@tyvm/knowhow 0.0.159 → 0.0.160

package/src/clients/attribution.ts

@@ -1,53 +0,0 @@
-import packageJson from "../../package.json";
-
-/**
- * Default Knowhow attribution headers.
- *
- * - `HTTP-Referer`       – OpenRouter analytics / billing attribution.
- * - `X-Title`            – Generic app-name header; supported by OpenRouter
- *                          and other providers that follow the same convention.
- * - `X-OpenRouter-Title` – Legacy OpenRouter-specific alias (kept for
- *                          back-compat; `X-Title` is preferred going forward).
- * - `User-Agent`         – Standard SDK identification.
- */
-export const KNOWHOW_ATTRIBUTION_HEADERS: Readonly<Record<string, string>> = Object.freeze({
-  "HTTP-Referer": "https://knowhow.tyvm.ai",
-  "X-Title": "Knowhow",
-  "X-OpenRouter-Title": "Knowhow",
-  "User-Agent": `knowhow/${packageJson.version}`,
-});
-
-/**
- * Adds Knowhow's application attribution to an outbound OpenRouter request.
- * Later header sets win case-insensitively, so callers can override defaults
- * without producing duplicate keys.
- */
-export function withKnowhowAttribution(
-  /**
-   * Common overrides:
-   *   - `"X-Title"` / `"X-OpenRouter-Title"` – set your own app name
-   *   - `"HTTP-Referer"` – set your own referer URL
-   *   - any additional custom headers are merged in as-is
-   */
-  ...headerSets: Array<Record<string, string> | undefined>
-): Record<string, string> {
-  const result: Record<string, string> = { ...KNOWHOW_ATTRIBUTION_HEADERS };
-
-  for (const headers of headerSets) {
-    if (!headers) continue;
-    for (const [key, value] of Object.entries(headers)) {
-      const existingKey = Object.keys(result).find(
-        (candidate) => candidate.toLowerCase() === key.toLowerCase()
-      );
-      if (existingKey) {
-        // Keep the original casing so SDK-owned defaults such as User-Agent
-        // are replaced rather than duplicated in their intermediate objects.
-        result[existingKey] = value;
-      } else {
-        result[key] = value;
-      }
-    }
-  }
-
-  return result;
-}

package/src/clients/cerebras.ts

@@ -1,10 +0,0 @@
-import { HttpClient } from "./http";
-import { CerebrasTextPricing } from "./pricing/cerebras";
-
-export class GenericCerebrasClient extends HttpClient {
-  constructor(apiKey: string) {
-    super("https://api.cerebras.ai");
-    this.setJwt(apiKey);
-    this.setPrices(CerebrasTextPricing);
-  }
-}

package/src/clients/contextLimits.ts

@@ -1,263 +0,0 @@
-import { Models, EmbeddingModels } from "../types";
-import { FireworksModels } from "./pricing/fireworks";
-
-/**
- * Context window limits (in tokens) for all supported models.
- * Sources:
- * - OpenAI: https://platform.openai.com/docs/models
- * - Anthropic: https://docs.anthropic.com/en/docs/about-claude/models
- * - Google: https://ai.google.dev/gemini-api/docs/models
- * - xAI: https://docs.x.ai/developers/models
- */
-export const ContextLimits: Record<string, number> = {
-  // ─── OpenAI ───────────────────────────────────────────────────────────────
-  [Models.openai.GPT_56_Sol]: 1_000_000,
-  [Models.openai.GPT_56_Terra]: 1_000_000,
-  [Models.openai.GPT_56_Luna]: 1_000_000,
-  [Models.openai.GPT_55]: 1_000_000,
-  [Models.openai.GPT_54]: 1_000_000,
-  [Models.openai.GPT_54_Mini]: 400_000,
-  [Models.openai.GPT_54_Nano]: 400_000,
-  [Models.openai.GPT_54_Pro]: 1_000_000,
-  [Models.openai.GPT_53_Chat]: 1_000_000,
-  [Models.openai.GPT_53_Codex]: 1_000_000,
-  [Models.openai.GPT_5]: 1_000_000,
-  [Models.openai.GPT_5_Mini]: 1_000_000,
-  [Models.openai.GPT_5_Nano]: 1_000_000,
-  [Models.openai.GPT_51]: 1_000_000,
-  [Models.openai.GPT_52]: 1_000_000,
-  [Models.openai.GPT_41]: 1_047_576,
-  [Models.openai.GPT_41_Mini]: 1_047_576,
-  [Models.openai.GPT_41_Nano]: 1_047_576,
-  [Models.openai.GPT_45]: 128_000,
-  [Models.openai.GPT_4o]: 128_000,
-  [Models.openai.GPT_4o_Mini]: 128_000,
-  [Models.openai.GPT_4o_Audio]: 128_000,
-  [Models.openai.GPT_4o_Realtime]: 128_000,
-  [Models.openai.GPT_4o_Mini_Audio]: 128_000,
-  [Models.openai.GPT_4o_Mini_Realtime]: 128_000,
-  [Models.openai.GPT_4o_Mini_Search]: 128_000,
-  [Models.openai.GPT_4o_Search]: 128_000,
-  [Models.openai.o1]: 200_000,
-  [Models.openai.o1_Mini]: 128_000,
-  [Models.openai.o1_Pro]: 200_000,
-  [Models.openai.o3]: 200_000,
-  [Models.openai.o3_Pro]: 200_000,
-  [Models.openai.o3_Mini]: 200_000,
-  [Models.openai.o4_Mini]: 200_000,
-  // ─── OpenAI (aliases / deprecated / non-text) ─────────────────────────────
-  [Models.openai.GPT_55_Pro]: 1_000_000,
-  [Models.openai.GPT_53_Codex_Spark]: 1_000_000,
-  [Models.openai.GPT_52_Chat]: 1_000_000,
-  [Models.openai.GPT_52_Codex]: 1_000_000,
-  [Models.openai.GPT_52_Pro]: 1_000_000,
-  [Models.openai.GPT_51_Chat]: 1_000_000,
-  [Models.openai.GPT_51_Codex]: 1_000_000,
-  [Models.openai.GPT_51_Codex_Max]: 1_000_000,
-  [Models.openai.GPT_51_Codex_Mini]: 1_000_000,
-  [Models.openai.GPT_5_Pro]: 1_000_000,
-  [Models.openai.GPT_5_Chat]: 1_000_000,
-  [Models.openai.GPT_5_Codex]: 1_000_000,
-  [Models.openai.o1_Preview]: 128_000,
-  [Models.openai.o3_Deep_Research]: 200_000,
-  [Models.openai.o4_Mini_Deep_Research]: 200_000,
-  [Models.openai.GPT_4o_Transcribe]: 128_000,
-  [Models.openai.GPT_4o_Mini_Transcribe]: 128_000,
-  [Models.openai.GPT_Realtime_15]: 128_000,
-  [Models.openai.GPT_Realtime_Mini]: 128_000,
-  [Models.openai.GPT_4o_2024_05_13]: 128_000,
-  [Models.openai.GPT_4o_2024_11_20]: 128_000,
-  [Models.openai.GPT_35_Turbo]: 16_385,
-  [Models.openai.GPT_4]: 8_192,
-  [Models.openai.GPT_4_Turbo]: 128_000,
-  // OpenAI image/video/audio models — no text context window
-  [Models.openai.GPT_Image_2]: 0,
-  [Models.openai.GPT_Image_15]: 0,
-  [Models.openai.GPT_Image_1]: 0,
-  [Models.openai.GPT_Image_1_Mini]: 0,
-  [Models.openai.ChatGPT_Image]: 0,
-  [Models.openai.TTS_1]: 0,
-  [Models.openai.Whisper_1]: 0,
-  [Models.openai.DALL_E_3]: 0,
-  [Models.openai.DALL_E_2]: 0,
-  [Models.openai.Sora]: 0,
-  [Models.openai.Sora_2]: 0,
-  [Models.openai.Sora_2_Pro]: 0,
-
-  // ─── Anthropic ────────────────────────────────────────────────────────────
-  [Models.anthropic.Opus4_8Fast]: 1_000_000,
-  [Models.anthropic.Opus4_8]: 1_000_000,
-  // Claude Opus 5 (launched 2026-06-09, 1M context window)
-  [Models.anthropic.Opus5]: 1_000_000,
-  [Models.anthropic.Opus5Fast]: 1_000_000,
-  // Claude 5 generation (launched 2026-06-09, 1M context window)
-  [Models.anthropic.Fable5]: 1_000_000,
-  [Models.anthropic.Sonnet5]: 1_000_000,
-  [Models.anthropic.Mythos5]: 1_000_000,
-  [Models.anthropic.MythosPreview]: 1_000_000,
-  [Models.anthropic.Opus4_7]: 1_000_000,
-  [Models.anthropic.Opus4_6]: 1_000_000,
-  [Models.anthropic.Opus4_6Fast]: 1_000_000,
-  [Models.anthropic.Sonnet4_6]: 1_000_000,
-  [Models.anthropic.Opus4_5]: 1_000_000,
-  [Models.anthropic.Opus4]: 200_000,
-  [Models.anthropic.Opus4_1]: 200_000,
-  [Models.anthropic.Sonnet4]: 200_000,
-  [Models.anthropic.Sonnet4_5]: 200_000,
-  [Models.anthropic.Haiku4_5]: 200_000,
-  // Short aliases (resolve to their latest dated version)
-  [Models.anthropic.Sonnet4_5Short]: 200_000,
-  [Models.anthropic.Haiku4_5Short]: 200_000,
-  [Models.anthropic.Opus4_5Short]: 1_000_000,
-  [Models.anthropic.Sonnet3_7]: 200_000,
-  [Models.anthropic.Sonnet3_5]: 200_000,
-  [Models.anthropic.Opus3]: 200_000,
-  [Models.anthropic.Haiku3]: 200_000,
-  [Models.anthropic.Haiku3_5]: 200_000,
-
-  // ─── Anthropic (aliases / deprecated) ────────────────────────────────────
-  [Models.anthropic.Sonnet3_5_20240620]: 200_000,
-  [Models.anthropic.Haiku3_5_Latest]: 200_000,
-  [Models.anthropic.Sonnet3]: 200_000,
-  [Models.anthropic.Opus4_0]: 200_000,
-  [Models.anthropic.Sonnet4_0]: 200_000,
-
-  // ─── Google ───────────────────────────────────────────────────────────────
-  // Gemini 3.7 / 3.6 / 3.5 (text)
-  [Models.google.Gemini_37_Flash]: 1_048_576,
-  [Models.google.Gemini_36_Flash]: 1_000_000,
-  [Models.google.Gemini_35_Flash]: 1_000_000,
-  [Models.google.Gemini_35_Flash_Lite]: 1_000_000,
-  [Models.google.Gemini_35_Live_Translate_Preview]: 1_000_000,
-  // Gemini 3.x (text / multimodal reasoning)
-  [Models.google.Gemini_31_Flash_Lite]: 1_000_000,
-  [Models.google.Gemini_31_Flash_TTS_Preview]: 1_000_000,
-  [Models.google.Gemini_3_Pro_Preview]: 1_000_000,
-  [Models.google.Gemini_Omni_Flash_Preview]: 1_000_000,
-  [Models.google.Gemini_25_Computer_Use_Preview]: 1_000_000,
-  [Models.google.Gemini_20_Flash_Lite]: 1_000_000,
-  [Models.google.Gemini_Robotics_ER_16_Preview]: 1_000_000,
-  [Models.google.Gemini_Robotics_ER_15_Preview]: 1_000_000,
-  [Models.google.Gemini_31_Pro_Preview]: 1_000_000,
-  [Models.google.Gemini_31_Flash_Image_Preview]: 1_000_000,
-  [Models.google.Gemini_31_Flash_Lite_Preview]: 1_000_000,
-  [Models.google.Gemini_31_Flash_Live_Preview]: 1_000_000,
-  [Models.google.Gemini_3_Flash_Preview]: 1_000_000,
-  [Models.google.Gemini_3_Pro_Image_Preview]: 1_000_000,
-  [Models.google.Gemini_25_Pro]: 1_000_000,
-  [Models.google.Gemini_25_Flash]: 1_000_000,
-  [Models.google.Gemini_25_Flash_Lite]: 1_000_000,
-  [Models.google.Gemini_25_Flash_Preview]: 1_000_000,
-  [Models.google.Gemini_25_Pro_Preview]: 1_000_000,
-  [Models.google.Gemini_25_Flash_Image]: 1_000_000,
-  [Models.google.Gemini_25_Flash_Live]: 1_000_000,
-  [Models.google.Gemini_25_Flash_Preview_0417]: 1_000_000,
-  [Models.google.Gemini_25_Flash_Image_Preview]: 1_000_000,
-  [Models.google.Gemini_25_Flash_Native_Audio]: 1_000_000,
-  [Models.google.Gemini_25_Flash_TTS]: 1_000_000,
-  [Models.google.Gemini_25_Pro_TTS]: 1_000_000,
-  // Google image/video generation models — no text context window; use 0
-  [Models.google.Imagen_3]: 0,
-  [Models.google.Imagen_4_Fast]: 0,
-  [Models.google.Imagen_4_Ultra]: 0,
-  [Models.google.Veo_2]: 0,
-  [Models.google.Veo_3]: 0,
-  [Models.google.Veo_3_Fast]: 0,
-  [Models.google.Veo_3_1]: 0,
-  [Models.google.Veo_3_1_Fast]: 0,
-  [Models.google.Veo_3_1_Lite]: 0,
-  // Google image generation models — no text context window; use 0
-  [Models.google.Gemini_31_Flash_Image]: 0,
-  [Models.google.Gemini_31_Flash_Lite_Image]: 0,
-  [Models.google.Gemini_3_Pro_Image]: 0,
-  [Models.google.Gemini_20_Flash]: 1_000_000,
-  [Models.google.Gemini_20_Flash_Preview_Image_Generation]: 1_000_000,
-  [Models.google.Gemini_20_Flash_Live]: 1_000_000,
-  [Models.google.Gemini_20_Flash_TTS]: 1_000_000,
-  [Models.google.Gemini_15_Flash]: 1_000_000,
-  [Models.google.Gemini_15_Flash_8B]: 1_000_000,
-  [Models.google.Gemini_15_Pro]: 2_000_000,
-
-  // ─── xAI ──────────────────────────────────────────────────────────────────
-  [Models.xai.Grok4_6]: 500_000,
-  [Models.xai.Grok4_1_Fast_Reasoning]: 2_000_000,
-  [Models.xai.Grok4_1_Fast_NonReasoning]: 2_000_000,
-  [Models.xai.Grok_4_20_Reasoning]: 131_072,
-  [Models.xai.Grok_4_20_NonReasoning]: 131_072,
-  [Models.xai.Grok_4_20_MultiAgent]: 2_000_000,
-  [Models.xai.GrokCodeFast]: 2_000_000,
-  [Models.xai.Grok4]: 131_072,
-  [Models.xai.Grok3Beta]: 131_072,
-  [Models.xai.Grok3MiniBeta]: 131_072,
-  [Models.xai.Grok3FastBeta]: 131_072,
-  [Models.xai.Grok3MiniFastBeta]: 131_072,
-  [Models.xai.Grok21212]: 131_072,
-  [Models.xai.Grok2Vision1212]: 131_072,
-  // ─── xAI (aliases / deprecated / image / video) ───────────────────────────
-  [Models.xai.Grok2Latest]: 131_072,
-  [Models.xai.Grok2VisionLatest]: 131_072,
-  [Models.xai.Grok3Latest]: 131_072,
-  [Models.xai.Grok3FastLatest]: 131_072,
-  [Models.xai.Grok3MiniLatest]: 131_072,
-  [Models.xai.Grok3MiniFastLatest]: 131_072,
-  [Models.xai.GrokBeta]: 131_072,
-  [Models.xai.GrokVisionBeta]: 131_072,
-  [Models.xai.Grok4_1_Fast]: 2_000_000,
-  [Models.xai.Grok4_5]: 500_000,
-  [Models.xai.Grok4Fast]: 2_000_000,
-  [Models.xai.Grok4FastNonReasoning]: 2_000_000,
-  // xAI image/video models — no text context window
-  [Models.xai.GrokImagineImage]: 0,
-  [Models.xai.GrokImagineVideo]: 0,
-  [Models.xai.Grok2Image1212]: 0,
-
-  // ─── Fireworks AI ──────────────────────────────────────────────────────────
-  // Moonshot AI (Kimi) — context windows per fireworks.ai/pricing
-  [FireworksModels.KimiK3]:       131_072,
-  [FireworksModels.KimiK2p7Code]: 131_072,
-  [FireworksModels.KimiK2_6]:     131_072,
-  [FireworksModels.KimiK2_5]:     131_072,
-
-  // MiniMax
-  [FireworksModels.MinimaxM3]:    1_000_000,
-  [FireworksModels.MinimaxM2_7]:  1_000_000,
-  [FireworksModels.MinimaxM2_5]:  1_000_000,
-
-  // Qwen
-  [FireworksModels.Qwen3p7Plus]:  131_072,
-  [FireworksModels.Qwen3_6Plus]:  131_072,
-  [FireworksModels.Qwen3p6Plus]:  131_072,
-  [FireworksModels.Qwen3Vl30bA3bThinking]: 32_768,
-  [FireworksModels.Qwen3Vl30bA3bInstruct]: 32_768,
-  [FireworksModels.Qwen3_8b]:     131_072,
-
-  // Z.ai (GLM)
-  [FireworksModels.Glm5p2]: 131_072,
-  [FireworksModels.Glm5p1]: 131_072,
-  [FireworksModels.Glm5]:   131_072,
-  [FireworksModels.Glm4_7]: 131_072,
-
-  // DeepSeek AI
-  [FireworksModels.DeepseekV4Pro]:   163_840,
-  [FireworksModels.DeepseekV4Flash]: 163_840,
-  [FireworksModels.DeepseekV3_2]:    163_840,
-  [FireworksModels.DeepseekV3_1]:    163_840,
-
-  // OpenAI OSS
-  [FireworksModels.GptOss120b]: 131_072,
-
-  // Fireworks embedding/reranker models — 40k context
-  [FireworksModels.Qwen3Embedding8b]: 40_960,
-  [FireworksModels.Qwen3Reranker8b]:  40_960,
-};
-
-/** Default fallback context window limit (tokens) used when a model is not found. */
-export const DEFAULT_CONTEXT_LIMIT = 30_000;
-
-/**
- * Returns the context window limit (in tokens) for a given model.
- * Falls back to DEFAULT_CONTEXT_LIMIT if the model is not recognized.
- */
-export function getModelContextLimit(model: string): number {
-  return ContextLimits[model] ?? DEFAULT_CONTEXT_LIMIT;
-}

package/src/clients/copilot.ts

@@ -1,23 +0,0 @@
-import { HttpClient } from "./http";
-import { CopilotTextPricing } from "./pricing/copilot";
-
-/**
- * GitHub Copilot client — OpenAI-compatible API
- * https://docs.github.com/en/copilot/reference/ai-models/supported-models
- *
- * GitHub Copilot exposes an OpenAI-compatible endpoint at https://api.githubcopilot.com
- * that allows subscribers to use premium models (Claude Opus, GPT-5.x, Gemini, Grok etc.)
- * via their Copilot subscription's premium request allowance — no per-token charges.
- *
- * Authentication: uses a GitHub token (same as GITHUB_TOKEN / a personal access token
- * or OAuth token with copilot scope).
- *
- * Set env var GITHUB_COPILOT_TOKEN (preferred) or GITHUB_TOKEN to enable.
- */
-export class GitHubCopilotClient extends HttpClient {
-  constructor(apiKey = process.env.GITHUB_COPILOT_TOKEN ?? process.env.GITHUB_TOKEN) {
-    super("https://api.githubcopilot.com");
-    if (apiKey) this.setJwt(apiKey);
-    this.setPrices(CopilotTextPricing);
-  }
-}

package/src/clients/deepseek.ts

@@ -1,16 +0,0 @@
-import { HttpClient } from "./http";
-import { DeepSeekTextPricing } from "./pricing/deepseek";
-
-/**
- * DeepSeek client — OpenAI-compatible API
- * https://platform.deepseek.com/api-docs/
- * Industry-leading reasoning (R1) and coding (V3) models at very low cost.
- * Set env var DEEPSEEK_API_KEY to enable.
- */
-export class GenericDeepSeekClient extends HttpClient {
-  constructor(apiKey = process.env.DEEPSEEK_API_KEY) {
-    super("https://api.deepseek.com");
-    if (apiKey) this.setJwt(apiKey);
-    this.setPrices(DeepSeekTextPricing);
-  }
-}

package/src/clients/fireworks.ts

@@ -1,71 +0,0 @@
-import { HttpClient } from "./http";
-import { FireworksTextPricing, FireworksEmbeddingPricing } from "./pricing/fireworks";
-import { CompletionOptions, CompletionResponse } from "./types";
-
-type ModelInfo = { id: string; object: string; owned_by: string };
-
-/**
- * Fireworks AI client — OpenAI-compatible API (fast serverless inference)
- * https://docs.fireworks.ai/api-reference/introduction
- * Set env var FIREWORKS_API_KEY to enable.
- */
-export class GenericFireworksClient extends HttpClient {
-  constructor(apiKey = process.env.FIREWORKS_API_KEY) {
-    super("https://api.fireworks.ai/inference");
-    if (apiKey) this.setJwt(apiKey);
-    this.setPrices(FireworksTextPricing);
-  }
-
-  /**
-   * Skip the live /v1/models API call entirely — Fireworks' model listing
-   * endpoint frequently returns 500s and triggers slow exponential backoff.
-   * Our pricing map is the authoritative source of truth for available models.
-   */
-  async getModels(_type = "all"): Promise<ModelInfo[]> {
-    // Return embedding/reranker models when requested
-    if (_type === "embedding") {
-      return Object.keys(FireworksEmbeddingPricing).map((id) => ({
-        id,
-        object: "model",
-        owned_by: "fireworks",
-      }));
-    }
-
-    return Object.keys(FireworksTextPricing).map((id) => ({
-      id,
-      object: "model",
-      owned_by: "fireworks",
-    }));
-  }
-
-  /**
-   * Sanitize the request before sending to Fireworks.
-   * Some models (e.g. kimi-k2p7-code) reject extra fields like:
-   *   - tools[N].function.returns  (non-standard extension)
-   *   - long_ttl_cache             (Anthropic-specific cache flag)
-   */
-  async createChatCompletion(
-    options: CompletionOptions
-  ): Promise<CompletionResponse> {
-    const sanitized: CompletionOptions = {
-      ...options,
-      // Strip Anthropic-specific field not accepted by Fireworks
-      long_ttl_cache: undefined,
-      // This is translated by the OpenAI/Anthropic clients, but is not part of
-      // Fireworks' OpenAI-compatible chat completion schema.
-      reasoning_summary: undefined,
-    };
-
-    if (sanitized.tools) {
-      sanitized.tools = sanitized.tools.map((tool) => {
-        const { returns, ...fnRest } = tool.function as any;
-        return {
-          ...tool,
-          function: fnRest,
-        };
-      });
-    }
-
-    return super.createChatCompletion(sanitized);
-  }
-}