZodString, z.ZodArray<z.ZodString, "many">]>;
 }, "strip", z.ZodTypeAny, {
     input: string | string[];
-    purpose?: string | undefined;
     tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
+    purpose?: string | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1511,8 +1522,8 @@ declare const moderationInputSchema: z.ZodObject<{
     labels?: Record<string, string> | undefined;
 }, {
     input: string | string[];
-    purpose?: string | undefined;
     tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
+    purpose?: string | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1573,8 +1584,8 @@ declare const podcastInputSchema: z.ZodObject<{
         speaker: string;
     }[];
     voices: Record<string, string>;
-    purpose?: string | undefined;
     tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
+    purpose?: string | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1593,8 +1604,8 @@ declare const podcastInputSchema: z.ZodObject<{
         speaker: string;
     }[];
     voices: Record<string, string>;
-    purpose?: string | undefined;
     tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
+    purpose?: string | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1641,7 +1652,15 @@ declare const ttsInputSchema: z.ZodObject<{
     text: z.ZodString;
     voice: z.ZodString;
     /** F037: voice to use if `voice` is one we know the provider has retired. Without
-     *  it a retired voice throws VoiceUnavailableError rather than reaching the API. */
+     *  it a retired voice throws VoiceUnavailableError rather than reaching the API.
+     *
+     *  THINK BEFORE SETTING THIS. It is right for batch work where any acceptable voice
+     *  will do (a generated podcast, a bulk render). It is WRONG where the voice is an
+     *  identity a human recognises — a brand voice, a house narrator: falling back means
+     *  your product suddenly speaks with a stranger's voice and says nothing about why.
+     *  There, leave it unset and let the throw alarm you. Silence WITH a message beats
+     *  the wrong audio WITHOUT one. (torrent-search-api's call on their own house voice,
+     *  2026-08-11 — they took the alarm over the fallback, deliberately.) */
     voiceFallback: z.ZodOptional<z.ZodString>;
     lang: z.ZodOptional<z.ZodString>;
     format: z.ZodOptional<z.ZodString>;
@@ -1649,8 +1668,8 @@ declare const ttsInputSchema: z.ZodObject<{
 }, "strip", z.ZodTypeAny, {
     text: string;
     voice: string;
-    purpose?: string | undefined;
     tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
+    purpose?: string | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -1669,8 +1688,8 @@ declare const ttsInputSchema: z.ZodObject<{
 }, {
     text: string;
     voice: string;
-    purpose?: string | undefined;
     tier?: "fast" | "smart" | "powerful" | "cheap" | "vision" | "video" | "embedding" | undefined;
+    purpose?: string | undefined;
     override?: {
         provider?: string | undefined;
         model?: string | undefined;
@@ -2053,8 +2072,8 @@ declare const falStubAdapter: ProviderAdapter;
  *  wires the live adapters. */
 declare const stubProviders: Record<string, ProviderAdapter>;
 
-declare const VERSION: "0.28.0";
-declare const SDK_TAG: "@broberg/ai-sdk@0.28.0";
+declare const VERSION: "0.30.0";
+declare const SDK_TAG: "@broberg/ai-sdk@0.30.0";
 
 /** Built-in defaults. Every entry is overridable via AiConfig.defaults or a
  *  per-call override.