1 line
111 KiB
JSON
1 line
111 KiB
JSON
{"repo": "choeng-rayu/rayu-cli", "files": {"rayu/src/utils/rayuProviders.ts": {"caca4043c11f75763539d7ad2adb24acfc0b11d8": {"t": 1785674297.7926712, "n": 31973, "h": "6ef75baacac8431c", "c": "// Shared registry of built-in provider presets for Rayu-CLI, used by both the\n// onboarding flow and the /connect command. Also imports API keys from known\n// environment variables (and a local .env) into ~/.rayu/providers.json so keys\n// the user already has in .env become first-class config entries.\nimport { existsSync, readFileSync } from 'fs'\nimport { join } from 'path'\nimport {\n type ProviderKind,\n type RayuProvider,\n getActiveProvider,\n loadRayuConfig,\n saveRayuConfig,\n} from './rayuConfig.js'\nimport { loadDotEnv } from './envUtils.js'\nimport { OLLAMA_CLOUD_BASE_URL, OLLAMA_CLOUD_PROVIDER_ID } from '../services/api/ollamaCloud.js'\n\n\nexport type ProviderPreset = {\n /** Stable provider id (also the config id). */\n id: string\n label: string\n kind: ProviderKind\n /** Base URL for openai-compatible providers (omit for bedrock / local-prompt). */\n baseURL?: string\n /** Sensible default model id (used until the live catalog is fetched). */\n defaultModel?: string\n /** Provider-native small/fast model for lightweight work and haiku aliases. */\n smallFastModel?: string\n /** Env var names whose value is the API key for this provider. */\n envKeys?: string[]\n /** True for endpoints where the user must type the base URL (no fixed host). */\n promptBaseURL?: boolean\n /**\n * True for presets authenticated via Google OAuth / Application Default\n * Credentials rather than a typed API key (e.g. Gemini on Vertex AI). The\n * /connect flow runs ADC detection / loopback login + project/region prompts\n * instead of the API-key step.\n */\n requiresOAuth?: boolean\n}\n\n// --- AWS Bedrock (API key / bearer token) -----------------------------------\n// AWS Bedrock exposes an OpenAI-compatible Chat Completions API on the\n// `bedrock-mantle` endpoint (AWS-recommended). It is region-scoped and\n// authenticated with a Bedrock API key (the AWS_BEARER_TOKEN_BEDROCK bearer\n// token). Both `GET /v1/models` and `POST /v1/chat/completions` work with that\n// token, so Rayu reuses its OpenAI-compatible adapter + model-catalog fetch\n// verbatim. Mantle (unlike the bare bedrock-runtime OpenAI surface) properly\n// separates reasoning models' chain-of-thought into `reasoning_content` and\n// parses native tool calls — required for Kimi K2 Thinking et al.\n\n/** Default AWS region used for Bedrock when none is specified. */\nexport const DEFAULT_BEDROCK_REGION = 'us-east-1'\n\n/** Build the OpenAI-compatible Bedrock base URL for a given AWS region. */\nexport function bedrockBaseURL(region: string): string {\n const r = (region || DEFAULT_BEDROCK_REGION).trim()\n // The recommended `bedrock-mantle` endpoint serves the OpenAI Chat Completions\n // API at /v1/chat/completions and the model catalog at /v1/models. Model IDs\n // on mantle use the `<vendor>.<model>` form (e.g. moonshotai.kimi-k2-thinking).\n return `https://bedrock-mantle.${r}.api.aws/v1`\n}\n\n/** AWS regions that commonly support the Bedrock runtime endpoint. */\nexport const BEDROCK_REGIONS: Array<{ id: string; label: string }> = [\n { id: 'us-east-1', label: 'US East (N. Virginia) · us-east-1' },\n { id: 'us-east-2', label: 'US East (Ohio) · us-east-2' },\n { id: 'us-west-2', label: 'US West (Oregon) · us-west-2' },\n { id: 'ap-south-1', label: 'Asia Pacific (Mumbai) · ap-south-1' },\n { id: 'ap-southeast-1', label: 'Asia Pacific (Singapore) · ap-southeast-1' },\n { id: 'ap-southeast-2', label: 'Asia Pacific (Sydney) · ap-southeast-2' },\n { id: 'ap-northeast-1', label: 'Asia Pacific (Tokyo) · ap-northeast-1' },\n { id: 'eu-central-1', label: 'Europe (Frankfurt) · eu-central-1' },\n { id: 'eu-west-1', label: 'Europe (Ireland) · eu-west-1' },\n { id: 'eu-west-3', label: 'Europe (Paris) · eu-west-3' },\n]\n\n// --- Google Vertex AI (Gemini via OAuth / ADC) ------------------------------\n// Vertex AI exposes an OpenAI-compatible Chat Completions endpoint under the\n// per-region `…/endpoints/openapi` path, authenticated with a Google Cloud\n// OAuth bearer token (cloud-platform scope) rather than a static API key. Rayu\n// reuses its OpenAI-compatible adapter with a fetch wrapper that injects a\n// freshly-minted token. The provider is project + region scoped.\n\n/** Stable provider id for the Gemini/Vertex (OAuth) preset. */\nexport const GEMINI_VERTEX_PROVIDER_ID = 'gemini-vertex'\n\n/** Default GCP location for Vertex chat. `global` serves the newest Gemini\n * models (3.x) and also the 2.x line, so it's the most compatible default. */\nexport const DEFAULT_VERTEX_REGION = 'global'\n\n/** Vertex AI host for a location: the `global` location uses the un-prefixed\n * host, every regional location uses the `{region}-` prefix. */\nexport function vertexHost(region: string): string {\n const r = (region || DEFAULT_VERTEX_REGION).trim()\n return r === 'global'\n ? 'aiplatform.googleapis.com'\n : `${r}-aiplatform.googleapis.com`\n}\n\n/**\n * Build the OpenAI-compatible Vertex AI base URL for a project + region. The\n * Chat Completions endpoint lives at `{base}/chat/completions`; model ids are\n * sent with a `google/` publisher prefix (handled at request time).\n */\nexport function vertexBaseURL(project: string, region: string): string {\n const r = (region || DEFAULT_VERTEX_REGION).trim()\n const p = project.trim()\n return `https://${vertexHost(r)}/v1beta1/projects/${p}/locations/${r}/endpoints/openapi`\n}\n\n/** GCP locations that commonly serve Gemini on Vertex AI. `global` first — it\n * hosts the newest Gemini models and is the recommended default. */\nexport const VERTEX_REGIONS: Array<{ id: string; label: string }> = [\n { id: 'global', label: 'Global (recommended — newest Gemini models) · global' },\n { id: 'us-central1', label: 'US Central (Iowa) · us-central1' },\n { id: 'us-east1', label: 'US East (S. Carolina) · us-east1' },\n { id: 'us-east4', label: 'US East (N. Virginia) · us-east4' },\n { id: 'us-west1', label: 'US West (Oregon) · us-west1' },\n { id: 'europe-west1', label: 'Europe West (Belgium) · europe-west1' },\n { id: 'europe-west4', label: 'Europe West (Netherlands) · europe-west4' },\n { id: 'asia-northeast1', label: 'Asia Northeast (Tokyo) · asia-northeast1' },\n { id: 'asia-southeast1', label: 'Asia Southeast (Singapore) · asia-southeast1' },\n]\n\n\n// --- Ollama (local) ---------------------------------------------------------\n// Ollama serves an OpenAI-compatible API at http://localhost:11434/v1 (plus its\n// native /api surface). No API key is required. The address can be overridden\n// with the standard OLLAMA_HOST env var (a bare port, host:port, or full URL),\n// matching the Ollama CLI, so Rayu auto-detects a server started elsewhere.\nexport const OLLAMA_DEFAULT_BASE_URL = 'http://localhost:11434/v1'\n\n/** OpenAI-compatible base URL for a local Ollama server, honoring OLLAMA_HOST. */\nexport function ollamaBaseURL(): string {\n let raw = (process.env.OLLAMA_HOST || '').trim()\n if (!raw) return OLLAMA_DEFAULT_BASE_URL\n if (/^\\d+$/.test(raw)) raw = `localhost:${raw}` // bare port, e.g. \"11434\"\n if (!/^https?:\\/\\//i.test(raw)) raw = `http://${raw}` // host or host:port\n raw = raw.replace(/\\/+$/, '')\n if (!/\\/v1(\\/.*)?$/.test(raw)) raw = `${raw}/v1` // ensure the OpenAI /v1 base\n return raw\n}\n\n\n// All confirmed OpenAI-compatible providers (tool calling + /v1/models), plus\n// AWS Bedrock via its OpenAI-compatible bedrock-runtime endpoint (authenticated\n// with a Bedrock API key / bearer token). Rayu connects to OpenAI-compatible\n// endpoints; Bedrock is routed through the same adapter once a region + API key\n// are supplied via /connect.\n/** Stable provider id for the Rayu-hosted gateway provider (registered on\n * login for paid users; not connected manually via /connect). */\nexport const RAYU_HOSTED_PROVIDER_ID = 'rayu-hosted'\nexport const RAYU_HOSTED_PROVIDER_LABEL = 'Rayu (hosted)'\n\n/**\n * Built-in providers that support storing MULTIPLE API keys with automatic\n * rate-limit key rotation:\n * • NVIDIA / OpenRouter — openai-compatible → rotation in openaiAdapter\n * (withKeyRotation).\n * • Ollama Cloud — anthropic-compatible → rotation in anthropicMessagesClient\n * (makeKeyRotatingFetch), since it uses the native Anthropic SDK path.\n * The rotation mechanism is generic per client kind, so enabling another\n * provider is just adding its id here — or, with no code change, via the\n * RAYU_MULTI_KEY_PROVIDERS env var (see envMultiKeyProviderIds below). Always\n * Basic-plan gated via isMultiApiKeyAllowed().\n */\nexport const MULTI_KEY_PROVIDER_IDS: ReadonlySet<string> = new Set([\n 'nvidia',\n 'openrouter',\n 'ollama-cloud',\n])\n\n/**\n * Extra multi-key provider ids from the RAYU_MULTI_KEY_PROVIDERS env var\n * (comma- or space-separated), unioned with the built-in set above. This is the\n * self-serve knob: to give ANOTHER provider multi-API-key rotation you just add\n * its id in .env — no code change, no new release. e.g.\n * RAYU_MULTI_KEY_PROVIDERS=deepseek,groq\n * (Still Basic-plan gated via isMultiApiKeyAllowed(), same as the built-ins.)\n */\nfunction envMultiKeyProviderIds(): string[] {\n const raw = process.env.RAYU_MULTI_KEY_PROVIDERS\n if (!raw) return []\n return raw\n .split(/[\\s,]+/)\n .map(s => s.trim())\n .filter(Boolean)\n}\n\n/**\n * Provider kinds whose auth is a ROTATABLE user-supplied API key sent as a\n * request header. Multi-key rotation is ONLY wired for these two client paths\n * (openaiAdapter.withKeyRotation + anthropicMessagesClient.makeKeyRotatingFetch).\n *\n * Every other kind is intentionally excluded, because their credential is NOT a\n * rotatable user API key:\n * • 'anthropic' — first-party Console key (single x-api-key)\n * • 'kiro' — Kiro/CodeWhisperer (an apikey OR kiro-cli OAuth token)\n * • 'vertex' — Google OAuth / ADC\n * • 'genai' — Login-with-Gemini OAuth\n * • 'copilot' — GitHub OAuth device flow\n * • 'bedrock' — AWS creds / bearer token\n * • 'rayu-hosted' — Rayu account JWT via the gateway\n * This is the HARD guard behind the user requirement: multi-key is for BYO\n * API-key providers only — never Kiro, never an Auth/managed-credential provider.\n */\nconst MULTI_KEY_PROVIDER_KINDS: ReadonlySet<ProviderKind> = new Set<ProviderKind>([\n 'openai-compatible',\n 'anthropic-compatible',\n])\n\n/**\n * Resolve a provider's kind by id: the configured provider first (covers custom\n * / local ids the user added), then the built-in preset. Undefined only for a\n * brand-new id that is neither configured nor a preset.\n */\nfunction providerKindForId(providerId: string): ProviderKind | undefined {\n const configured = loadRayuConfig().providers.find(p => p.id === providerId)\n if (configured) return configured.kind\n return PROVIDER_PRESETS.find(p => p.id === providerId)?.kind\n}\n\n/**\n * True when a provider supports the multi-API-key manager + rotation.\n *\n * Two conditions, BOTH required:\n * 1. the id is explicitly allowed — a built-in (nvidia/openrouter/ollama-cloud)\n * or opted-in via RAYU_MULTI_KEY_PROVIDERS; AND\n * 2. its kind authenticates with a rotatable user API key (openai-compatible or\n * anthropic-compatible).\n *\n * Condition 2 is the safety net: even if someone mistakenly adds `kiro` or an\n * OAuth provider id to RAYU_MULTI_KEY_PROVIDERS, it can NEVER become multi-key,\n * because rotation only exists for — and only makes sense for — user-API-key\n * providers. An id whose kind can't be resolved (neither configured nor a known\n * preset) is DENIED — we only enable multi-key once we can confirm the provider\n * authenticates with a rotatable user API key.\n */\nexport function supportsMultiApiKey(providerId: string | undefined): boolean {\n if (!providerId) return false\n const listed =\n MULTI_KEY_PROVIDER_IDS.has(providerId) ||\n envMultiKeyProviderIds().includes(providerId)\n if (!listed) return false\n const kind = providerKindForId(providerId)\n return kind !== undefined && MULTI_KEY_PROVIDER_KINDS.has(kind)\n}\n\nexport const PROVIDER_PRESETS: ProviderPreset[] = [\n {\n // First-party Anthropic API (the Console — console.anthropic.com). Uses the\n // native Anthropic Messages API via getAnthropicClient() (kind:'anthropic'\n // → getAPIProvider() === 'anthropic'), the SAME path the app uses for\n // first-party Claude. That means extended THINKING and the full CONTEXT\n // window (incl. the 1M-token context beta) are supported NATIVELY — not\n // emulated through the OpenAI-compatible adapter. Authenticated with a\n // Console API key (sk-ant-…) sent as x-api-key directly to\n // api.anthropic.com. No baseURL: first-party endpoint is the SDK default.\n id: 'anthropic',\n label: 'Anthropic — Claude (console.anthropic.com API key) · native thinking + context',\n kind: 'anthropic',\n defaultModel: 'claude-sonnet-4-6',\n smallFastModel: 'claude-haiku-4-5-20251001',\n envKeys: ['ANTHROPIC_API_KEY'],\n },\n {\n // LongCat — Meituan's LongCat-2.0, exposed via an Anthropic-compatible\n // Messages API at https://api.longcat.chat/anthropic/v1/messages,\n // authenticated with `Authorization: Bearer <key>` (NOT x-api-key). Rayu\n // talks to it with the NATIVE Anthropic SDK (kind:'anthropic-compatible')\n // pointed at that baseURL, with the key sent as a Bearer authToken — so\n // extended THINKING and the full 1M-token CONTEXT work NATIVELY, not\n // emulated through the OpenAI-compatible adapter (the same reasoning as the\n // first-party 'anthropic' preset above). One model today: LongCat-2.0 (1M\n // context — see KNOWN_MODEL_CONTEXT in rayuConfig.ts). LongCat also offers an\n // OpenAI-compatible surface, but the Anthropic path maps 1:1 with claude.ts.\n // Docs: https://longcat.chat/platform/docs/APIDocs.html\n id: 'longcat',\n label: 'LongCat — Meituan (api.longcat.chat) · LongCat-2.0 1M ctx · native thinking',\n kind: 'anthropic-compatible',\n baseURL: 'https://api.longcat.chat/anthropic',\n defaultModel: 'LongCat-2.0',\n envKeys: ['LONGCAT_API_KEY'],\n },\n {\n // Ollama Cloud — Ollama's HOSTED models (ollama.com), a SEPARATE connector\n // from the local `ollama` preset (localhost, no key). Exposed via an\n // Anthropic-compatible Messages API at https://ollama.com/v1/messages,\n // authenticated with `Authorization: Bearer <ollama.com API key>`. Rayu uses\n // the NATIVE Anthropic SDK (kind:'anthropic-compatible' → the key is sent as\n // a Bearer authToken) so thinking + tools map 1:1 with claude.ts. Unlike a\n // single-model preset, the /connect flow FETCHES the user's account models\n // (fetchOllamaCloudModels → GET /v1/models, /api/tags fallback) and their\n // real context windows (POST /api/show → model_info.<arch>.context_length;\n // the KNOWN_MODEL_CONTEXT table is the fallback). Cloud model ids carry a\n // `-cloud`/`:cloud` tag, e.g. gpt-oss:120b-cloud, qwen3-coder:cloud.\n // Docs: https://docs.ollama.com/api/anthropic-compatibility\n id: OLLAMA_CLOUD_PROVIDER_ID,\n label: 'Ollama Cloud (ollama.com) · hosted models · fetches your account models',\n kind: 'anthropic-compatible',\n baseURL: OLLAMA_CLOUD_BASE_URL,\n defaultModel: 'gpt-oss:120b-cloud',\n envKeys: ['OLLAMA_CLOUD_API_KEY'],\n },\n {\n id: 'nvidia',\n label: 'NVIDIA NIM (integrate.api.nvidia.com)',\n kind: 'openai-compatible',\n baseURL: 'https://integrate.api.nvidia.com/v1',\n defaultModel: 'meta/llama-3.3-70b-instruct',\n smallFastModel: 'nvidia/llama-3.1-nemotron-nano-8b-v1',\n envKeys: ['NVIDIA_API_KEY'],\n },\n {\n id: 'doubleword',\n label: 'Doubleword (api.doubleword.ai)',\n kind: 'openai-compatible',\n baseURL: 'https://api.doubleword.ai/v1',\n defaultModel: 'moonshotai/Kimi-K2.6',\n smallFastModel: 'Qwen/Qwen3.5-9B',\n envKeys: ['DOUBLE_WORD_API_KEY', 'DOUBLEWORD_API_KEY'],\n },\n {\n id: 'deepseek',\n label: 'DeepSeek (api.deepseek.com)',\n kind: 'openai-compatible',\n baseURL: 'https://api.deepseek.com/v1',\n defaultModel: 'deepseek-chat',\n smallFastModel: 'deepseek-v4-flash',\n envKeys: ['DEEPSEEK_API_KEY'],\n },\n {\n // Z.ai — GLM family (Zhipu AI). OpenAI-compatible Chat Completions at\n // https://api.z.ai/api/paas/v4/chat/completions, authenticated with a Bearer\n // API key. GLM-5.2 is the current flagship coding/agent model with a 1M-token\n // context window; GLM-4.6 is 200K; GLM-4.5 / 4.5-Air are 128K. Thinking is\n // NATIVE: GLM-4.5+ emit chain-of-thought in the `reasoning_content` field,\n // which the OpenAI adapter surfaces as a thinking block. Z.ai has NO\n // OpenAI-style `GET /models`, so the full lineup comes from the curated\n // catalog in curatedProviderModels.ts (merged with any live fetch). Per-model\n // context windows live in KNOWN_MODEL_CONTEXT (rayuConfig.ts). Docs:\n // https://docs.z.ai/api-reference/llm/chat-completion\n id: 'glm',\n label: 'GLM — Z.ai (api.z.ai) · GLM-5.2 1M ctx / GLM-4.6 200K · native thinking',\n kind: 'openai-compatible',\n baseURL: 'https://api.z.ai/api/paas/v4',\n defaultModel: 'glm-5.2',\n smallFastModel: 'glm-4.5-air',\n envKeys: ['ZAI_API_KEY', 'ZHIPUAI_API_KEY', 'GLM_API_KEY'],\n },\n {\n // MiniMax — M-series language models. OpenAI-compatible Chat Completions at\n // https://api.minimax.io/v1/chat/completions (Bearer MINIMAX_API_KEY).\n // MiniMax-M3 is the frontier coding/agent model with a 1M-token context;\n // MiniMax-M2 / M2.x are 204,800. Thinking is NATIVE and ON by default\n // (M2.x always think; M3 thinks unless disabled): reasoning is returned\n // either as `reasoning_content` (reasoning_split) or inline <think>…</think>\n // at the start of the message — the OpenAI adapter surfaces BOTH as a\n // thinking block (processInlineThink). Per-model context lives in\n // KNOWN_MODEL_CONTEXT (rayuConfig.ts). Docs:\n // https://platform.minimax.io/docs/api-reference/text-openai-api\n id: 'minimax',\n label: 'MiniMax (api.minimax.io) · M3 1M ctx / M2 reasoning · native thinking',\n kind: 'openai-compatible',\n baseURL: 'https://api.minimax.io/v1',\n defaultModel: 'MiniMax-M2',\n smallFastModel: 'MiniMax-M2',\n envKeys: ['MINIMAX_API_KEY'],\n },\n {\n id: 'kimi-moonshot',\n label: 'Kimi / Moonshot (api.moonshot.ai)',\n kind: 'openai-compatible',\n baseURL: 'https://api.moonshot.ai/v1',\n defaultModel: 'kimi-k2.6',\n smallFastModel: 'kimi-k2.6',\n envKeys: ['KIMI_API_KEY', 'MOONSHOT_API_KEY'],\n },\n {\n id: 'kimi-for-code',\n label: 'Kimi for Code (api.kimi.com/coding)',\n kind: 'openai-compatible',\n baseURL: 'https://api.kimi.com/coding/v1',\n defaultModel: 'kimi-for-coding',\n smallFastModel: 'kimi-for-coding',\n envKeys: ['KIMI_FOR_CODE_API_KEY'],\n },\n {\n // Fugu — Sakana AI's multi-agent system (Japan). Although it's an agentic\n // multi-agent system under the hood, it's exposed like a standard LLM\n // through the OpenAI-compatible Sakana API: `POST /v1/chat/completions`\n // (and the Responses API) at https://api.sakana.ai/v1, authenticated with a\n // Bearer SAKANA_API_KEY. Rayu reuses its OpenAI-compatible adapter + live\n // `GET /v1/models` catalog verbatim. Two models ship: `fugu` (default;\n // routes across providers) and `fugu-ultra` (premium). Both have a 1M-token\n // context window — see KNOWN_MODEL_CONTEXT in rayuConfig.ts. Docs:\n // https://console.sakana.ai/get-started\n id: 'fugu',\n label: 'Fugu — Sakana AI (api.sakana.ai)',\n kind: 'openai-compatible',\n baseURL: 'https://api.sakana.ai/v1',\n defaultModel: 'fugu',\n smallFastModel: 'fugu',\n envKeys: ['SAKANA_API_KEY'],\n },\n {\n id: 'openai',\n label: 'OpenAI (api.openai.com)',\n kind: 'openai-compatible',\n baseURL: 'https://api.openai.com/v1',\n defaultModel: 'gpt-4o',\n smallFastModel: 'gpt-4o-mini',\n envKeys: ['OPENAI_API_KEY'],\n },\n {\n // Google Gemini via its OpenAI-compatible surface. The\n // generativelanguage endpoint exposes /chat/completions and /models under\n // the /v1beta/openai path, authenticated with a Gemini API key, so Rayu can\n // reuse its OpenAI-compatible adapter + live model-catalog fetch verbatim.\n // (The OAuth/Vertex path is a separate preset — see GEMINI_VERTEX_PRESET.)\n // No hardcoded defaultModel: the catalog is fetched live and the picker\n // chooses (e.g. gemini-2.5-flash / gemini-3.x-flash as they ship).\n id: 'gemini',\n label: 'Google Gemini — API key (generativelanguage.googleapis.com)',\n kind: 'openai-compatible',\n baseURL: 'https://generativelanguage.googleapis.com/v1beta/openai',\n envKeys: ['GEMINI_API_KEY', 'GOOGLE_API_KEY'],\n },\n {\n // Gemini on Google Vertex AI, authenticated with Google OAuth /\n // Application Default Credentials (no static API key). Project + region\n // scoped; the base URL is computed from the values chosen/detected in\n // /connect (vertexBaseURL). Chat is served through the OpenAI-compatible\n // adapter via a fetch wrapper that injects a fresh bearer token.\n id: GEMINI_VERTEX_PROVIDER_ID,\n label: 'Google Gemini — Vertex AI (OAuth / ADC) · recommended for heavy use',\n kind: 'vertex',\n requiresOAuth: true,\n },\n {\n // \"Login with Gemini\" — interactive Google sign-in (user's Desktop OAuth\n // client). Served by the @google/genai adapter in Vertex mode on the\n // `global` location, using the project from the client/login. No API key\n // or gcloud/ADC required.\n id: 'gemini-login',\n label: 'Login with Gemini (Google account) · free/Pro consumer quota',\n kind: 'genai',\n requiresOAuth: true,\n },\n {\n // GitHub Copilot — sign in with GitHub (OAuth device flow) and use your\n // Copilot subscription's models (Claude, GPT, Gemini, …) via Copilot's\n // OpenAI-compatible endpoint (api.githubcopilot.com). No API key to paste:\n // the GitHub OAuth token is exchanged for a short-lived Copilot token that\n // is auto-refreshed. Reuses the OpenAI adapter through a token-injecting\n // fetch wrapper (handled by kind:'copilot' in client.ts).\n id: 'copilot',\n label: 'GitHub Copilot (sign in with GitHub) · uses your Copilot subscription',\n kind: 'copilot',\n },\n {\n id: 'openrouter',\n label: 'OpenRouter (openrouter.ai)',\n kind: 'openai-compatible',\n baseURL: 'https://openrouter.ai/api/v1',\n defaultModel: 'anthropic/claude-3.5-sonnet',\n smallFastModel: 'openai/gpt-4o-mini',\n envKeys: ['OPENROUTER_API_KEY'],\n },\n {\n id: 'xai',\n label: 'xAI / Grok (api.x.ai)',\n kind: 'openai-compatible',\n baseURL: 'https://api.x.ai/v1',\n defaultModel: 'grok-4',\n smallFastModel: 'grok-3-mini',\n envKeys: ['XAI_API_KEY'],\n },\n {\n id: 'groq',\n label: 'Groq (api.groq.com)',\n kind: 'openai-compatible',\n baseURL: 'https://api.groq.com/openai/v1',\n defaultModel: 'llama-3.3-70b-versatile',\n smallFastModel: 'llama-3.1-8b-instant',\n envKeys: ['GROQ_API_KEY'],\n },\n {\n id: 'fireworks',\n label: 'Fireworks AI (api.fireworks.ai)',\n kind: 'openai-compatible',\n baseURL: 'https://api.fireworks.ai/inference/v1',\n defaultModel: 'accounts/fireworks/models/llama-v3p3-70b-instruct',\n envKeys: ['FIREWORKS_API_KEY'],\n },\n {\n id: 'togetherai',\n label: 'Together AI (api.together.xyz)',\n kind: 'openai-compatible',\n baseURL: 'https://api.together.xyz/v1',\n defaultModel: 'meta-llama/Llama-3.3-70B-Instruct-Turbo',\n envKeys: ['TOGETHER_API_KEY', 'TOGETHERAI_API_KEY'],\n },\n {\n id: 'cerebras',\n label: 'Cerebras (api.cerebras.ai)',\n kind: 'openai-compatible',\n baseURL: 'https://api.cerebras.ai/v1',\n defaultModel: 'llama-3.3-70b',\n envKeys: ['CEREBRAS_API_KEY'],\n },\n {\n id: 'baseten',\n label: 'Baseten (inference.baseten.co)',\n kind: 'openai-compatible',\n baseURL: 'https://inference.baseten.co/v1',\n envKeys: ['BASETEN_API_KEY'],\n },\n {\n id: 'deepinfra',\n label: 'DeepInfra (api.deepinfra.com)',\n kind: 'openai-compatible',\n baseURL: 'https://api.deepinfra.com/v1/openai',\n defaultModel: 'meta-llama/Llama-3.3-70B-Instruct',\n envKeys: ['DEEPINFRA_API_KEY'],\n },\n {\n // Hugging Face Inference Providers — an OpenAI-compatible router\n // (router.huggingface.co/v1) that fans a single `<org>/<model>` id out to\n // the available inference providers (novita, together, fireworks, …).\n // `GET /v1/models` returns the live cross-provider catalog, so Rayu's\n // generic openai-compatible adapter + model fetch work verbatim. Auth is a\n // Hugging Face access token (hf_…), read from the standard HF env vars.\n id: 'huggingface',\n label: 'Hugging Face — Inference Providers (router.huggingface.co)',\n kind: 'openai-compatible',\n baseURL: 'https://router.huggingface.co/v1',\n defaultModel: 'deepseek-ai/DeepSeek-V4-Pro',\n smallFastModel: 'deepseek-ai/DeepSeek-V4-Flash',\n envKeys: ['HF_TOKEN', 'HUGGINGFACE_API_KEY', 'HUGGING_FACE_HUB_TOKEN'],\n },\n {\n id: 'bedrock',\n label: 'AWS Bedrock — all models (Claude + open-weight)',\n kind: 'bedrock',\n // ONE Bedrock provider for the whole catalog. The wire format is chosen per\n // MODEL by resolveWireFormat(): Claude inference profiles speak the Anthropic\n // Messages API (bedrock-runtime invoke endpoints), and the models Bedrock\n // flags `openAiChatCompletions` speak OpenAI Chat Completions on the\n // bedrock-mantle endpoint. Region-scoped, so no fixed baseURL here — the\n // wizard sets the mantle baseURL after the region is chosen. Models are\n // fetched live from the Bedrock control plane.\n envKeys: ['AWS_BEARER_TOKEN_BEDROCK'],\n },\n {\n id: 'azure',\n label: 'Microsoft Azure — all deployments (Claude + Azure OpenAI)',\n kind: 'azure',\n // ONE Azure resource serving both wire formats, chosen per MODEL:\n // Claude deployments → Anthropic Messages at {origin}/anthropic;\n // everything else → Azure OpenAI v1 Responses at {origin}/openai/v1.\n // The endpoint is derived from the resource name entered in /connect, so no\n // fixed baseURL here. Deployments are fetched live.\n envKeys: ['ANTHROPIC_FOUNDRY_API_KEY', 'AZURE_OPENAI_API_KEY'],\n },\n {\n id: 'kiro',\n label: 'Kiro — Claude via AWS (API key or kiro-cli login)',\n kind: 'kiro',\n defaultModel: 'claude-sonnet-4.6',\n smallFastModel: 'claude-haiku-4.5',\n envKeys: ['KIRO_API_KEY'],\n // No fixed baseURL: region-scoped AWS CodeWhisperer endpoint, reached via a\n // dedicated adapter (src/services/api/kiroAdapter.ts). Auth + the adapter\n // are lazy-loaded only when the active provider is kind:'kiro'.\n },\n {\n id: 'ollama',\n label: 'Ollama (local · auto-detect)',\n kind: 'openai-compatible',\n baseURL: OLLAMA_DEFAULT_BASE_URL,\n // No envKeys / API key: Ollama needs none. Connected via /connect →\n // Localhost → Ollama, which probes the server and lists its pulled models.\n },\n {\n id: 'local',\n label: 'Local / custom OpenAI-compatible endpoint',\n kind: 'openai-compatible',\n promptBaseURL: true,\n },\n]\n\n\n\n/**\n * Import API keys from known env vars / .env into the provider config. Adds a\n * provider entry for each env key found that isn't already configured with a\n * key. Sets the first imported provider active only if none is active yet.\n * SECURITY: keys are written to the 0600 config file; never logged.\n */\nexport function migrateEnvKeysToConfig(): void {\n loadDotEnv()\n const cfg = loadRayuConfig()\n let changed = false\n for (const provider of cfg.providers) {\n const preset = PROVIDER_PRESETS.find(p => p.id === provider.id)\n if (!preset) continue\n if (!provider.baseURL && preset.baseURL) {\n provider.baseURL = preset.baseURL\n changed = true\n }\n if (!provider.defaultModel && preset.defaultModel) {\n provider.defaultModel = preset.defaultModel\n changed = true\n }\n if (!provider.smallFastModel && preset.smallFastModel) {\n provider.smallFastModel = preset.smallFastModel\n changed = true\n }\n }\n for (const preset of PROVIDER_PRESETS) {\n const key = preset.envKeys\n ?.map(k => process.env[k])\n .find(v => typeof v === 'string' && v.length > 0)\n if (!key) continue\n const existing = cfg.providers.find(p => p.id === preset.id)\n if (existing?.apiKey) continue // already configured with a key\n // Bedrock is region-scoped: derive region from AWS env (or default) and\n // compute the OpenAI-compatible bedrock-runtime base URL from it.\n const isBedrock = preset.kind === 'bedrock'\n const region = isBedrock\n ? (process.env.AWS_REGION ||\n process.env.AWS_DEFAULT_REGION ||\n DEFAULT_BEDROCK_REGION)\n : undefined\n const baseURL = isBedrock ? bedrockBaseURL(region as string) : preset.baseURL\n if (existing) {\n existing.apiKey = key\n existing.baseURL ??= baseURL\n existing.defaultModel ??= preset.defaultModel\n existing.smallFastModel ??= preset.smallFastModel\n if (isBedrock) existing.awsRegion ??= region\n } else {\n cfg.providers.push({\n id: preset.id,\n kind: preset.kind,\n apiKey: key,\n ...(baseURL ? { baseURL } : {}),\n ...(preset.defaultModel ? { defaultModel: preset.defaultModel } : {}),\n ...(preset.smallFastModel ? { smallFastModel: preset.smallFastModel } : {}),\n ...(isBedrock ? { awsRegion: region } : {}),\n })\n }\n changed = true\n }\n if (changed) {\n cfg.activeProvider ??= cfg.providers[0]?.id\n saveRayuConfig(cfg)\n }\n}\n\n\n// --- Provider display names (header / status line) --------------------------\n// Short, human-friendly names shown next to the model in the logo/header (e.g.\n// \"gemini-3.5-flash · Vertex AI\") in place of the generic \"API Usage Billing\".\n\n/** Short, branded names keyed by provider id. */\nconst PROVIDER_DISPLAY_NAMES: Record<string, string> = {\n anthropic: 'Anthropic',\n longcat: 'LongCat',\n nvidia: 'NVIDIA',\n doubleword: 'Doubleword',\n deepseek: 'DeepSeek',\n glm: 'GLM',\n minimax: 'MiniMax',\n 'kimi-moonshot': 'Kimi',\n 'kimi-for-code': 'Kimi for Code',\n fugu: 'Fugu',\n openai: 'OpenAI',\n gemini: 'Gemini',\n [GEMINI_VERTEX_PROVIDER_ID]: 'Vertex AI',\n 'gemini-login': 'Gemini',\n openrouter: 'OpenRouter',\n xai: 'xAI',\n groq: 'Groq',\n fireworks: 'Fireworks AI',\n togetherai: 'Together AI',\n cerebras: 'Cerebras',\n baseten: 'Baseten',\n deepinfra: 'DeepInfra',\n huggingface: 'Hugging Face',\n bedrock: 'AWS Bedrock',\n 'bedrock-openai': 'AWS Bedrock',\n 'bedrock-anthropic': 'AWS Bedrock',\n kiro: 'Kiro',\n copilot: 'GitHub Copilot',\n 'rayu-hosted': 'Rayu',\n ollama: 'Ollama',\n 'ollama-cloud': 'Ollama Cloud',\n local: 'Local',\n}\n\n/** Fallback name derived from the provider kind when the id isn't known. */\nfunction providerKindName(kind: ProviderKind): string {\n switch (kind) {\n case 'vertex':\n return 'Vertex AI'\n case 'bedrock':\n return 'AWS Bedrock'\n case 'genai':\n return 'Gemini'\n case 'kiro':\n return 'Kiro'\n case 'copilot':\n return 'GitHub Copilot'\n case 'anthropic':\n return 'Anthropic'\n case 'anthropic-compatible':\n return 'Anthropic-compatible'\n default:\n return 'OpenAI-compatible'\n }\n}\n\n/**\n * Human-friendly display name for a configured provider — id-aware, with a\n * kind-based fallback, plus a localhost hint so custom Ollama endpoints still\n * read as \"Ollama\".\n */\nexport function providerDisplayName(provider: RayuProvider): string {\n const known = PROVIDER_DISPLAY_NAMES[provider.id]\n if (known) return known\n if (\n provider.kind === 'openai-compatible' &&\n provider.baseURL &&\n /(?:\\/\\/|@)(?:localhost|127\\.0\\.0\\.1)(?::\\d+)?|:11434(?:\\/|$)/.test(\n provider.baseURL,\n )\n ) {\n return 'Ollama'\n }\n return providerKindName(provider.kind)\n}\n\n/**\n * Short display name for the ACTIVE provider (e.g. \"Vertex AI\", \"Ollama\",\n * \"NVIDIA\"), or undefined when no provider is configured yet.\n */\nexport function getActiveProviderDisplayName(): string | undefined {\n const p = getActiveProvider()\n return p ? providerDisplayName(p) : undefined\n}\n"}}, "rayu-backend/src/models/models.constants.ts": {"caca4043c11f75763539d7ad2adb24acfc0b11d8": {"t": 1785745603.355017, "n": 12759, "h": "54c58cf15e897564", "c": "import type { PlanCode } from '../common/enums'\n\nexport interface HostedModelSeed {\n code: string\n label: string\n /**\n * Name of the provider row (providers.name) this model routes through. All\n * routing config — base URL, wire format, auth scheme, key env — lives on that\n * provider, so a model only needs to name it plus its own upstream model id.\n */\n providerName: string\n upstreamModelId: string\n inputPricePer1MCents: number\n outputPricePer1MCents: number\n /** Credit charge for INPUT tokens (credits per 1M). */\n creditMultiplier: number\n /** Credit charge for OUTPUT tokens; defaults to the input charge. */\n outputCreditMultiplier?: number\n /** Credit charge for cache-HIT prompt tokens (absolute, not a ratio). */\n cacheReadCreditMultiplier?: number\n /** Credit charge for cache-CREATION prompt tokens; defaults to input. */\n cacheWriteCreditMultiplier?: number\n allowedPlanCodes: PlanCode[]\n /**\n * Context window in TOKENS (e.g. 200_000, 1_000_000). Optional: when omitted\n * the CLI falls back to its own default for the model. Admin-editable.\n */\n contextWindow?: number\n /** Model speaks a thinking/reasoning parameter. */\n supportsReasoning: boolean\n /** Model accepts image content blocks. */\n supportsImage: boolean\n /** Model accepts tool/function definitions. */\n supportsTools?: boolean\n enabled: boolean\n}\n\n// Provider names must match rows in the provider registry (see\n// providers/providers.constants.ts PROVIDER_SEED and migration\n// 0000000000009_providers). Renaming a provider in the dashboard detaches these\n// seeds — seedDefaults then logs a warning and skips, rather than failing boot.\nconst OLLAMA_PROVIDER = 'rayu-ollama'\n\n// First-time defaults only. Prices/multipliers/access are all admin-editable in\n// the dashboard afterwards; the seed is non-destructive (create-if-missing).\n// Rayu resells these via the (Phase 2) gateway using its own purchased keys.\nexport const MODEL_SEED: HostedModelSeed[] = [\n // DeepSeek V4 routed DIRECT to DeepSeek's own API via its Anthropic-compatible\n // endpoint (provider 'deepseek'). The gateway's built-in default for this\n // provider is Auth:x-api-key + Endpoint:anthropic (see knownProviderDefaults in\n // rayu-gateway/internal/config/config.go), so handleAnthropicMessages forwards\n // to https://api.deepseek.com/anthropic/v1/messages keyed by the gateway's\n // DEEPSEEK_API_KEY. This is the NATIVE rayu-hosted path — the CLI talks the\n // Anthropic SDK to the gateway's /anthropic base (see rayuHostedClient.ts), so\n // thinking / tools / prompt-cache usage map 1:1 with no OpenAI translation.\n // upstreamModelId is DeepSeek's own model id (NOT an Ollama `:cloud` tag):\n // deepseek-v4-pro → deepseek-v4-pro\n // deepseek-v4-flash → deepseek-v4-flash\n // NOTE: for this to route, the 'deepseek' provider row must be enabled (Admin →\n // Providers) and DEEPSEEK_API_KEY must be set in the gateway env.\n // input==output price → FLAT billing, so creditMultiplier is exactly credits\n // per 1M tokens (pro = 1.0 → 1 credit/1M; flash = 0.33). The CLI resolves the\n // deepseek-v4 codes to 1M context — add a per-model override if the real\n // model's window is smaller.\n {\n code: 'deepseek-v4-flash',\n label: 'DeepSeek V4 Flash',\n providerName: 'deepseek',\n upstreamModelId: 'deepseek-v4-flash', // DeepSeek API model id (Anthropic-compatible)\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40, // input==output → flat 0.33 credits / 1M tokens\n creditMultiplier: 0.33, // cheaper tier — ~1/3 the credit cost of Pro\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 1_000_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n {\n code: 'deepseek-v4-pro',\n label: 'DeepSeek V4 Pro',\n providerName: 'deepseek',\n upstreamModelId: 'deepseek-v4-pro', // DeepSeek API model id (Anthropic-compatible)\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40, // input==output → flat, exactly 1 credit / 1M tokens\n creditMultiplier: 1, // reference tier (1 credit / 1M tokens at baseline)\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 1_000_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n {\n // LongCat 2.0 (Meituan) — 1M-context, Anthropic-compatible API at\n // https://api.longcat.chat/anthropic/v1/messages (auth: Authorization:\n // Bearer, handled per-provider in the gateway). Real provider pricing:\n // uncached input $0.75/1M, cached input $0.015/1M (≈0.02× input), output\n // $2.95/1M. creditMultiplier is the USER-FACING credit charge and is only a\n // sensible starting default here — the admin tunes it (and every price) in\n // the dashboard; nothing about the charge is hardcoded downstream.\n code: 'longcat-2',\n label: 'LongCat 2.0',\n providerName: 'longcat',\n upstreamModelId: 'LongCat-2.0',\n inputPricePer1MCents: 75, // $0.75 / 1M (uncached input)\n outputPricePer1MCents: 295, // $2.95 / 1M\n cacheReadCreditMultiplier: 0.02, // cached $0.015 / uncached $0.75 ≈ 0.02\n creditMultiplier: 0.5, // starting default — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 1_000_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n\n // --- Ollama Cloud (provider = OLLAMA_PROVIDER, default 'rayu-ollama') -------\n // Ollama's HOSTED models (ollama.com), resold via the gateway with Rayu's own\n // OLLAMA_API_KEY. They speak Ollama's NATIVE Anthropic Messages API — but at\n // `{host}/v1/messages` (NO /anthropic path segment, unlike DeepSeek/LongCat)\n // with `Authorization: Bearer` — both handled per-provider in the gateway\n // (anthropicUpstream + anthropicUsesBearerAuth). Ollama does NOT do prompt\n // caching, so every prompt token bills at the input rate.\n //\n // creditMultiplier is the USER-FACING charge = CREDITS PER 1M TOKENS (baseline\n // is 1 credit / 1M tokens): 2.5 → 2.5 credits/1M, 0.75 → 0.75 credits/1M, etc.\n // input==output price keeps the charge FLAT (output tokens bill at the same\n // rate as input), matching the spec (\"1M tokens = N credits\"). Prices here are\n // internal cost-ledger placeholders (advisory) and, like every field, are\n // admin-editable in the dashboard — nothing is hardcoded downstream.\n //\n // NOTE: upstreamModelId must match the EXACT id in the Rayu Ollama Cloud\n // account (verify with `ollama list` / the Ollama dashboard); adjust per model\n // in the admin panel if a tag differs. Llama-4 and MiniMax-M3 credit rates were\n // not specified — defaulted (1.0 / 2.5) and are admin-tunable.\n {\n code: 'glm-5.2',\n label: 'GLM-5.2 (Ollama Cloud)',\n providerName: OLLAMA_PROVIDER,\n upstreamModelId: 'glm-5.2:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // 2.5 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 1_000_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n {\n code: 'kimi-k2.7',\n label: 'Kimi K2.7 Code (Ollama Cloud)',\n providerName: OLLAMA_PROVIDER,\n // Ollama's real id is 'kimi-k2.7-code' (NOT 'kimi-k2.7:cloud' — that 404s).\n // Verified against ollama.com/api/tags. NOTE: this model is subscription-gated\n // on Ollama (403 \"requires a subscription\") until the account's plan includes it.\n upstreamModelId: 'kimi-k2.7-code',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // 2.5 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 256_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n {\n code: 'minimax-m3',\n label: 'MiniMax M3 (Ollama Cloud)',\n providerName: OLLAMA_PROVIDER,\n upstreamModelId: 'minimax-m3:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // default (rate unspecified) — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 128_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n {\n code: 'llama-4',\n label: 'Llama 4 (Ollama Cloud)',\n providerName: OLLAMA_PROVIDER,\n upstreamModelId: 'llama4:cloud',\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40,\n creditMultiplier: 1, // default (rate unspecified) — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 128_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: false,\n supportsImage: true,\n enabled: true,\n },\n {\n code: 'gpt-oss-120b',\n label: 'GPT-OSS 120B (Ollama Cloud)',\n providerName: OLLAMA_PROVIDER,\n upstreamModelId: 'gpt-oss:120b-cloud',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 128_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n {\n code: 'qwen3.5-397b',\n label: 'Qwen3.5 397B (Ollama Cloud)',\n providerName: OLLAMA_PROVIDER,\n upstreamModelId: 'qwen3.5:397b-cloud',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 256_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n {\n code: 'qwen3.5-122b',\n label: 'Qwen3.5 122B (Ollama Cloud)',\n providerName: OLLAMA_PROVIDER,\n upstreamModelId: 'qwen3.5:122b',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 256_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n]\n"}, "9a3b056dbfac675c333915dd4a46a2b04559ec3b": {"t": 1785754691.44766, "n": 12759, "h": "54c58cf15e897564", "c": "import type { PlanCode } from '../common/enums'\n\nexport interface HostedModelSeed {\n code: string\n label: string\n /**\n * Name of the provider row (providers.name) this model routes through. All\n * routing config — base URL, wire format, auth scheme, key env — lives on that\n * provider, so a model only needs to name it plus its own upstream model id.\n */\n providerName: string\n upstreamModelId: string\n inputPricePer1MCents: number\n outputPricePer1MCents: number\n /** Credit charge for INPUT tokens (credits per 1M). */\n creditMultiplier: number\n /** Credit charge for OUTPUT tokens; defaults to the input charge. */\n outputCreditMultiplier?: number\n /** Credit charge for cache-HIT prompt tokens (absolute, not a ratio). */\n cacheReadCreditMultiplier?: number\n /** Credit charge for cache-CREATION prompt tokens; defaults to input. */\n cacheWriteCreditMultiplier?: number\n allowedPlanCodes: PlanCode[]\n /**\n * Context window in TOKENS (e.g. 200_000, 1_000_000). Optional: when omitted\n * the CLI falls back to its own default for the model. Admin-editable.\n */\n contextWindow?: number\n /** Model speaks a thinking/reasoning parameter. */\n supportsReasoning: boolean\n /** Model accepts image content blocks. */\n supportsImage: boolean\n /** Model accepts tool/function definitions. */\n supportsTools?: boolean\n enabled: boolean\n}\n\n// Provider names must match rows in the provider registry (see\n// providers/providers.constants.ts PROVIDER_SEED and migration\n// 0000000000009_providers). Renaming a provider in the dashboard detaches these\n// seeds — seedDefaults then logs a warning and skips, rather than failing boot.\nconst OLLAMA_PROVIDER = 'rayu-ollama'\n\n// First-time defaults only. Prices/multipliers/access are all admin-editable in\n// the dashboard afterwards; the seed is non-destructive (create-if-missing).\n// Rayu resells these via the (Phase 2) gateway using its own purchased keys.\nexport const MODEL_SEED: HostedModelSeed[] = [\n // DeepSeek V4 routed DIRECT to DeepSeek's own API via its Anthropic-compatible\n // endpoint (provider 'deepseek'). The gateway's built-in default for this\n // provider is Auth:x-api-key + Endpoint:anthropic (see knownProviderDefaults in\n // rayu-gateway/internal/config/config.go), so handleAnthropicMessages forwards\n // to https://api.deepseek.com/anthropic/v1/messages keyed by the gateway's\n // DEEPSEEK_API_KEY. This is the NATIVE rayu-hosted path — the CLI talks the\n // Anthropic SDK to the gateway's /anthropic base (see rayuHostedClient.ts), so\n // thinking / tools / prompt-cache usage map 1:1 with no OpenAI translation.\n // upstreamModelId is DeepSeek's own model id (NOT an Ollama `:cloud` tag):\n // deepseek-v4-pro → deepseek-v4-pro\n // deepseek-v4-flash → deepseek-v4-flash\n // NOTE: for this to route, the 'deepseek' provider row must be enabled (Admin →\n // Providers) and DEEPSEEK_API_KEY must be set in the gateway env.\n // input==output price → FLAT billing, so creditMultiplier is exactly credits\n // per 1M tokens (pro = 1.0 → 1 credit/1M; flash = 0.33). The CLI resolves the\n // deepseek-v4 codes to 1M context — add a per-model override if the real\n // model's window is smaller.\n {\n code: 'deepseek-v4-flash',\n label: 'DeepSeek V4 Flash',\n providerName: 'deepseek',\n upstreamModelId: 'deepseek-v4-flash', // DeepSeek API model id (Anthropic-compatible)\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40, // input==output → flat 0.33 credits / 1M tokens\n creditMultiplier: 0.33, // cheaper tier — ~1/3 the credit cost of Pro\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 1_000_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n {\n code: 'deepseek-v4-pro',\n label: 'DeepSeek V4 Pro',\n providerName: 'deepseek',\n upstreamModelId: 'deepseek-v4-pro', // DeepSeek API model id (Anthropic-compatible)\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40, // input==output → flat, exactly 1 credit / 1M tokens\n creditMultiplier: 1, // reference tier (1 credit / 1M tokens at baseline)\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 1_000_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n {\n // LongCat 2.0 (Meituan) — 1M-context, Anthropic-compatible API at\n // https://api.longcat.chat/anthropic/v1/messages (auth: Authorization:\n // Bearer, handled per-provider in the gateway). Real provider pricing:\n // uncached input $0.75/1M, cached input $0.015/1M (≈0.02× input), output\n // $2.95/1M. creditMultiplier is the USER-FACING credit charge and is only a\n // sensible starting default here — the admin tunes it (and every price) in\n // the dashboard; nothing about the charge is hardcoded downstream.\n code: 'longcat-2',\n label: 'LongCat 2.0',\n providerName: 'longcat',\n upstreamModelId: 'LongCat-2.0',\n inputPricePer1MCents: 75, // $0.75 / 1M (uncached input)\n outputPricePer1MCents: 295, // $2.95 / 1M\n cacheReadCreditMultiplier: 0.02, // cached $0.015 / uncached $0.75 ≈ 0.02\n creditMultiplier: 0.5, // starting default — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 1_000_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n\n // --- Ollama Cloud (provider = OLLAMA_PROVIDER, default 'rayu-ollama') -------\n // Ollama's HOSTED models (ollama.com), resold via the gateway with Rayu's own\n // OLLAMA_API_KEY. They speak Ollama's NATIVE Anthropic Messages API — but at\n // `{host}/v1/messages` (NO /anthropic path segment, unlike DeepSeek/LongCat)\n // with `Authorization: Bearer` — both handled per-provider in the gateway\n // (anthropicUpstream + anthropicUsesBearerAuth). Ollama does NOT do prompt\n // caching, so every prompt token bills at the input rate.\n //\n // creditMultiplier is the USER-FACING charge = CREDITS PER 1M TOKENS (baseline\n // is 1 credit / 1M tokens): 2.5 → 2.5 credits/1M, 0.75 → 0.75 credits/1M, etc.\n // input==output price keeps the charge FLAT (output tokens bill at the same\n // rate as input), matching the spec (\"1M tokens = N credits\"). Prices here are\n // internal cost-ledger placeholders (advisory) and, like every field, are\n // admin-editable in the dashboard — nothing is hardcoded downstream.\n //\n // NOTE: upstreamModelId must match the EXACT id in the Rayu Ollama Cloud\n // account (verify with `ollama list` / the Ollama dashboard); adjust per model\n // in the admin panel if a tag differs. Llama-4 and MiniMax-M3 credit rates were\n // not specified — defaulted (1.0 / 2.5) and are admin-tunable.\n {\n code: 'glm-5.2',\n label: 'GLM-5.2 (Ollama Cloud)',\n providerName: OLLAMA_PROVIDER,\n upstreamModelId: 'glm-5.2:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // 2.5 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 1_000_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n {\n code: 'kimi-k2.7',\n label: 'Kimi K2.7 Code (Ollama Cloud)',\n providerName: OLLAMA_PROVIDER,\n // Ollama's real id is 'kimi-k2.7-code' (NOT 'kimi-k2.7:cloud' — that 404s).\n // Verified against ollama.com/api/tags. NOTE: this model is subscription-gated\n // on Ollama (403 \"requires a subscription\") until the account's plan includes it.\n upstreamModelId: 'kimi-k2.7-code',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // 2.5 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 256_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n {\n code: 'minimax-m3',\n label: 'MiniMax M3 (Ollama Cloud)',\n providerName: OLLAMA_PROVIDER,\n upstreamModelId: 'minimax-m3:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // default (rate unspecified) — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 128_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n {\n code: 'llama-4',\n label: 'Llama 4 (Ollama Cloud)',\n providerName: OLLAMA_PROVIDER,\n upstreamModelId: 'llama4:cloud',\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40,\n creditMultiplier: 1, // default (rate unspecified) — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 128_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: false,\n supportsImage: true,\n enabled: true,\n },\n {\n code: 'gpt-oss-120b',\n label: 'GPT-OSS 120B (Ollama Cloud)',\n providerName: OLLAMA_PROVIDER,\n upstreamModelId: 'gpt-oss:120b-cloud',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 128_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n {\n code: 'qwen3.5-397b',\n label: 'Qwen3.5 397B (Ollama Cloud)',\n providerName: OLLAMA_PROVIDER,\n upstreamModelId: 'qwen3.5:397b-cloud',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 256_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n {\n code: 'qwen3.5-122b',\n label: 'Qwen3.5 122B (Ollama Cloud)',\n providerName: OLLAMA_PROVIDER,\n upstreamModelId: 'qwen3.5:122b',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n // Context window in tokens (admin-editable; the CLI budgets against it).\n contextWindow: 256_000,\n // Capability flags: the gateway rejects an image block / thinking field\n // for a model whose flag is false, and the CLI warns the user to switch\n // models. Admin-tunable per model in the dashboard.\n supportsReasoning: true,\n supportsImage: false,\n enabled: true,\n },\n]\n"}, "cea6b95d7872983602641db216bcd339c7cb8ce4": {"t": 1785754693.8501897, "n": 8522, "h": "e907a947e2bec1be", "c": "import type { PlanCode } from '../common/enums'\n\nexport interface HostedModelSeed {\n code: string\n label: string\n provider: string\n upstreamBaseUrl: string\n upstreamModelId: string\n inputPricePer1MCents: number\n outputPricePer1MCents: number\n creditMultiplier: number\n cacheReadCreditMultiplier?: number\n cacheWriteCreditMultiplier?: number\n allowedPlanCodes: PlanCode[]\n enabled: boolean\n}\n\n// Ollama Cloud's provider name is configurable via OLLAMA_PROVIDER_NAME (default\n// 'rayu-ollama') so it can be renamed without a code change. It MUST match the\n// gateway's OLLAMA_PROVIDER_NAME env (the gateway keys its Ollama routing +\n// OLLAMA_API_KEY off the same value); read once here at module load.\nconst OLLAMA_PROVIDER = process.env.OLLAMA_PROVIDER_NAME?.trim() || 'rayu-ollama'\n\n// First-time defaults only. Prices/multipliers/access are all admin-editable in\n// the dashboard afterwards; the seed is non-destructive (create-if-missing).\n// Rayu resells these via the (Phase 2) gateway using its own purchased keys.\nexport const MODEL_SEED: HostedModelSeed[] = [\n // DeepSeek V4 routed DIRECT to DeepSeek's own API via its Anthropic-compatible\n // endpoint (provider 'deepseek'). The gateway's built-in default for this\n // provider is Auth:x-api-key + Endpoint:anthropic (see knownProviderDefaults in\n // rayu-gateway/internal/config/config.go), so handleAnthropicMessages forwards\n // to https://api.deepseek.com/anthropic/v1/messages keyed by the gateway's\n // DEEPSEEK_API_KEY. This is the NATIVE rayu-hosted path — the CLI talks the\n // Anthropic SDK to the gateway's /anthropic base (see rayuHostedClient.ts), so\n // thinking / tools / prompt-cache usage map 1:1 with no OpenAI translation.\n // upstreamModelId is DeepSeek's own model id (NOT an Ollama `:cloud` tag):\n // deepseek-v4-pro → deepseek-v4-pro\n // deepseek-v4-flash → deepseek-v4-flash\n // NOTE: for this to route, 'deepseek' must NOT be in RAYU_DISABLED_PROVIDERS\n // and DEEPSEEK_API_KEY must be set in the gateway env.\n // input==output price → FLAT billing, so creditMultiplier is exactly credits\n // per 1M tokens (pro = 1.0 → 1 credit/1M; flash = 0.33). The CLI resolves the\n // deepseek-v4 codes to 1M context — add a per-model override if the real\n // model's window is smaller.\n {\n code: 'deepseek-v4-flash',\n label: 'DeepSeek V4 Flash',\n provider: 'deepseek',\n upstreamBaseUrl: 'https://api.deepseek.com',\n upstreamModelId: 'deepseek-v4-flash', // DeepSeek API model id (Anthropic-compatible)\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40, // input==output → flat 0.33 credits / 1M tokens\n creditMultiplier: 0.33, // cheaper tier — ~1/3 the credit cost of Pro\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'deepseek-v4-pro',\n label: 'DeepSeek V4 Pro',\n provider: 'deepseek',\n upstreamBaseUrl: 'https://api.deepseek.com',\n upstreamModelId: 'deepseek-v4-pro', // DeepSeek API model id (Anthropic-compatible)\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40, // input==output → flat, exactly 1 credit / 1M tokens\n creditMultiplier: 1, // reference tier (1 credit / 1M tokens at baseline)\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n // LongCat 2.0 (Meituan) — 1M-context, Anthropic-compatible API at\n // https://api.longcat.chat/anthropic/v1/messages (auth: Authorization:\n // Bearer, handled per-provider in the gateway). Real provider pricing:\n // uncached input $0.75/1M, cached input $0.015/1M (≈0.02× input), output\n // $2.95/1M. creditMultiplier is the USER-FACING credit charge and is only a\n // sensible starting default here — the admin tunes it (and every price) in\n // the dashboard; nothing about the charge is hardcoded downstream.\n code: 'longcat-2',\n label: 'LongCat 2.0',\n provider: 'longcat',\n upstreamBaseUrl: 'https://api.longcat.chat',\n upstreamModelId: 'LongCat-2.0',\n inputPricePer1MCents: 75, // $0.75 / 1M (uncached input)\n outputPricePer1MCents: 295, // $2.95 / 1M\n cacheReadCreditMultiplier: 0.02, // cached $0.015 / uncached $0.75 ≈ 0.02\n creditMultiplier: 0.5, // starting default — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n\n // --- Ollama Cloud (provider = OLLAMA_PROVIDER, default 'rayu-ollama') -------\n // Ollama's HOSTED models (ollama.com), resold via the gateway with Rayu's own\n // OLLAMA_API_KEY. They speak Ollama's NATIVE Anthropic Messages API — but at\n // `{host}/v1/messages` (NO /anthropic path segment, unlike DeepSeek/LongCat)\n // with `Authorization: Bearer` — both handled per-provider in the gateway\n // (anthropicUpstream + anthropicUsesBearerAuth). Ollama does NOT do prompt\n // caching, so every prompt token bills at the input rate.\n //\n // creditMultiplier is the USER-FACING charge = CREDITS PER 1M TOKENS (baseline\n // is 1 credit / 1M tokens): 2.5 → 2.5 credits/1M, 0.75 → 0.75 credits/1M, etc.\n // input==output price keeps the charge FLAT (output tokens bill at the same\n // rate as input), matching the spec (\"1M tokens = N credits\"). Prices here are\n // internal cost-ledger placeholders (advisory) and, like every field, are\n // admin-editable in the dashboard — nothing is hardcoded downstream.\n //\n // NOTE: upstreamModelId must match the EXACT id in the Rayu Ollama Cloud\n // account (verify with `ollama list` / the Ollama dashboard); adjust per model\n // in the admin panel if a tag differs. Llama-4 and MiniMax-M3 credit rates were\n // not specified — defaulted (1.0 / 2.5) and are admin-tunable.\n {\n code: 'glm-5.2',\n label: 'GLM-5.2 (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'glm-5.2:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // 2.5 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'kimi-k2.7',\n label: 'Kimi K2.7 Code (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n // Ollama's real id is 'kimi-k2.7-code' (NOT 'kimi-k2.7:cloud' — that 404s).\n // Verified against ollama.com/api/tags. NOTE: this model is subscription-gated\n // on Ollama (403 \"requires a subscription\") until the account's plan includes it.\n upstreamModelId: 'kimi-k2.7-code',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // 2.5 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'minimax-m3',\n label: 'MiniMax M3 (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'minimax-m3:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // default (rate unspecified) — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'llama-4',\n label: 'Llama 4 (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'llama4:cloud',\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40,\n creditMultiplier: 1, // default (rate unspecified) — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'gpt-oss-120b',\n label: 'GPT-OSS 120B (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'gpt-oss:120b-cloud',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'qwen3.5-397b',\n label: 'Qwen3.5 397B (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'qwen3.5:397b-cloud',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'qwen3.5-122b',\n label: 'Qwen3.5 122B (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'qwen3.5:122b',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n]\n"}, "2fae27b0f561997f7fb5919ec09dc6f8ca11478d": {"t": 1785754696.8194618, "n": 7862, "h": "424b4920af232775", "c": "import type { PlanCode } from '../common/enums'\n\nexport interface HostedModelSeed {\n code: string\n label: string\n provider: string\n upstreamBaseUrl: string\n upstreamModelId: string\n inputPricePer1MCents: number\n outputPricePer1MCents: number\n creditMultiplier: number\n cacheReadCreditMultiplier?: number\n cacheWriteCreditMultiplier?: number\n allowedPlanCodes: PlanCode[]\n enabled: boolean\n}\n\n// Ollama Cloud's provider name is configurable via OLLAMA_PROVIDER_NAME (default\n// 'rayu-ollama') so it can be renamed without a code change. It MUST match the\n// gateway's OLLAMA_PROVIDER_NAME env (the gateway keys its Ollama routing +\n// OLLAMA_API_KEY off the same value); read once here at module load.\nconst OLLAMA_PROVIDER = process.env.OLLAMA_PROVIDER_NAME?.trim() || 'rayu-ollama'\n\n// First-time defaults only. Prices/multipliers/access are all admin-editable in\n// the dashboard afterwards; the seed is non-destructive (create-if-missing).\n// Rayu resells these via the (Phase 2) gateway using its own purchased keys.\nexport const MODEL_SEED: HostedModelSeed[] = [\n // DeepSeek V4 routed through OLLAMA CLOUD (provider OLLAMA_PROVIDER, Rayu's\n // OLLAMA_API_KEY) — the official DeepSeek provider is DISABLED/dropped. The\n // upstreamModelId is the EXACT Ollama Cloud tag (`ollama run <id>`):\n // deepseek-v4-pro → deepseek-v4-pro:cloud\n // deepseek-v4-flash → deepseek-v4-flash:cloud\n // input==output price → FLAT billing, so creditMultiplier is exactly credits\n // per 1M tokens (pro = 1.0 → 1 credit/1M; flash = 0.33). The CLI resolves the\n // deepseek-v4 codes to 1M context — add a per-model override if the Ollama\n // model's window is smaller.\n {\n code: 'deepseek-v4-flash',\n label: 'DeepSeek V4 Flash',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'deepseek-v4-flash:cloud', // Ollama Cloud tag\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40, // input==output → flat 0.33 credits / 1M tokens\n creditMultiplier: 0.33, // cheaper tier — ~1/3 the credit cost of Pro\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'deepseek-v4-pro',\n label: 'DeepSeek V4 Pro',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'deepseek-v4-pro:cloud', // Ollama Cloud tag\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40, // input==output → flat, exactly 1 credit / 1M tokens\n creditMultiplier: 1, // reference tier (1 credit / 1M tokens at baseline)\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n // LongCat 2.0 (Meituan) — 1M-context, Anthropic-compatible API at\n // https://api.longcat.chat/anthropic/v1/messages (auth: Authorization:\n // Bearer, handled per-provider in the gateway). Real provider pricing:\n // uncached input $0.75/1M, cached input $0.015/1M (≈0.02× input), output\n // $2.95/1M. creditMultiplier is the USER-FACING credit charge and is only a\n // sensible starting default here — the admin tunes it (and every price) in\n // the dashboard; nothing about the charge is hardcoded downstream.\n code: 'longcat-2',\n label: 'LongCat 2.0',\n provider: 'longcat',\n upstreamBaseUrl: 'https://api.longcat.chat',\n upstreamModelId: 'LongCat-2.0',\n inputPricePer1MCents: 75, // $0.75 / 1M (uncached input)\n outputPricePer1MCents: 295, // $2.95 / 1M\n cacheReadCreditMultiplier: 0.02, // cached $0.015 / uncached $0.75 ≈ 0.02\n creditMultiplier: 0.5, // starting default — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n\n // --- Ollama Cloud (provider = OLLAMA_PROVIDER, default 'rayu-ollama') -------\n // Ollama's HOSTED models (ollama.com), resold via the gateway with Rayu's own\n // OLLAMA_API_KEY. They speak Ollama's NATIVE Anthropic Messages API — but at\n // `{host}/v1/messages` (NO /anthropic path segment, unlike DeepSeek/LongCat)\n // with `Authorization: Bearer` — both handled per-provider in the gateway\n // (anthropicUpstream + anthropicUsesBearerAuth). Ollama does NOT do prompt\n // caching, so every prompt token bills at the input rate.\n //\n // creditMultiplier is the USER-FACING charge = CREDITS PER 1M TOKENS (baseline\n // is 1 credit / 1M tokens): 2.5 → 2.5 credits/1M, 0.75 → 0.75 credits/1M, etc.\n // input==output price keeps the charge FLAT (output tokens bill at the same\n // rate as input), matching the spec (\"1M tokens = N credits\"). Prices here are\n // internal cost-ledger placeholders (advisory) and, like every field, are\n // admin-editable in the dashboard — nothing is hardcoded downstream.\n //\n // NOTE: upstreamModelId must match the EXACT id in the Rayu Ollama Cloud\n // account (verify with `ollama list` / the Ollama dashboard); adjust per model\n // in the admin panel if a tag differs. Llama-4 and MiniMax-M3 credit rates were\n // not specified — defaulted (1.0 / 2.5) and are admin-tunable.\n {\n code: 'glm-5.2',\n label: 'GLM-5.2 (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'glm-5.2:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // 2.5 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'kimi-k2.7',\n label: 'Kimi K2.7 Code (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n // Ollama's real id is 'kimi-k2.7-code' (NOT 'kimi-k2.7:cloud' — that 404s).\n // Verified against ollama.com/api/tags. NOTE: this model is subscription-gated\n // on Ollama (403 \"requires a subscription\") until the account's plan includes it.\n upstreamModelId: 'kimi-k2.7-code',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // 2.5 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'minimax-m3',\n label: 'MiniMax M3 (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'minimax-m3:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // default (rate unspecified) — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'llama-4',\n label: 'Llama 4 (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'llama4:cloud',\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40,\n creditMultiplier: 1, // default (rate unspecified) — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'gpt-oss-120b',\n label: 'GPT-OSS 120B (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'gpt-oss:120b-cloud',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'qwen3.5-397b',\n label: 'Qwen3.5 397B (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'qwen3.5:397b-cloud',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'qwen3.5-122b',\n label: 'Qwen3.5 122B (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'qwen3.5:122b',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n]\n"}, "8e6898484890bf5cc6d7fb499042f3e27000a2d7": {"t": 1785754702.3661318, "n": 7836, "h": "ecc0c69d8035d524", "c": "import type { PlanCode } from '../common/enums'\n\nexport interface HostedModelSeed {\n code: string\n label: string\n provider: string\n upstreamBaseUrl: string\n upstreamModelId: string\n inputPricePer1MCents: number\n outputPricePer1MCents: number\n creditMultiplier: number\n cacheReadCreditMultiplier?: number\n cacheWriteCreditMultiplier?: number\n allowedPlanCodes: PlanCode[]\n enabled: boolean\n}\n\n// Ollama Cloud's provider name is configurable via OLLAMA_PROVIDER_NAME (default\n// 'rayu-ollama') so it can be renamed without a code change. It MUST match the\n// gateway's OLLAMA_PROVIDER_NAME env (the gateway keys its Ollama routing +\n// OLLAMA_API_KEY off the same value); read once here at module load.\nconst OLLAMA_PROVIDER = process.env.OLLAMA_PROVIDER_NAME?.trim() || 'rayu-ollama'\n\n// First-time defaults only. Prices/multipliers/access are all admin-editable in\n// the dashboard afterwards; the seed is non-destructive (create-if-missing).\n// Rayu resells these via the (Phase 2) gateway using its own purchased keys.\nexport const MODEL_SEED: HostedModelSeed[] = [\n // DeepSeek V4 routed through OLLAMA CLOUD (provider OLLAMA_PROVIDER, Rayu's\n // OLLAMA_API_KEY) — the official DeepSeek route is dropped. input==output price\n // → FLAT billing, so creditMultiplier is exactly credits per 1M tokens\n // (pro = 1.0 → 1 credit/1M; flash = 0.33). NOTE: set upstreamModelId to the\n // EXACT deepseek id in your Ollama account (Ollama serves deepseek as e.g.\n // deepseek-v3.1:671b-cloud). The CLI resolves the deepseek-v4 codes to 1M\n // context — add a per-model override if the Ollama model's window is smaller.\n {\n code: 'deepseek-v4-flash',\n label: 'DeepSeek V4 Flash',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'deepseek-v3.1:671b-cloud', // verify/set exact Ollama id\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40, // input==output → flat 0.33 credits / 1M tokens\n creditMultiplier: 0.33, // cheaper tier — ~1/3 the credit cost of Pro\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'deepseek-v4-pro',\n label: 'DeepSeek V4 Pro',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'deepseek-v3.1:671b-cloud', // verify/set exact Ollama id\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40, // input==output → flat, exactly 1 credit / 1M tokens\n creditMultiplier: 1, // reference tier (1 credit / 1M tokens at baseline)\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n // LongCat 2.0 (Meituan) — 1M-context, Anthropic-compatible API at\n // https://api.longcat.chat/anthropic/v1/messages (auth: Authorization:\n // Bearer, handled per-provider in the gateway). Real provider pricing:\n // uncached input $0.75/1M, cached input $0.015/1M (≈0.02× input), output\n // $2.95/1M. creditMultiplier is the USER-FACING credit charge and is only a\n // sensible starting default here — the admin tunes it (and every price) in\n // the dashboard; nothing about the charge is hardcoded downstream.\n code: 'longcat-2',\n label: 'LongCat 2.0',\n provider: 'longcat',\n upstreamBaseUrl: 'https://api.longcat.chat',\n upstreamModelId: 'LongCat-2.0',\n inputPricePer1MCents: 75, // $0.75 / 1M (uncached input)\n outputPricePer1MCents: 295, // $2.95 / 1M\n cacheReadCreditMultiplier: 0.02, // cached $0.015 / uncached $0.75 ≈ 0.02\n creditMultiplier: 0.5, // starting default — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n\n // --- Ollama Cloud (provider = OLLAMA_PROVIDER, default 'rayu-ollama') -------\n // Ollama's HOSTED models (ollama.com), resold via the gateway with Rayu's own\n // OLLAMA_API_KEY. They speak Ollama's NATIVE Anthropic Messages API — but at\n // `{host}/v1/messages` (NO /anthropic path segment, unlike DeepSeek/LongCat)\n // with `Authorization: Bearer` — both handled per-provider in the gateway\n // (anthropicUpstream + anthropicUsesBearerAuth). Ollama does NOT do prompt\n // caching, so every prompt token bills at the input rate.\n //\n // creditMultiplier is the USER-FACING charge = CREDITS PER 1M TOKENS (baseline\n // is 1 credit / 1M tokens): 2.5 → 2.5 credits/1M, 0.75 → 0.75 credits/1M, etc.\n // input==output price keeps the charge FLAT (output tokens bill at the same\n // rate as input), matching the spec (\"1M tokens = N credits\"). Prices here are\n // internal cost-ledger placeholders (advisory) and, like every field, are\n // admin-editable in the dashboard — nothing is hardcoded downstream.\n //\n // NOTE: upstreamModelId must match the EXACT id in the Rayu Ollama Cloud\n // account (verify with `ollama list` / the Ollama dashboard); adjust per model\n // in the admin panel if a tag differs. Llama-4 and MiniMax-M3 credit rates were\n // not specified — defaulted (1.0 / 2.5) and are admin-tunable.\n {\n code: 'glm-5.2',\n label: 'GLM-5.2 (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'glm-5.2:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // 2.5 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'kimi-k2.7',\n label: 'Kimi K2.7 Code (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n // Ollama's real id is 'kimi-k2.7-code' (NOT 'kimi-k2.7:cloud' — that 404s).\n // Verified against ollama.com/api/tags. NOTE: this model is subscription-gated\n // on Ollama (403 \"requires a subscription\") until the account's plan includes it.\n upstreamModelId: 'kimi-k2.7-code',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // 2.5 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'minimax-m3',\n label: 'MiniMax M3 (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'minimax-m3:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // default (rate unspecified) — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'llama-4',\n label: 'Llama 4 (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'llama4:cloud',\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40,\n creditMultiplier: 1, // default (rate unspecified) — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'gpt-oss-120b',\n label: 'GPT-OSS 120B (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'gpt-oss:120b-cloud',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'qwen3.5-397b',\n label: 'Qwen3.5 397B (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'qwen3.5:397b-cloud',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'qwen3.5-122b',\n label: 'Qwen3.5 122B (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'qwen3.5:122b',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n]\n"}, "51221478230df27bb07700b8d7b9456e022d3bf4": {"t": 1785754703.7486646, "n": 7480, "h": "b54d8d13ca82c111", "c": "import type { PlanCode } from '../common/enums'\n\nexport interface HostedModelSeed {\n code: string\n label: string\n provider: string\n upstreamBaseUrl: string\n upstreamModelId: string\n inputPricePer1MCents: number\n outputPricePer1MCents: number\n creditMultiplier: number\n cacheReadCreditMultiplier?: number\n cacheWriteCreditMultiplier?: number\n allowedPlanCodes: PlanCode[]\n enabled: boolean\n}\n\n// Ollama Cloud's provider name is configurable via OLLAMA_PROVIDER_NAME (default\n// 'rayu-ollama') so it can be renamed without a code change. It MUST match the\n// gateway's OLLAMA_PROVIDER_NAME env (the gateway keys its Ollama routing +\n// OLLAMA_API_KEY off the same value); read once here at module load.\nconst OLLAMA_PROVIDER = process.env.OLLAMA_PROVIDER_NAME?.trim() || 'rayu-ollama'\n\n// First-time defaults only. Prices/multipliers/access are all admin-editable in\n// the dashboard afterwards; the seed is non-destructive (create-if-missing).\n// Rayu resells these via the (Phase 2) gateway using its own purchased keys.\nexport const MODEL_SEED: HostedModelSeed[] = [\n // DeepSeek V4 routed through OLLAMA CLOUD (provider OLLAMA_PROVIDER, Rayu's\n // OLLAMA_API_KEY) — the official DeepSeek route is dropped. ONLY the routing\n // changed; prices/multipliers are the prior admin-editable defaults (tune for\n // Ollama's real cost in the dashboard). NOTE: set upstreamModelId to the EXACT\n // deepseek id in your Ollama account (Ollama serves deepseek as e.g.\n // deepseek-v3.1:671b-cloud). The CLI resolves the deepseek-v4 codes to 1M\n // context — add a per-model override if the Ollama model's window is smaller.\n {\n code: 'deepseek-v4-flash',\n label: 'DeepSeek V4 Flash',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'deepseek-v3.1:671b-cloud', // verify/set exact Ollama id\n inputPricePer1MCents: 14,\n outputPricePer1MCents: 28,\n creditMultiplier: 0.33, // cheaper tier — ~1/3 the credit cost of Pro\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'deepseek-v4-pro',\n label: 'DeepSeek V4 Pro',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'deepseek-v3.1:671b-cloud', // verify/set exact Ollama id\n inputPricePer1MCents: 174,\n outputPricePer1MCents: 348,\n creditMultiplier: 1, // reference tier (1 credit / 1M tokens at baseline)\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n // LongCat 2.0 (Meituan) — 1M-context, Anthropic-compatible API at\n // https://api.longcat.chat/anthropic/v1/messages (auth: Authorization:\n // Bearer, handled per-provider in the gateway). Real provider pricing:\n // uncached input $0.75/1M, cached input $0.015/1M (≈0.02× input), output\n // $2.95/1M. creditMultiplier is the USER-FACING credit charge and is only a\n // sensible starting default here — the admin tunes it (and every price) in\n // the dashboard; nothing about the charge is hardcoded downstream.\n code: 'longcat-2',\n label: 'LongCat 2.0',\n provider: 'longcat',\n upstreamBaseUrl: 'https://api.longcat.chat',\n upstreamModelId: 'LongCat-2.0',\n inputPricePer1MCents: 75, // $0.75 / 1M (uncached input)\n outputPricePer1MCents: 295, // $2.95 / 1M\n cacheReadCreditMultiplier: 0.02, // cached $0.015 / uncached $0.75 ≈ 0.02\n creditMultiplier: 0.5, // starting default — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n\n // --- Ollama Cloud (provider = OLLAMA_PROVIDER, default 'rayu-ollama') -------\n // Ollama's HOSTED models (ollama.com), resold via the gateway with Rayu's own\n // OLLAMA_API_KEY. They speak Ollama's NATIVE Anthropic Messages API — but at\n // `{host}/v1/messages` (NO /anthropic path segment, unlike DeepSeek/LongCat)\n // with `Authorization: Bearer` — both handled per-provider in the gateway\n // (anthropicUpstream + anthropicUsesBearerAuth). Ollama does NOT do prompt\n // caching, so every prompt token bills at the input rate.\n //\n // creditMultiplier is the USER-FACING charge = CREDITS PER 1M TOKENS (baseline\n // is 1 credit / 1M tokens): 2.5 → 2.5 credits/1M, 0.75 → 0.75 credits/1M, etc.\n // input==output price keeps the charge FLAT (output tokens bill at the same\n // rate as input), matching the spec (\"1M tokens = N credits\"). Prices here are\n // internal cost-ledger placeholders (advisory) and, like every field, are\n // admin-editable in the dashboard — nothing is hardcoded downstream.\n //\n // NOTE: upstreamModelId must match the EXACT id in the Rayu Ollama Cloud\n // account (verify with `ollama list` / the Ollama dashboard); adjust per model\n // in the admin panel if a tag differs. Llama-4 and MiniMax-M3 credit rates were\n // not specified — defaulted (1.0 / 2.5) and are admin-tunable.\n {\n code: 'glm-5.2',\n label: 'GLM-5.2 (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'glm-5.2:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // 2.5 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'kimi-k2.7',\n label: 'Kimi K2.7 (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'kimi-k2.7:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // 2.5 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'minimax-m3',\n label: 'MiniMax M3 (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'minimax-m3:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // default (rate unspecified) — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'llama-4',\n label: 'Llama 4 (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'llama4:cloud',\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40,\n creditMultiplier: 1, // default (rate unspecified) — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'gpt-oss-120b',\n label: 'GPT-OSS 120B (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'gpt-oss:120b-cloud',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'qwen3.5-397b',\n label: 'Qwen3.5 397B (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'qwen3.5:397b-cloud',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'qwen3.5-122b',\n label: 'Qwen3.5 122B (Ollama Cloud)',\n provider: OLLAMA_PROVIDER,\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'qwen3.5:122b',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n]\n"}, "e76461b828d1a301a30d8a4c2d9afeea946a996e": {"t": 1785754705.0895526, "n": 6515, "h": "9f41f882bb385f45", "c": "import type { PlanCode } from '../common/enums'\n\nexport interface HostedModelSeed {\n code: string\n label: string\n provider: string\n upstreamBaseUrl: string\n upstreamModelId: string\n inputPricePer1MCents: number\n outputPricePer1MCents: number\n creditMultiplier: number\n cacheReadCreditMultiplier?: number\n cacheWriteCreditMultiplier?: number\n allowedPlanCodes: PlanCode[]\n enabled: boolean\n}\n\n// First-time defaults only. Prices/multipliers/access are all admin-editable in\n// the dashboard afterwards; the seed is non-destructive (create-if-missing).\n// Rayu resells these via the (Phase 2) gateway using its own purchased keys.\nexport const MODEL_SEED: HostedModelSeed[] = [\n {\n code: 'deepseek-v4-flash',\n label: 'DeepSeek V4 Flash',\n provider: 'deepseek',\n upstreamBaseUrl: 'https://api.deepseek.com/v1',\n upstreamModelId: 'deepseek-v4-flash',\n inputPricePer1MCents: 14, // $0.14 / 1M\n outputPricePer1MCents: 28, // $0.28 / 1M\n creditMultiplier: 0.33, // cheaper model — ~1/3 the credit cost of Pro\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'deepseek-v4-pro',\n label: 'DeepSeek V4 Pro',\n provider: 'deepseek',\n upstreamBaseUrl: 'https://api.deepseek.com/v1',\n upstreamModelId: 'deepseek-v4-pro',\n inputPricePer1MCents: 174, // $1.74 / 1M\n outputPricePer1MCents: 348, // $3.48 / 1M\n creditMultiplier: 1, // reference model (1 credit = 1e6/baselineCreditsPer1M tokens)\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n // LongCat 2.0 (Meituan) — 1M-context, Anthropic-compatible API at\n // https://api.longcat.chat/anthropic/v1/messages (auth: Authorization:\n // Bearer, handled per-provider in the gateway). Real provider pricing:\n // uncached input $0.75/1M, cached input $0.015/1M (≈0.02× input), output\n // $2.95/1M. creditMultiplier is the USER-FACING credit charge and is only a\n // sensible starting default here — the admin tunes it (and every price) in\n // the dashboard; nothing about the charge is hardcoded downstream.\n code: 'longcat-2',\n label: 'LongCat 2.0',\n provider: 'longcat',\n upstreamBaseUrl: 'https://api.longcat.chat',\n upstreamModelId: 'LongCat-2.0',\n inputPricePer1MCents: 75, // $0.75 / 1M (uncached input)\n outputPricePer1MCents: 295, // $2.95 / 1M\n cacheReadCreditMultiplier: 0.02, // cached $0.015 / uncached $0.75 ≈ 0.02\n creditMultiplier: 0.5, // starting default — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n\n // --- Ollama Cloud (provider 'rayu-ollama') ---------------------------\n // Ollama's HOSTED models (ollama.com), resold via the gateway with Rayu's own\n // OLLAMA_API_KEY. They speak Ollama's NATIVE Anthropic Messages API — but at\n // `{host}/v1/messages` (NO /anthropic path segment, unlike DeepSeek/LongCat)\n // with `Authorization: Bearer` — both handled per-provider in the gateway\n // (anthropicUpstream + anthropicUsesBearerAuth). Ollama does NOT do prompt\n // caching, so every prompt token bills at the input rate.\n //\n // creditMultiplier is the USER-FACING charge = CREDITS PER 1M TOKENS (baseline\n // is 1 credit / 1M tokens): 2.5 → 2.5 credits/1M, 0.75 → 0.75 credits/1M, etc.\n // input==output price keeps the charge FLAT (output tokens bill at the same\n // rate as input), matching the spec (\"1M tokens = N credits\"). Prices here are\n // internal cost-ledger placeholders (advisory) and, like every field, are\n // admin-editable in the dashboard — nothing is hardcoded downstream.\n //\n // NOTE: upstreamModelId must match the EXACT id in the Rayu Ollama Cloud\n // account (verify with `ollama list` / the Ollama dashboard); adjust per model\n // in the admin panel if a tag differs. Llama-4 and MiniMax-M3 credit rates were\n // not specified — defaulted (1.0 / 2.5) and are admin-tunable.\n {\n code: 'glm-5.2',\n label: 'GLM-5.2 (Ollama Cloud)',\n provider: 'rayu-ollama',\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'glm-5.2:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // 2.5 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'kimi-k2.7',\n label: 'Kimi K2.7 (Ollama Cloud)',\n provider: 'rayu-ollama',\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'kimi-k2.7:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // 2.5 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'minimax-m3',\n label: 'MiniMax M3 (Ollama Cloud)',\n provider: 'rayu-ollama',\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'minimax-m3:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // default (rate unspecified) — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'llama-4',\n label: 'Llama 4 (Ollama Cloud)',\n provider: 'rayu-ollama',\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'llama4:cloud',\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40,\n creditMultiplier: 1, // default (rate unspecified) — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'gpt-oss-120b',\n label: 'GPT-OSS 120B (Ollama Cloud)',\n provider: 'rayu-ollama',\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'gpt-oss:120b-cloud',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'qwen3.5-397b',\n label: 'Qwen3.5 397B (Ollama Cloud)',\n provider: 'rayu-ollama',\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'qwen3.5:397b-cloud',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'qwen3.5-122b',\n label: 'Qwen3.5 122B (Ollama Cloud)',\n provider: 'rayu-ollama',\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'qwen3.5:122b',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n]\n"}, "4e1f5068245d82111a1971373cc0b5f9c9869339": {"t": 1785754706.5963821, "n": 6563, "h": "3a829d68713d3314", "c": "import type { PlanCode } from '../common/enums'\n\nexport interface HostedModelSeed {\n code: string\n label: string\n provider: string\n upstreamBaseUrl: string\n upstreamModelId: string\n inputPricePer1MCents: number\n outputPricePer1MCents: number\n creditMultiplier: number\n cacheReadCreditMultiplier?: number\n cacheWriteCreditMultiplier?: number\n allowedPlanCodes: PlanCode[]\n enabled: boolean\n}\n\n// First-time defaults only. Prices/multipliers/access are all admin-editable in\n// the dashboard afterwards; the seed is non-destructive (create-if-missing).\n// Rayu resells these via the (Phase 2) gateway using its own purchased keys.\nexport const MODEL_SEED: HostedModelSeed[] = [\n {\n code: 'deepseek-v4-flash',\n label: 'DeepSeek V4 Flash',\n provider: 'deepseek',\n upstreamBaseUrl: 'https://api.deepseek.com/v1',\n upstreamModelId: 'deepseek-v4-flash',\n inputPricePer1MCents: 14, // $0.14 / 1M\n outputPricePer1MCents: 28, // $0.28 / 1M\n creditMultiplier: 0.33, // cheaper model — ~1/3 the credit cost of Pro\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'deepseek-v4-pro',\n label: 'DeepSeek V4 Pro',\n provider: 'deepseek',\n upstreamBaseUrl: 'https://api.deepseek.com/v1',\n upstreamModelId: 'deepseek-v4-pro',\n inputPricePer1MCents: 174, // $1.74 / 1M\n outputPricePer1MCents: 348, // $3.48 / 1M\n creditMultiplier: 1, // reference model (1 credit = 1e6/baselineCreditsPer1M tokens)\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n // LongCat 2.0 (Meituan) — 1M-context, Anthropic-compatible API at\n // https://api.longcat.chat/anthropic/v1/messages (auth: Authorization:\n // Bearer, handled per-provider in the gateway). Real provider pricing:\n // uncached input $0.75/1M, cached input $0.015/1M (≈0.02× input), output\n // $2.95/1M. creditMultiplier is the USER-FACING credit charge and is only a\n // sensible starting default here — the admin tunes it (and every price) in\n // the dashboard; nothing about the charge is hardcoded downstream.\n code: 'longcat-2',\n label: 'LongCat 2.0',\n provider: 'longcat',\n upstreamBaseUrl: 'https://api.longcat.chat',\n upstreamModelId: 'LongCat-2.0',\n inputPricePer1MCents: 75, // $0.75 / 1M (uncached input)\n outputPricePer1MCents: 295, // $2.95 / 1M\n cacheReadCreditMultiplier: 0.02, // cached $0.015 / uncached $0.75 ≈ 0.02\n creditMultiplier: 0.5, // starting default — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n\n // --- Ollama Cloud (provider 'rayu-ollama-cloud') ---------------------------\n // Ollama's HOSTED models (ollama.com), resold via the gateway with Rayu's own\n // OLLAMA_API_KEY. They speak Ollama's NATIVE Anthropic Messages API — but at\n // `{host}/v1/messages` (NO /anthropic path segment, unlike DeepSeek/LongCat)\n // with `Authorization: Bearer` — both handled per-provider in the gateway\n // (anthropicUpstream + anthropicUsesBearerAuth). Ollama does NOT do prompt\n // caching, so every prompt token bills at the input rate.\n //\n // creditMultiplier is the USER-FACING charge = CREDITS PER 1M TOKENS (baseline\n // is 1 credit / 1M tokens): 2.5 → 2.5 credits/1M, 0.75 → 0.75 credits/1M, etc.\n // input==output price keeps the charge FLAT (output tokens bill at the same\n // rate as input), matching the spec (\"1M tokens = N credits\"). Prices here are\n // internal cost-ledger placeholders (advisory) and, like every field, are\n // admin-editable in the dashboard — nothing is hardcoded downstream.\n //\n // NOTE: upstreamModelId must match the EXACT id in the Rayu Ollama Cloud\n // account (verify with `ollama list` / the Ollama dashboard); adjust per model\n // in the admin panel if a tag differs. Llama-4 and MiniMax-M3 credit rates were\n // not specified — defaulted (1.0 / 2.5) and are admin-tunable.\n {\n code: 'glm-5.2',\n label: 'GLM-5.2 (Ollama Cloud)',\n provider: 'rayu-ollama-cloud',\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'glm-5.2:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // 2.5 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'kimi-k2.7',\n label: 'Kimi K2.7 (Ollama Cloud)',\n provider: 'rayu-ollama-cloud',\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'kimi-k2.7:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // 2.5 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'minimax-m3',\n label: 'MiniMax M3 (Ollama Cloud)',\n provider: 'rayu-ollama-cloud',\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'minimax-m3:cloud',\n inputPricePer1MCents: 60,\n outputPricePer1MCents: 60,\n creditMultiplier: 2.5, // default (rate unspecified) — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'llama-4',\n label: 'Llama 4 (Ollama Cloud)',\n provider: 'rayu-ollama-cloud',\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'llama4:cloud',\n inputPricePer1MCents: 40,\n outputPricePer1MCents: 40,\n creditMultiplier: 1, // default (rate unspecified) — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'gpt-oss-120b',\n label: 'GPT-OSS 120B (Ollama Cloud)',\n provider: 'rayu-ollama-cloud',\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'gpt-oss:120b-cloud',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'qwen3.5-397b',\n label: 'Qwen3.5 397B (Ollama Cloud)',\n provider: 'rayu-ollama-cloud',\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'qwen3.5:397b-cloud',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'qwen3.5-122b',\n label: 'Qwen3.5 122B (Ollama Cloud)',\n provider: 'rayu-ollama-cloud',\n upstreamBaseUrl: 'https://ollama.com',\n upstreamModelId: 'qwen3.5:122b',\n inputPricePer1MCents: 30,\n outputPricePer1MCents: 30,\n creditMultiplier: 0.75, // 0.75 credits / 1M tokens\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n]\n"}, "dabb71d7ff1e4bd2504de29c3fd50b5fa01bd4fa": {"t": 1785754708.8189723, "n": 2569, "h": "ed89bfb2777d4057", "c": "import type { PlanCode } from '../common/enums'\n\nexport interface HostedModelSeed {\n code: string\n label: string\n provider: string\n upstreamBaseUrl: string\n upstreamModelId: string\n inputPricePer1MCents: number\n outputPricePer1MCents: number\n creditMultiplier: number\n cacheReadCreditMultiplier?: number\n cacheWriteCreditMultiplier?: number\n allowedPlanCodes: PlanCode[]\n enabled: boolean\n}\n\n// First-time defaults only. Prices/multipliers/access are all admin-editable in\n// the dashboard afterwards; the seed is non-destructive (create-if-missing).\n// Rayu resells these via the (Phase 2) gateway using its own purchased keys.\nexport const MODEL_SEED: HostedModelSeed[] = [\n {\n code: 'deepseek-v4-flash',\n label: 'DeepSeek V4 Flash',\n provider: 'deepseek',\n upstreamBaseUrl: 'https://api.deepseek.com/v1',\n upstreamModelId: 'deepseek-v4-flash',\n inputPricePer1MCents: 14, // $0.14 / 1M\n outputPricePer1MCents: 28, // $0.28 / 1M\n creditMultiplier: 0.33, // cheaper model — ~1/3 the credit cost of Pro\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'deepseek-v4-pro',\n label: 'DeepSeek V4 Pro',\n provider: 'deepseek',\n upstreamBaseUrl: 'https://api.deepseek.com/v1',\n upstreamModelId: 'deepseek-v4-pro',\n inputPricePer1MCents: 174, // $1.74 / 1M\n outputPricePer1MCents: 348, // $3.48 / 1M\n creditMultiplier: 1, // reference model (1 credit = 1e6/baselineCreditsPer1M tokens)\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n // LongCat 2.0 (Meituan) — 1M-context, Anthropic-compatible API at\n // https://api.longcat.chat/anthropic/v1/messages (auth: Authorization:\n // Bearer, handled per-provider in the gateway). Real provider pricing:\n // uncached input $0.75/1M, cached input $0.015/1M (≈0.02× input), output\n // $2.95/1M. creditMultiplier is the USER-FACING credit charge and is only a\n // sensible starting default here — the admin tunes it (and every price) in\n // the dashboard; nothing about the charge is hardcoded downstream.\n code: 'longcat-2',\n label: 'LongCat 2.0',\n provider: 'longcat',\n upstreamBaseUrl: 'https://api.longcat.chat',\n upstreamModelId: 'LongCat-2.0',\n inputPricePer1MCents: 75, // $0.75 / 1M (uncached input)\n outputPricePer1MCents: 295, // $2.95 / 1M\n cacheReadCreditMultiplier: 0.02, // cached $0.015 / uncached $0.75 ≈ 0.02\n creditMultiplier: 0.5, // starting default — admin-tunable\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n]\n"}, "72ac5f58c44c3c6b970447d63e5c1f3cb9117e46": {"t": 1785754709.963612, "n": 1550, "h": "e7863ea405bd4a02", "c": "import type { PlanCode } from '../common/enums'\n\nexport interface HostedModelSeed {\n code: string\n label: string\n provider: string\n upstreamBaseUrl: string\n upstreamModelId: string\n inputPricePer1MCents: number\n outputPricePer1MCents: number\n creditMultiplier: number\n cacheReadCreditMultiplier?: number\n cacheWriteCreditMultiplier?: number\n allowedPlanCodes: PlanCode[]\n enabled: boolean\n}\n\n// First-time defaults only. Prices/multipliers/access are all admin-editable in\n// the dashboard afterwards; the seed is non-destructive (create-if-missing).\n// Rayu resells these via the (Phase 2) gateway using its own purchased keys.\nexport const MODEL_SEED: HostedModelSeed[] = [\n {\n code: 'deepseek-v4-flash',\n label: 'DeepSeek V4 Flash',\n provider: 'deepseek',\n upstreamBaseUrl: 'https://api.deepseek.com/v1',\n upstreamModelId: 'deepseek-v4-flash',\n inputPricePer1MCents: 14, // $0.14 / 1M\n outputPricePer1MCents: 28, // $0.28 / 1M\n creditMultiplier: 0.33, // cheaper model — ~1/3 the credit cost of Pro\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'deepseek-v4-pro',\n label: 'DeepSeek V4 Pro',\n provider: 'deepseek',\n upstreamBaseUrl: 'https://api.deepseek.com/v1',\n upstreamModelId: 'deepseek-v4-pro',\n inputPricePer1MCents: 174, // $1.74 / 1M\n outputPricePer1MCents: 348, // $3.48 / 1M\n creditMultiplier: 1, // reference model (1 credit = 1e6/baselineCreditsPer1M tokens)\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n]\n"}, "c61c6ddc5e3359a9f42e0808685b525d125784d7": {"t": 1785754711.375495, "n": 1475, "h": "767aec87c5d335ba", "c": "import type { PlanCode } from '../common/enums'\n\nexport interface HostedModelSeed {\n code: string\n label: string\n provider: string\n upstreamBaseUrl: string\n upstreamModelId: string\n inputPricePer1MCents: number\n outputPricePer1MCents: number\n creditMultiplier: number\n allowedPlanCodes: PlanCode[]\n enabled: boolean\n}\n\n// First-time defaults only. Prices/multipliers/access are all admin-editable in\n// the dashboard afterwards; the seed is non-destructive (create-if-missing).\n// Rayu resells these via the (Phase 2) gateway using its own purchased keys.\nexport const MODEL_SEED: HostedModelSeed[] = [\n {\n code: 'deepseek-v4-flash',\n label: 'DeepSeek V4 Flash',\n provider: 'deepseek',\n upstreamBaseUrl: 'https://api.deepseek.com/v1',\n upstreamModelId: 'deepseek-v4-flash',\n inputPricePer1MCents: 14, // $0.14 / 1M\n outputPricePer1MCents: 28, // $0.28 / 1M\n creditMultiplier: 0.33, // cheaper model — ~1/3 the credit cost of Pro\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'deepseek-v4-pro',\n label: 'DeepSeek V4 Pro',\n provider: 'deepseek',\n upstreamBaseUrl: 'https://api.deepseek.com/v1',\n upstreamModelId: 'deepseek-v4-pro',\n inputPricePer1MCents: 174, // $1.74 / 1M\n outputPricePer1MCents: 348, // $3.48 / 1M\n creditMultiplier: 1, // reference model (1 credit = 1e6/baselineCreditsPer1M tokens)\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n]\n"}, "d6ae938492fecea149595f05dd13d82ddd993932": {"t": 1785754713.0222273, "n": 1361, "h": "0a998228135c84dd", "c": "import type { PlanCode } from '../common/enums'\n\nexport interface HostedModelSeed {\n code: string\n label: string\n provider: string\n upstreamBaseUrl: string\n upstreamModelId: string\n inputPricePer1MCents: number\n outputPricePer1MCents: number\n creditMultiplier: number\n allowedPlanCodes: PlanCode[]\n enabled: boolean\n}\n\n// First-time defaults only. Prices/multipliers/access are all admin-editable in\n// the dashboard afterwards; the seed is non-destructive (create-if-missing).\n// Rayu resells these via the (Phase 2) gateway using its own purchased keys.\nexport const MODEL_SEED: HostedModelSeed[] = [\n {\n code: 'deepseek-v4-flash',\n label: 'DeepSeek V4 Flash',\n provider: 'deepseek',\n upstreamBaseUrl: 'https://api.deepseek.com/v1',\n upstreamModelId: 'deepseek-v4-flash',\n inputPricePer1MCents: 14, // $0.14 / 1M\n outputPricePer1MCents: 28, // $0.28 / 1M\n creditMultiplier: 1,\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n {\n code: 'deepseek-v4-pro',\n label: 'DeepSeek V4 Pro',\n provider: 'deepseek',\n upstreamBaseUrl: 'https://api.deepseek.com/v1',\n upstreamModelId: 'deepseek-v4-pro',\n inputPricePer1MCents: 174, // $1.74 / 1M\n outputPricePer1MCents: 348, // $3.48 / 1M\n creditMultiplier: 3,\n allowedPlanCodes: ['pro', 'pro_plus', 'max'],\n enabled: true,\n },\n]\n"}}}} |