promptimizer 0.1.0

package/README.md

@@ -0,0 +1,63 @@
+# promptimizer
+
+OpenAI-compatible TypeScript SDK for [Promptimizer](https://hackathon-omega-liart.vercel.app). Route each request to the cheapest adequate model, cache repeated prefixes, and keep a quality gate so savings do not silently degrade hard answers.
+
+```bash
+npm install promptimizer
+```
+
+## Account + BYOK
+
+Create a `pmz_live_` key at `/account`, connect a provider in the console or CLI, then:
+
+```ts
+import { Promptimizer } from "promptimizer";
+
+const client = new Promptimizer({
+  apiKey: process.env.PROMPTIMIZER_API_KEY,
+});
+
+const completion = await client.chat.completions.create({
+  messages: [{ role: "user", content: "What is 17 * 24?" }],
+});
+
+console.log(completion.choices[0].message.content);
+console.log(completion.promptimizer);
+console.log(completion.usage.cost);
+```
+
+The default gateway is the hosted app. Override with `gatewayURL` or `PROMPTIMIZER_URL`.
+
+Connect a provider from code if it is not already saved on the account:
+
+```ts
+await client.connect({
+  provider: "baseten",
+  apiKey: process.env.BASETEN_API_KEY,
+});
+```
+
+## Drop-in with the official OpenAI SDK
+
+```ts
+import OpenAI from "openai";
+
+const openai = new OpenAI({
+  apiKey: process.env.PROMPTIMIZER_API_KEY,
+  baseURL: "https://hackathon-omega-liart.vercel.app/api/v1",
+});
+
+await openai.chat.completions.create({
+  model: "auto",
+  messages: [{ role: "user", content: "Explain REST in two sentences." }],
+});
+```
+
+## Local classification (no network)
+
+```ts
+import { classifyText } from "promptimizer";
+
+classifyText("Design a rate limiter for 1 million QPS");
+// recommended_tier: "frontier"
+```

package/dist/classify.d.ts

@@ -0,0 +1,4 @@
+import type { ChatMessage, Classification, Tier } from "./types";
+export declare function classifyMessages(messages: ChatMessage[]): Classification;
+export declare function classifyText(text: string): Classification;
+export declare function difficultyTier(complexity: number): Tier;

package/dist/client.d.ts

@@ -0,0 +1,54 @@
+import type { ChatCompletion, ChatCompletionRequest, ConnectOptions, PromptimizerOptions, SavingsSummary, Session } from "./types";
+export declare class PromptimizerError extends Error {
+    readonly status: number;
+    readonly body?: unknown | undefined;
+    constructor(message: string, status: number, body?: unknown | undefined);
+}
+export declare const DEFAULT_GATEWAY: string;
+export declare class Promptimizer {
+    readonly gatewayURL: string;
+    sessionId?: string;
+    apiKey?: string;
+    private readonly fetcher;
+    constructor(options?: PromptimizerOptions);
+    static connect(options: ConnectOptions & {
+        gatewayURL?: string;
+    }): Promise<{
+        client: Promptimizer;
+        session: Session;
+    }>;
+    connect(options: ConnectOptions): Promise<Session>;
+    providers(): Promise<{
+        object: string;
+        data: Array<{
+            id: string;
+            label: string;
+            base_url: string;
+            env: string;
+        }>;
+    }>;
+    savings(): Promise<SavingsSummary>;
+    session(): Promise<Session>;
+    models(): Promise<{
+        object: string;
+        data: Session["models"];
+        baseline_model: string | null;
+    }>;
+    updateFleet(body: {
+        overrides?: Record<string, string>;
+        selected?: Record<string, boolean>;
+        baseline_model?: string;
+    }): Promise<Session>;
+    classify(input: {
+        messages?: ChatCompletionRequest["messages"];
+        prompt?: string;
+    }): Promise<unknown>;
+    benchmark(compareAlwaysFrontier?: boolean): Promise<unknown>;
+    readonly chat: {
+        completions: {
+            create: (request: ChatCompletionRequest) => Promise<ChatCompletion>;
+        };
+    };
+    createChatCompletion(request: ChatCompletionRequest): Promise<ChatCompletion>;
+    private request;
+}

package/dist/index.cjs

@@ -0,0 +1,315 @@
+"use strict";
+var __defProp = Object.defineProperty;
+var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
+var __getOwnPropNames = Object.getOwnPropertyNames;
+var __hasOwnProp = Object.prototype.hasOwnProperty;
+var __export = (target, all) => {
+  for (var name in all)
+    __defProp(target, name, { get: all[name], enumerable: true });
+};
+var __copyProps = (to, from, except, desc) => {
+  if (from && typeof from === "object" || typeof from === "function") {
+    for (let key of __getOwnPropNames(from))
+      if (!__hasOwnProp.call(to, key) && key !== except)
+        __defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
+  }
+  return to;
+};
+var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
+
+// src/index.ts
+var index_exports = {};
+__export(index_exports, {
+  DEFAULT_GATEWAY: () => DEFAULT_GATEWAY,
+  PROVIDERS: () => PROVIDERS,
+  Promptimizer: () => Promptimizer,
+  PromptimizerError: () => PromptimizerError,
+  classifyMessages: () => classifyMessages,
+  classifyText: () => classifyText,
+  difficultyTier: () => difficultyTier,
+  findProvider: () => findProvider,
+  publicCatalog: () => publicCatalog,
+  resolveBaseURL: () => resolveBaseURL
+});
+module.exports = __toCommonJS(index_exports);
+
+// src/client.ts
+var PromptimizerError = class extends Error {
+  constructor(message, status, body) {
+    super(message);
+    this.status = status;
+    this.body = body;
+    this.name = "PromptimizerError";
+  }
+};
+var DEFAULT_GATEWAY = (process.env.PROMPTIMIZER_URL || "https://hackathon-omega-liart.vercel.app/api").replace(/\/$/, "");
+var Promptimizer = class _Promptimizer {
+  gatewayURL;
+  sessionId;
+  apiKey;
+  fetcher;
+  constructor(options = {}) {
+    this.gatewayURL = (options.gatewayURL ?? options.baseURL ?? DEFAULT_GATEWAY).replace(/\/$/, "");
+    this.sessionId = options.sessionId;
+    this.apiKey = options.apiKey;
+    this.fetcher = options.fetch ?? fetch;
+  }
+  static async connect(options) {
+    const client = new _Promptimizer({
+      gatewayURL: options.gatewayURL,
+      apiKey: options.accountKey
+    });
+    const session = await client.connect(options);
+    return { client, session };
+  }
+  async connect(options) {
+    const session = await this.request("/v1/providers/connect", {
+      method: "POST",
+      body: JSON.stringify({
+        mode: options.mode ?? "byok",
+        label: options.label,
+        provider: options.provider,
+        base_url: options.baseURL,
+        api_key: options.apiKey
+      })
+    });
+    this.sessionId = session.session_id;
+    return session;
+  }
+  async providers() {
+    return this.request(
+      "/v1/providers"
+    );
+  }
+  async savings() {
+    return this.request("/v1/savings");
+  }
+  async session() {
+    return this.request("/v1/session");
+  }
+  async models() {
+    return this.request(
+      "/v1/models"
+    );
+  }
+  async updateFleet(body) {
+    return this.request("/v1/models", { method: "PATCH", body: JSON.stringify(body) });
+  }
+  async classify(input) {
+    return this.request("/v1/classify", { method: "POST", body: JSON.stringify(input) });
+  }
+  async benchmark(compareAlwaysFrontier = true) {
+    return this.request("/v1/benchmark/run", {
+      method: "POST",
+      body: JSON.stringify({ compare_always_frontier: compareAlwaysFrontier })
+    });
+  }
+  chat = {
+    completions: {
+      create: (request) => this.createChatCompletion(request)
+    }
+  };
+  async createChatCompletion(request) {
+    return this.request("/v1/chat/completions", {
+      method: "POST",
+      body: JSON.stringify(request)
+    });
+  }
+  async request(path, init = {}) {
+    const headers = new Headers(init.headers);
+    headers.set("Content-Type", "application/json");
+    if (this.apiKey) {
+      headers.set("Authorization", `Bearer ${this.apiKey}`);
+    } else if (this.sessionId) {
+      headers.set("X-Promptimizer-Session", this.sessionId);
+      headers.set("Authorization", `Bearer ${this.sessionId}`);
+    }
+    const response = await this.fetcher(`${this.gatewayURL}${path}`, { ...init, headers });
+    const data = await response.json().catch(() => ({}));
+    if (!response.ok) {
+      const detail = typeof data === "object" && data && "detail" in data ? String(data.detail) : response.statusText;
+      throw new PromptimizerError(detail, response.status, data);
+    }
+    return data;
+  }
+};
+
+// src/classify.ts
+var HIGH_RISK = /* @__PURE__ */ new Set(["system_design", "safety_sensitive", "code_debug", "reasoning"]);
+function textFrom(messages) {
+  return messages.map((m) => {
+    if (typeof m.content === "string") return m.content;
+    return m.content.map((b) => b.text ?? "").join("\n");
+  }).join("\n").trim();
+}
+function classifyMessages(messages) {
+  return classifyText(textFrom(messages));
+}
+function classifyText(text) {
+  const words = text.split(/\s+/).filter(Boolean).length;
+  const lines = text.split("\n").length;
+  const features = {
+    code_fence: /```/.test(text),
+    code_kw: /\b(def |class |function |import |fn |pub |async |SELECT |goroutine|mutex|traceback)\b/i.test(
+      text
+    ),
+    math: /(\$\$|\\frac|prove that|expected value|O\([nN]\)|\d+\s*[\*\^]\s*\d+)/i.test(text),
+    design: /\b(design|architect|rate limiter|distributed|consistency|shard|1 million QPS)\b/i.test(
+      text
+    ),
+    reason: /\b(prove|why does|walk through|step by step|derive|contradiction|p-value)\b/i.test(
+      text
+    ),
+    debug: /\b(bug|race|panic|fails on|diagnose|deadlock)\b/i.test(text),
+    summarize: /\b(summarize|tl;dr|in two sentences|eli5)\b/i.test(text),
+    translate: /\b(translate|traduce)\b/i.test(text),
+    creative: /\b(write a (poem|story|song)|haiku)\b/i.test(text),
+    safety: /\b(refund|legal|medical|hipaa|lawsuit|diagnosis)\b/i.test(text),
+    analysis: /\b(compare|trade-?off|versus|analyse|analyze|evaluate|should we)\b/i.test(text),
+    constraints: (text.match(/\b(must|include|constraints?|requirements?)\b/gi) ?? []).length,
+    words,
+    lines,
+    question_marks: (text.match(/\?/g) ?? []).length
+  };
+  const category = categoryOf(features);
+  const complexity = complexityOf(features, category);
+  const p_small_quality = pSmallQuality(features, category, complexity);
+  const quality_risk = riskOf(category, complexity, p_small_quality);
+  const recommended_tier = tierFromP(p_small_quality);
+  const signals = [
+    features.code_fence,
+    features.code_kw,
+    features.math,
+    features.design,
+    features.reason,
+    features.debug,
+    features.safety
+  ].filter(Boolean).length;
+  const confidence = category === "factual_recall" && words < 16 ? 0.9 : Math.min(0.95, 0.55 + 0.12 * signals);
+  return {
+    complexity,
+    category,
+    confidence: Number(confidence.toFixed(3)),
+    recommended_tier,
+    quality_risk,
+    p_small_quality,
+    uncertainty: Number((1 - p_small_quality).toFixed(3)),
+    structured_output: Boolean(features.code_fence || features.constraints >= 1),
+    context_tokens_est: Math.max(1, Math.round(text.length / 4)),
+    rationale: `${category.replaceAll("_", " ")} L${complexity}. P(quality|small)=${p_small_quality}. Route to ${recommended_tier}.`,
+    features
+  };
+}
+function categoryOf(h) {
+  if (h.safety) return "safety_sensitive";
+  if (h.design) return "system_design";
+  if (h.debug && (h.code_fence || h.code_kw)) return "code_debug";
+  if (h.code_fence || h.code_kw) return "code_generation";
+  if (h.math && h.reason) return "reasoning";
+  if (h.math) return "math";
+  if (h.reason) return "reasoning";
+  if (h.translate) return "translation";
+  if (h.summarize) return "summarization";
+  if (h.creative) return "creative";
+  if (h.analysis) return "analysis";
+  if (h.words < 24 && h.question_marks >= 1) return "factual_recall";
+  return h.words > 80 ? "analysis" : "factual_recall";
+}
+function complexityOf(h, category) {
+  let score = 1;
+  if (h.words > 40) score += 1;
+  if (h.words > 120) score += 1;
+  if (h.lines > 12 || h.code_fence) score += 1;
+  if (h.constraints >= 2) score += 1;
+  if (h.design || h.reason) score += 1;
+  if (h.debug) score += 1;
+  if (["system_design", "reasoning", "safety_sensitive"].includes(category)) score = Math.max(score + 1, 4);
+  if (category === "code_generation") score = Math.max(score, 3);
+  if (category === "code_debug") score = Math.max(score, 4);
+  if (category === "factual_recall" && h.words < 20) score = Math.min(score, 2);
+  return Math.max(1, Math.min(5, score));
+}
+function pSmallQuality(h, category, complexity) {
+  let p = 0.96;
+  if (h.design || category === "system_design") p -= 0.28;
+  if (h.safety || category === "safety_sensitive") p -= 0.3;
+  if (h.debug || category === "code_debug") p -= 0.22;
+  if (h.reason || category === "reasoning") p -= 0.18;
+  if (complexity >= 5) p -= 0.22;
+  else if (complexity >= 4) p -= 0.14;
+  else if (complexity === 3) p -= 0.06;
+  if (h.words > 120) p -= 0.07;
+  if (h.constraints >= 2) p -= 0.07;
+  if (category === "code_generation") p -= 0.05;
+  if (category === "factual_recall" && h.words < 24) p = Math.max(p, 0.94);
+  return Number(Math.min(0.99, Math.max(0.05, p)).toFixed(3));
+}
+function riskOf(category, complexity, p) {
+  if (HIGH_RISK.has(category) || complexity >= 5 || p < 0.72) return "high";
+  if (complexity >= 3 || p < 0.9) return "medium";
+  return "low";
+}
+function tierFromP(p) {
+  if (p >= 0.9) return "economy";
+  if (p >= 0.72) return "standard";
+  return "frontier";
+}
+function difficultyTier(complexity) {
+  if (complexity <= 2) return "economy";
+  if (complexity === 3) return "standard";
+  return "frontier";
+}
+
+// src/providers.ts
+var PROVIDERS = [
+  { id: "openai", label: "OpenAI", baseURL: "https://api.openai.com/v1", env: "OPENAI_API_KEY", hint: "sk-..." },
+  { id: "groq", label: "Groq", baseURL: "https://api.groq.com/openai/v1", env: "GROQ_API_KEY", hint: "gsk_..." },
+  { id: "baseten", label: "Baseten", baseURL: "https://inference.baseten.co/v1", env: "BASETEN_API_KEY", hint: "baseten key" },
+  { id: "openrouter", label: "OpenRouter", baseURL: "https://openrouter.ai/api/v1", env: "OPENROUTER_API_KEY", hint: "sk-or-..." },
+  { id: "together", label: "Together", baseURL: "https://api.together.xyz/v1", env: "TOGETHER_API_KEY", hint: "together key" },
+  { id: "fireworks", label: "Fireworks", baseURL: "https://api.fireworks.ai/inference/v1", env: "FIREWORKS_API_KEY", hint: "fw_..." },
+  { id: "deepseek", label: "DeepSeek", baseURL: "https://api.deepseek.com/v1", env: "DEEPSEEK_API_KEY", hint: "sk-..." },
+  { id: "mistral", label: "Mistral", baseURL: "https://api.mistral.ai/v1", env: "MISTRAL_API_KEY", hint: "mistral key" },
+  { id: "cerebras", label: "Cerebras", baseURL: "https://api.cerebras.ai/v1", env: "CEREBRAS_API_KEY", hint: "csk-..." },
+  { id: "xai", label: "xAI", baseURL: "https://api.x.ai/v1", env: "XAI_API_KEY", hint: "xai-..." },
+  { id: "google", label: "Google", baseURL: "https://generativelanguage.googleapis.com/v1beta/openai", env: "GOOGLE_API_KEY", hint: "AIza..." },
+  { id: "perplexity", label: "Perplexity", baseURL: "https://api.perplexity.ai", env: "PERPLEXITY_API_KEY", hint: "pplx-..." },
+  { id: "nvidia", label: "NVIDIA NIM", baseURL: "https://integrate.api.nvidia.com/v1", env: "NVIDIA_API_KEY", hint: "nvapi-..." },
+  { id: "sambanova", label: "SambaNova", baseURL: "https://api.sambanova.ai/v1", env: "SAMBANOVA_API_KEY", hint: "samba key" },
+  { id: "hyperbolic", label: "Hyperbolic", baseURL: "https://api.hyperbolic.xyz/v1", env: "HYPERBOLIC_API_KEY", hint: "hyperbolic key" },
+  { id: "moonshot", label: "Moonshot", baseURL: "https://api.moonshot.ai/v1", env: "MOONSHOT_API_KEY", hint: "sk-..." },
+  { id: "ollama", label: "Ollama", baseURL: "http://localhost:11434/v1", env: "OLLAMA_API_KEY", hint: "optional" }
+];
+function findProvider(input) {
+  const needle = input.trim().toLowerCase();
+  return PROVIDERS.find((p) => p.id === needle || p.label.toLowerCase() === needle) ?? null;
+}
+function publicCatalog() {
+  return PROVIDERS.map(({ id, label, baseURL, env }) => ({
+    id,
+    label,
+    base_url: baseURL,
+    env
+  }));
+}
+function resolveBaseURL(input) {
+  if (input.baseURL?.trim()) return { baseURL: input.baseURL.trim().replace(/\/$/, ""), provider: findProvider(input.provider ?? "") };
+  if (input.provider) {
+    const provider = findProvider(input.provider);
+    if (provider) return { baseURL: provider.baseURL, provider };
+  }
+  return { baseURL: null, provider: null };
+}
+// Annotate the CommonJS export names for ESM import in node:
+0 && (module.exports = {
+  DEFAULT_GATEWAY,
+  PROVIDERS,
+  Promptimizer,
+  PromptimizerError,
+  classifyMessages,
+  classifyText,
+  difficultyTier,
+  findProvider,
+  publicCatalog,
+  resolveBaseURL
+});

package/dist/index.d.ts

@@ -0,0 +1,5 @@
+export { DEFAULT_GATEWAY, Promptimizer, PromptimizerError } from "./client";
+export { classifyMessages, classifyText, difficultyTier } from "./classify";
+export { PROVIDERS, findProvider, publicCatalog, resolveBaseURL } from "./providers";
+export type { ProviderPreset } from "./providers";
+export type { ChatCompletion, ChatCompletionRequest, ChatMessage, Classification, ConnectOptions, CostBreakdown, ModelInfo, PromptimizerOptions, SavingsSummary, Session, Tier, } from "./types.js";

package/dist/index.js

@@ -0,0 +1,279 @@
+// src/client.ts
+var PromptimizerError = class extends Error {
+  constructor(message, status, body) {
+    super(message);
+    this.status = status;
+    this.body = body;
+    this.name = "PromptimizerError";
+  }
+};
+var DEFAULT_GATEWAY = (process.env.PROMPTIMIZER_URL || "https://hackathon-omega-liart.vercel.app/api").replace(/\/$/, "");
+var Promptimizer = class _Promptimizer {
+  gatewayURL;
+  sessionId;
+  apiKey;
+  fetcher;
+  constructor(options = {}) {
+    this.gatewayURL = (options.gatewayURL ?? options.baseURL ?? DEFAULT_GATEWAY).replace(/\/$/, "");
+    this.sessionId = options.sessionId;
+    this.apiKey = options.apiKey;
+    this.fetcher = options.fetch ?? fetch;
+  }
+  static async connect(options) {
+    const client = new _Promptimizer({
+      gatewayURL: options.gatewayURL,
+      apiKey: options.accountKey
+    });
+    const session = await client.connect(options);
+    return { client, session };
+  }
+  async connect(options) {
+    const session = await this.request("/v1/providers/connect", {
+      method: "POST",
+      body: JSON.stringify({
+        mode: options.mode ?? "byok",
+        label: options.label,
+        provider: options.provider,
+        base_url: options.baseURL,
+        api_key: options.apiKey
+      })
+    });
+    this.sessionId = session.session_id;
+    return session;
+  }
+  async providers() {
+    return this.request(
+      "/v1/providers"
+    );
+  }
+  async savings() {
+    return this.request("/v1/savings");
+  }
+  async session() {
+    return this.request("/v1/session");
+  }
+  async models() {
+    return this.request(
+      "/v1/models"
+    );
+  }
+  async updateFleet(body) {
+    return this.request("/v1/models", { method: "PATCH", body: JSON.stringify(body) });
+  }
+  async classify(input) {
+    return this.request("/v1/classify", { method: "POST", body: JSON.stringify(input) });
+  }
+  async benchmark(compareAlwaysFrontier = true) {
+    return this.request("/v1/benchmark/run", {
+      method: "POST",
+      body: JSON.stringify({ compare_always_frontier: compareAlwaysFrontier })
+    });
+  }
+  chat = {
+    completions: {
+      create: (request) => this.createChatCompletion(request)
+    }
+  };
+  async createChatCompletion(request) {
+    return this.request("/v1/chat/completions", {
+      method: "POST",
+      body: JSON.stringify(request)
+    });
+  }
+  async request(path, init = {}) {
+    const headers = new Headers(init.headers);
+    headers.set("Content-Type", "application/json");
+    if (this.apiKey) {
+      headers.set("Authorization", `Bearer ${this.apiKey}`);
+    } else if (this.sessionId) {
+      headers.set("X-Promptimizer-Session", this.sessionId);
+      headers.set("Authorization", `Bearer ${this.sessionId}`);
+    }
+    const response = await this.fetcher(`${this.gatewayURL}${path}`, { ...init, headers });
+    const data = await response.json().catch(() => ({}));
+    if (!response.ok) {
+      const detail = typeof data === "object" && data && "detail" in data ? String(data.detail) : response.statusText;
+      throw new PromptimizerError(detail, response.status, data);
+    }
+    return data;
+  }
+};
+
+// src/classify.ts
+var HIGH_RISK = /* @__PURE__ */ new Set(["system_design", "safety_sensitive", "code_debug", "reasoning"]);
+function textFrom(messages) {
+  return messages.map((m) => {
+    if (typeof m.content === "string") return m.content;
+    return m.content.map((b) => b.text ?? "").join("\n");
+  }).join("\n").trim();
+}
+function classifyMessages(messages) {
+  return classifyText(textFrom(messages));
+}
+function classifyText(text) {
+  const words = text.split(/\s+/).filter(Boolean).length;
+  const lines = text.split("\n").length;
+  const features = {
+    code_fence: /```/.test(text),
+    code_kw: /\b(def |class |function |import |fn |pub |async |SELECT |goroutine|mutex|traceback)\b/i.test(
+      text
+    ),
+    math: /(\$\$|\\frac|prove that|expected value|O\([nN]\)|\d+\s*[\*\^]\s*\d+)/i.test(text),
+    design: /\b(design|architect|rate limiter|distributed|consistency|shard|1 million QPS)\b/i.test(
+      text
+    ),
+    reason: /\b(prove|why does|walk through|step by step|derive|contradiction|p-value)\b/i.test(
+      text
+    ),
+    debug: /\b(bug|race|panic|fails on|diagnose|deadlock)\b/i.test(text),
+    summarize: /\b(summarize|tl;dr|in two sentences|eli5)\b/i.test(text),
+    translate: /\b(translate|traduce)\b/i.test(text),
+    creative: /\b(write a (poem|story|song)|haiku)\b/i.test(text),
+    safety: /\b(refund|legal|medical|hipaa|lawsuit|diagnosis)\b/i.test(text),
+    analysis: /\b(compare|trade-?off|versus|analyse|analyze|evaluate|should we)\b/i.test(text),
+    constraints: (text.match(/\b(must|include|constraints?|requirements?)\b/gi) ?? []).length,
+    words,
+    lines,
+    question_marks: (text.match(/\?/g) ?? []).length
+  };
+  const category = categoryOf(features);
+  const complexity = complexityOf(features, category);
+  const p_small_quality = pSmallQuality(features, category, complexity);
+  const quality_risk = riskOf(category, complexity, p_small_quality);
+  const recommended_tier = tierFromP(p_small_quality);
+  const signals = [
+    features.code_fence,
+    features.code_kw,
+    features.math,
+    features.design,
+    features.reason,
+    features.debug,
+    features.safety
+  ].filter(Boolean).length;
+  const confidence = category === "factual_recall" && words < 16 ? 0.9 : Math.min(0.95, 0.55 + 0.12 * signals);
+  return {
+    complexity,
+    category,
+    confidence: Number(confidence.toFixed(3)),
+    recommended_tier,
+    quality_risk,
+    p_small_quality,
+    uncertainty: Number((1 - p_small_quality).toFixed(3)),
+    structured_output: Boolean(features.code_fence || features.constraints >= 1),
+    context_tokens_est: Math.max(1, Math.round(text.length / 4)),
+    rationale: `${category.replaceAll("_", " ")} L${complexity}. P(quality|small)=${p_small_quality}. Route to ${recommended_tier}.`,
+    features
+  };
+}
+function categoryOf(h) {
+  if (h.safety) return "safety_sensitive";
+  if (h.design) return "system_design";
+  if (h.debug && (h.code_fence || h.code_kw)) return "code_debug";
+  if (h.code_fence || h.code_kw) return "code_generation";
+  if (h.math && h.reason) return "reasoning";
+  if (h.math) return "math";
+  if (h.reason) return "reasoning";
+  if (h.translate) return "translation";
+  if (h.summarize) return "summarization";
+  if (h.creative) return "creative";
+  if (h.analysis) return "analysis";
+  if (h.words < 24 && h.question_marks >= 1) return "factual_recall";
+  return h.words > 80 ? "analysis" : "factual_recall";
+}
+function complexityOf(h, category) {
+  let score = 1;
+  if (h.words > 40) score += 1;
+  if (h.words > 120) score += 1;
+  if (h.lines > 12 || h.code_fence) score += 1;
+  if (h.constraints >= 2) score += 1;
+  if (h.design || h.reason) score += 1;
+  if (h.debug) score += 1;
+  if (["system_design", "reasoning", "safety_sensitive"].includes(category)) score = Math.max(score + 1, 4);
+  if (category === "code_generation") score = Math.max(score, 3);
+  if (category === "code_debug") score = Math.max(score, 4);
+  if (category === "factual_recall" && h.words < 20) score = Math.min(score, 2);
+  return Math.max(1, Math.min(5, score));
+}
+function pSmallQuality(h, category, complexity) {
+  let p = 0.96;
+  if (h.design || category === "system_design") p -= 0.28;
+  if (h.safety || category === "safety_sensitive") p -= 0.3;
+  if (h.debug || category === "code_debug") p -= 0.22;
+  if (h.reason || category === "reasoning") p -= 0.18;
+  if (complexity >= 5) p -= 0.22;
+  else if (complexity >= 4) p -= 0.14;
+  else if (complexity === 3) p -= 0.06;
+  if (h.words > 120) p -= 0.07;
+  if (h.constraints >= 2) p -= 0.07;
+  if (category === "code_generation") p -= 0.05;
+  if (category === "factual_recall" && h.words < 24) p = Math.max(p, 0.94);
+  return Number(Math.min(0.99, Math.max(0.05, p)).toFixed(3));
+}
+function riskOf(category, complexity, p) {
+  if (HIGH_RISK.has(category) || complexity >= 5 || p < 0.72) return "high";
+  if (complexity >= 3 || p < 0.9) return "medium";
+  return "low";
+}
+function tierFromP(p) {
+  if (p >= 0.9) return "economy";
+  if (p >= 0.72) return "standard";
+  return "frontier";
+}
+function difficultyTier(complexity) {
+  if (complexity <= 2) return "economy";
+  if (complexity === 3) return "standard";
+  return "frontier";
+}
+
+// src/providers.ts
+var PROVIDERS = [
+  { id: "openai", label: "OpenAI", baseURL: "https://api.openai.com/v1", env: "OPENAI_API_KEY", hint: "sk-..." },
+  { id: "groq", label: "Groq", baseURL: "https://api.groq.com/openai/v1", env: "GROQ_API_KEY", hint: "gsk_..." },
+  { id: "baseten", label: "Baseten", baseURL: "https://inference.baseten.co/v1", env: "BASETEN_API_KEY", hint: "baseten key" },
+  { id: "openrouter", label: "OpenRouter", baseURL: "https://openrouter.ai/api/v1", env: "OPENROUTER_API_KEY", hint: "sk-or-..." },
+  { id: "together", label: "Together", baseURL: "https://api.together.xyz/v1", env: "TOGETHER_API_KEY", hint: "together key" },
+  { id: "fireworks", label: "Fireworks", baseURL: "https://api.fireworks.ai/inference/v1", env: "FIREWORKS_API_KEY", hint: "fw_..." },
+  { id: "deepseek", label: "DeepSeek", baseURL: "https://api.deepseek.com/v1", env: "DEEPSEEK_API_KEY", hint: "sk-..." },
+  { id: "mistral", label: "Mistral", baseURL: "https://api.mistral.ai/v1", env: "MISTRAL_API_KEY", hint: "mistral key" },
+  { id: "cerebras", label: "Cerebras", baseURL: "https://api.cerebras.ai/v1", env: "CEREBRAS_API_KEY", hint: "csk-..." },
+  { id: "xai", label: "xAI", baseURL: "https://api.x.ai/v1", env: "XAI_API_KEY", hint: "xai-..." },
+  { id: "google", label: "Google", baseURL: "https://generativelanguage.googleapis.com/v1beta/openai", env: "GOOGLE_API_KEY", hint: "AIza..." },
+  { id: "perplexity", label: "Perplexity", baseURL: "https://api.perplexity.ai", env: "PERPLEXITY_API_KEY", hint: "pplx-..." },
+  { id: "nvidia", label: "NVIDIA NIM", baseURL: "https://integrate.api.nvidia.com/v1", env: "NVIDIA_API_KEY", hint: "nvapi-..." },
+  { id: "sambanova", label: "SambaNova", baseURL: "https://api.sambanova.ai/v1", env: "SAMBANOVA_API_KEY", hint: "samba key" },
+  { id: "hyperbolic", label: "Hyperbolic", baseURL: "https://api.hyperbolic.xyz/v1", env: "HYPERBOLIC_API_KEY", hint: "hyperbolic key" },
+  { id: "moonshot", label: "Moonshot", baseURL: "https://api.moonshot.ai/v1", env: "MOONSHOT_API_KEY", hint: "sk-..." },
+  { id: "ollama", label: "Ollama", baseURL: "http://localhost:11434/v1", env: "OLLAMA_API_KEY", hint: "optional" }
+];
+function findProvider(input) {
+  const needle = input.trim().toLowerCase();
+  return PROVIDERS.find((p) => p.id === needle || p.label.toLowerCase() === needle) ?? null;
+}
+function publicCatalog() {
+  return PROVIDERS.map(({ id, label, baseURL, env }) => ({
+    id,
+    label,
+    base_url: baseURL,
+    env
+  }));
+}
+function resolveBaseURL(input) {
+  if (input.baseURL?.trim()) return { baseURL: input.baseURL.trim().replace(/\/$/, ""), provider: findProvider(input.provider ?? "") };
+  if (input.provider) {
+    const provider = findProvider(input.provider);
+    if (provider) return { baseURL: provider.baseURL, provider };
+  }
+  return { baseURL: null, provider: null };
+}
+export {
+  DEFAULT_GATEWAY,
+  PROVIDERS,
+  Promptimizer,
+  PromptimizerError,
+  classifyMessages,
+  classifyText,
+  difficultyTier,
+  findProvider,
+  publicCatalog,
+  resolveBaseURL
+};

package/dist/providers.cjs

@@ -0,0 +1,74 @@
+"use strict";
+var __defProp = Object.defineProperty;
+var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
+var __getOwnPropNames = Object.getOwnPropertyNames;
+var __hasOwnProp = Object.prototype.hasOwnProperty;
+var __export = (target, all) => {
+  for (var name in all)
+    __defProp(target, name, { get: all[name], enumerable: true });
+};
+var __copyProps = (to, from, except, desc) => {
+  if (from && typeof from === "object" || typeof from === "function") {
+    for (let key of __getOwnPropNames(from))
+      if (!__hasOwnProp.call(to, key) && key !== except)
+        __defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
+  }
+  return to;
+};
+var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
+
+// src/providers.ts
+var providers_exports = {};
+__export(providers_exports, {
+  PROVIDERS: () => PROVIDERS,
+  findProvider: () => findProvider,
+  publicCatalog: () => publicCatalog,
+  resolveBaseURL: () => resolveBaseURL
+});
+module.exports = __toCommonJS(providers_exports);
+var PROVIDERS = [
+  { id: "openai", label: "OpenAI", baseURL: "https://api.openai.com/v1", env: "OPENAI_API_KEY", hint: "sk-..." },
+  { id: "groq", label: "Groq", baseURL: "https://api.groq.com/openai/v1", env: "GROQ_API_KEY", hint: "gsk_..." },
+  { id: "baseten", label: "Baseten", baseURL: "https://inference.baseten.co/v1", env: "BASETEN_API_KEY", hint: "baseten key" },
+  { id: "openrouter", label: "OpenRouter", baseURL: "https://openrouter.ai/api/v1", env: "OPENROUTER_API_KEY", hint: "sk-or-..." },
+  { id: "together", label: "Together", baseURL: "https://api.together.xyz/v1", env: "TOGETHER_API_KEY", hint: "together key" },
+  { id: "fireworks", label: "Fireworks", baseURL: "https://api.fireworks.ai/inference/v1", env: "FIREWORKS_API_KEY", hint: "fw_..." },
+  { id: "deepseek", label: "DeepSeek", baseURL: "https://api.deepseek.com/v1", env: "DEEPSEEK_API_KEY", hint: "sk-..." },
+  { id: "mistral", label: "Mistral", baseURL: "https://api.mistral.ai/v1", env: "MISTRAL_API_KEY", hint: "mistral key" },
+  { id: "cerebras", label: "Cerebras", baseURL: "https://api.cerebras.ai/v1", env: "CEREBRAS_API_KEY", hint: "csk-..." },
+  { id: "xai", label: "xAI", baseURL: "https://api.x.ai/v1", env: "XAI_API_KEY", hint: "xai-..." },
+  { id: "google", label: "Google", baseURL: "https://generativelanguage.googleapis.com/v1beta/openai", env: "GOOGLE_API_KEY", hint: "AIza..." },
+  { id: "perplexity", label: "Perplexity", baseURL: "https://api.perplexity.ai", env: "PERPLEXITY_API_KEY", hint: "pplx-..." },
+  { id: "nvidia", label: "NVIDIA NIM", baseURL: "https://integrate.api.nvidia.com/v1", env: "NVIDIA_API_KEY", hint: "nvapi-..." },
+  { id: "sambanova", label: "SambaNova", baseURL: "https://api.sambanova.ai/v1", env: "SAMBANOVA_API_KEY", hint: "samba key" },
+  { id: "hyperbolic", label: "Hyperbolic", baseURL: "https://api.hyperbolic.xyz/v1", env: "HYPERBOLIC_API_KEY", hint: "hyperbolic key" },
+  { id: "moonshot", label: "Moonshot", baseURL: "https://api.moonshot.ai/v1", env: "MOONSHOT_API_KEY", hint: "sk-..." },
+  { id: "ollama", label: "Ollama", baseURL: "http://localhost:11434/v1", env: "OLLAMA_API_KEY", hint: "optional" }
+];
+function findProvider(input) {
+  const needle = input.trim().toLowerCase();
+  return PROVIDERS.find((p) => p.id === needle || p.label.toLowerCase() === needle) ?? null;
+}
+function publicCatalog() {
+  return PROVIDERS.map(({ id, label, baseURL, env }) => ({
+    id,
+    label,
+    base_url: baseURL,
+    env
+  }));
+}
+function resolveBaseURL(input) {
+  if (input.baseURL?.trim()) return { baseURL: input.baseURL.trim().replace(/\/$/, ""), provider: findProvider(input.provider ?? "") };
+  if (input.provider) {
+    const provider = findProvider(input.provider);
+    if (provider) return { baseURL: provider.baseURL, provider };
+  }
+  return { baseURL: null, provider: null };
+}
+// Annotate the CommonJS export names for ESM import in node:
+0 && (module.exports = {
+  PROVIDERS,
+  findProvider,
+  publicCatalog,
+  resolveBaseURL
+});

package/dist/providers.d.ts

@@ -0,0 +1,25 @@
+export type ProviderPreset = {
+    id: string;
+    label: string;
+    baseURL: string;
+    env: string;
+    hint: string;
+};
+export declare const PROVIDERS: ProviderPreset[];
+export declare function findProvider(input: string): ProviderPreset | null;
+export declare function publicCatalog(): {
+    id: string;
+    label: string;
+    base_url: string;
+    env: string;
+}[];
+export declare function resolveBaseURL(input: {
+    provider?: string;
+    baseURL?: string;
+}): {
+    baseURL: string;
+    provider: ProviderPreset | null;
+} | {
+    baseURL: null;
+    provider: null;
+};

package/dist/providers.js

@@ -0,0 +1,46 @@
+// src/providers.ts
+var PROVIDERS = [
+  { id: "openai", label: "OpenAI", baseURL: "https://api.openai.com/v1", env: "OPENAI_API_KEY", hint: "sk-..." },
+  { id: "groq", label: "Groq", baseURL: "https://api.groq.com/openai/v1", env: "GROQ_API_KEY", hint: "gsk_..." },
+  { id: "baseten", label: "Baseten", baseURL: "https://inference.baseten.co/v1", env: "BASETEN_API_KEY", hint: "baseten key" },
+  { id: "openrouter", label: "OpenRouter", baseURL: "https://openrouter.ai/api/v1", env: "OPENROUTER_API_KEY", hint: "sk-or-..." },
+  { id: "together", label: "Together", baseURL: "https://api.together.xyz/v1", env: "TOGETHER_API_KEY", hint: "together key" },
+  { id: "fireworks", label: "Fireworks", baseURL: "https://api.fireworks.ai/inference/v1", env: "FIREWORKS_API_KEY", hint: "fw_..." },
+  { id: "deepseek", label: "DeepSeek", baseURL: "https://api.deepseek.com/v1", env: "DEEPSEEK_API_KEY", hint: "sk-..." },
+  { id: "mistral", label: "Mistral", baseURL: "https://api.mistral.ai/v1", env: "MISTRAL_API_KEY", hint: "mistral key" },
+  { id: "cerebras", label: "Cerebras", baseURL: "https://api.cerebras.ai/v1", env: "CEREBRAS_API_KEY", hint: "csk-..." },
+  { id: "xai", label: "xAI", baseURL: "https://api.x.ai/v1", env: "XAI_API_KEY", hint: "xai-..." },
+  { id: "google", label: "Google", baseURL: "https://generativelanguage.googleapis.com/v1beta/openai", env: "GOOGLE_API_KEY", hint: "AIza..." },
+  { id: "perplexity", label: "Perplexity", baseURL: "https://api.perplexity.ai", env: "PERPLEXITY_API_KEY", hint: "pplx-..." },
+  { id: "nvidia", label: "NVIDIA NIM", baseURL: "https://integrate.api.nvidia.com/v1", env: "NVIDIA_API_KEY", hint: "nvapi-..." },
+  { id: "sambanova", label: "SambaNova", baseURL: "https://api.sambanova.ai/v1", env: "SAMBANOVA_API_KEY", hint: "samba key" },
+  { id: "hyperbolic", label: "Hyperbolic", baseURL: "https://api.hyperbolic.xyz/v1", env: "HYPERBOLIC_API_KEY", hint: "hyperbolic key" },
+  { id: "moonshot", label: "Moonshot", baseURL: "https://api.moonshot.ai/v1", env: "MOONSHOT_API_KEY", hint: "sk-..." },
+  { id: "ollama", label: "Ollama", baseURL: "http://localhost:11434/v1", env: "OLLAMA_API_KEY", hint: "optional" }
+];
+function findProvider(input) {
+  const needle = input.trim().toLowerCase();
+  return PROVIDERS.find((p) => p.id === needle || p.label.toLowerCase() === needle) ?? null;
+}
+function publicCatalog() {
+  return PROVIDERS.map(({ id, label, baseURL, env }) => ({
+    id,
+    label,
+    base_url: baseURL,
+    env
+  }));
+}
+function resolveBaseURL(input) {
+  if (input.baseURL?.trim()) return { baseURL: input.baseURL.trim().replace(/\/$/, ""), provider: findProvider(input.provider ?? "") };
+  if (input.provider) {
+    const provider = findProvider(input.provider);
+    if (provider) return { baseURL: provider.baseURL, provider };
+  }
+  return { baseURL: null, provider: null };
+}
+export {
+  PROVIDERS,
+  findProvider,
+  publicCatalog,
+  resolveBaseURL
+};

package/dist/types.d.ts

@@ -0,0 +1,123 @@
+export type Tier = "economy" | "standard" | "frontier";
+export type ChatMessage = {
+    role: "system" | "user" | "assistant" | "tool" | string;
+    content: string | Array<{
+        type: string;
+        text?: string;
+    }>;
+    name?: string;
+};
+export type Classification = {
+    complexity: number;
+    category: string;
+    confidence: number;
+    recommended_tier: Tier;
+    quality_risk: "low" | "medium" | "high";
+    p_small_quality: number;
+    uncertainty: number;
+    structured_output: boolean;
+    context_tokens_est: number;
+    rationale: string;
+    features: Record<string, unknown>;
+};
+export type ModelInfo = {
+    id: string;
+    owned_by?: string;
+    input_per_1m?: number | null;
+    output_per_1m?: number | null;
+    tier: Tier;
+    source?: string;
+    selected?: boolean;
+};
+export type ConnectOptions = {
+    mode?: "mock" | "byok";
+    label?: string;
+    provider?: string;
+    baseURL?: string;
+    apiKey?: string;
+    accountKey?: string;
+    gatewayURL?: string;
+};
+export type SavingsSummary = {
+    requests: number;
+    actual_usd: number;
+    baseline_usd: number;
+    saved_usd: number;
+    saved_pct: number;
+    routing_saved_usd: number;
+    cache_saved_usd: number;
+    cache_hits: number;
+    escalations: number;
+    avg_quality: number | null;
+    recent: Array<{
+        id: string;
+        model: string;
+        tier: string;
+        actual_usd: number;
+        baseline_usd: number;
+        saved_usd: number;
+        routing_saved_usd: number;
+        cache_saved_usd: number;
+        cache_hit: boolean;
+        escalated: boolean;
+        quality: number | null;
+        created_at: string;
+    }>;
+};
+export type PromptimizerOptions = {
+    gatewayURL?: string;
+    sessionId?: string;
+    apiKey?: string;
+    baseURL?: string;
+    fetch?: typeof fetch;
+};
+export type ChatCompletionRequest = {
+    messages: ChatMessage[];
+    model?: string;
+    stream?: boolean;
+    level_override?: number;
+    temperature?: number;
+    max_tokens?: number;
+};
+export type CostBreakdown = {
+    actual_usd: number;
+    baseline_usd: number;
+    saved_usd: number;
+    saved_pct: number;
+    routing_saved_usd: number;
+    cache_discount_usd: number;
+    prompt_tokens: number;
+    completion_tokens: number;
+    cached_tokens: number;
+};
+export type ChatCompletion = {
+    id: string;
+    object: string;
+    created: number;
+    model: string;
+    choices: Array<{
+        index: number;
+        message: {
+            role: string;
+            content: string;
+        };
+        finish_reason: string;
+    }>;
+    usage: {
+        prompt_tokens: number;
+        completion_tokens: number;
+        total_tokens: number;
+        cost?: CostBreakdown;
+    };
+    promptimizer?: Record<string, unknown>;
+};
+export type Session = {
+    session_id: string;
+    mode: string;
+    label: string;
+    base_url: string;
+    models: ModelInfo[];
+    baseline_model: string | null;
+    stats: Record<string, number>;
+    created_at: number;
+};

package/package.json

@@ -0,0 +1,74 @@
+{
+  "name": "promptimizer",
+  "version": "0.1.0",
+  "description": "OpenAI-compatible SDK for Promptimizer — BYOK model routing, classification, and prompt caching.",
+  "type": "module",
+  "main": "./src/index.ts",
+  "module": "./src/index.ts",
+  "types": "./src/index.ts",
+  "exports": {
+    ".": {
+      "types": "./src/index.ts",
+      "import": "./src/index.ts",
+      "default": "./src/index.ts"
+    },
+    "./providers": {
+      "types": "./src/providers.ts",
+      "import": "./src/providers.ts",
+      "default": "./src/providers.ts"
+    }
+  },
+  "files": [
+    "dist",
+    "README.md"
+  ],
+  "scripts": {
+    "build": "node scripts/build.mjs",
+    "dev": "node scripts/build.mjs",
+    "prepublishOnly": "npm run build",
+    "typecheck": "tsc --noEmit",
+    "test": "node --experimental-strip-types --test src/classify.test.ts src/providers.test.ts",
+    "publish:npm": "npm publish --access public"
+  },
+  "publishConfig": {
+    "access": "public",
+    "main": "./dist/index.cjs",
+    "module": "./dist/index.js",
+    "types": "./dist/index.d.ts",
+    "exports": {
+      ".": {
+        "types": "./dist/index.d.ts",
+        "import": "./dist/index.js",
+        "require": "./dist/index.cjs"
+      },
+      "./providers": {
+        "types": "./dist/providers.d.ts",
+        "import": "./dist/providers.js",
+        "require": "./dist/providers.cjs"
+      }
+    }
+  },
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/sreecharan-desu/promptimizer.git",
+    "directory": "packages/sdk"
+  },
+  "homepage": "https://hackathon-omega-liart.vercel.app",
+  "bugs": {
+    "url": "https://github.com/sreecharan-desu/promptimizer/issues"
+  },
+  "keywords": [
+    "llm",
+    "openai",
+    "router",
+    "prompt-cache",
+    "byok",
+    "cost-optimization"
+  ],
+  "license": "MIT",
+  "devDependencies": {
+    "@types/node": "^22.15.0",
+    "esbuild": "^0.25.9",
+    "typescript": "^5.8.0"
+  }
+}