@fudrouter/fsrouter 0.6.63 → 0.6.64

package/open-sse/utils/thinkTagParser.js

@@ -1,146 +0,0 @@
-/**
- * Think Tag Parser
- *
- * Parses <think>...</think> tags from LLM output and converts them
- * to a structured `reasoning_content` field.
- *
- * Used by providers like DeepSeek, Qwen, and Qoder that embed
- * chain-of-thought reasoning inside <think> tags.
- *
- * Usage:
- *   import { extractThinkTags, hasThinkTags } from "./thinkTagParser.js";
- *
- *   const { reasoning, content } = extractThinkTags(rawOutput);
- *   // reasoning = "step-by-step thinking..."
- *   // content = "final answer..."
- */
-
-const THINK_OPEN = "<think>";
-const THINK_CLOSE = "</think>";
-
-/**
- * Check if text contains think tags
- * @param {string} text - Raw output text
- * @returns {boolean}
- */
-export function hasThinkTags(text) {
-  if (!text) return false;
-  return text.includes(THINK_OPEN);
-}
-
-/**
- * Extract think tags from text
- * Returns the reasoning content (inside <think>) and the cleaned final content
- *
- * @param {string} text - Raw output text
- * @returns {{ reasoning: string|null, content: string }}
- */
-export function extractThinkTags(text) {
-  if (!text || !text.includes(THINK_OPEN)) {
-    return { reasoning: null, content: text || "" };
-  }
-
-  let reasoning = "";
-  let content = text;
-  let iterations = 0;
-  const maxIterations = 10; // safety limit
-
-  while (content.includes(THINK_OPEN) && iterations < maxIterations) {
-    const openIdx = content.indexOf(THINK_OPEN);
-    const closeIdx = content.indexOf(THINK_CLOSE, openIdx);
-
-    if (closeIdx === -1) {
-      // Unclosed think tag — treat everything after <think> as reasoning
-      reasoning += content.slice(openIdx + THINK_OPEN.length);
-      content = content.slice(0, openIdx);
-      break;
-    }
-
-    // Extract the think content
-    const thinkContent = content.slice(openIdx + THINK_OPEN.length, closeIdx);
-    reasoning += (reasoning ? "\n" : "") + thinkContent;
-
-    // Remove the think block from content
-    content = content.slice(0, openIdx) + content.slice(closeIdx + THINK_CLOSE.length);
-    iterations++;
-  }
-
-  return {
-    reasoning: reasoning.trim() || null,
-    content: content.trim(),
-  };
-}
-
-/**
- * Process a streaming delta chunk and extract think content.
- * Maintains state across chunks using a context object.
- *
- * @param {string} delta - New text chunk
- * @param {object} ctx - Mutable context object { insideThink, buffer }
- * @returns {{ reasoningDelta: string|null, contentDelta: string|null }}
- */
-export function processStreamingThinkDelta(delta, ctx) {
-  if (!ctx.buffer) ctx.buffer = "";
-
-  ctx.buffer += delta;
-  let reasoningDelta = "";
-  let contentDelta = "";
-
-  while (ctx.buffer.length > 0) {
-    if (ctx.insideThink) {
-      // Looking for closing tag
-      const closeIdx = ctx.buffer.indexOf(THINK_CLOSE);
-      if (closeIdx === -1) {
-        // Might be a partial tag at the end — keep last few chars
-        if (ctx.buffer.length > THINK_CLOSE.length) {
-          const safe = ctx.buffer.slice(0, -(THINK_CLOSE.length - 1));
-          reasoningDelta += safe;
-          ctx.buffer = ctx.buffer.slice(-(THINK_CLOSE.length - 1));
-        }
-        break;
-      }
-      reasoningDelta += ctx.buffer.slice(0, closeIdx);
-      ctx.buffer = ctx.buffer.slice(closeIdx + THINK_CLOSE.length);
-      ctx.insideThink = false;
-    } else {
-      // Looking for opening tag
-      const openIdx = ctx.buffer.indexOf(THINK_OPEN);
-      if (openIdx === -1) {
-        // Might be a partial tag at the end
-        if (ctx.buffer.length > THINK_OPEN.length) {
-          const safe = ctx.buffer.slice(0, -(THINK_OPEN.length - 1));
-          contentDelta += safe;
-          ctx.buffer = ctx.buffer.slice(-(THINK_OPEN.length - 1));
-        }
-        break;
-      }
-      contentDelta += ctx.buffer.slice(0, openIdx);
-      ctx.buffer = ctx.buffer.slice(openIdx + THINK_OPEN.length);
-      ctx.insideThink = true;
-    }
-  }
-
-  return {
-    reasoningDelta: reasoningDelta || null,
-    contentDelta: contentDelta || null,
-  };
-}
-
-/**
- * Flush remaining buffer content from streaming context.
- * Call this when the stream ends.
- *
- * @param {object} ctx - Mutable context object { insideThink, buffer }
- * @returns {{ reasoningDelta: string|null, contentDelta: string|null }}
- */
-export function flushThinkBuffer(ctx) {
-  if (!ctx.buffer) return { reasoningDelta: null, contentDelta: null };
-
-  const remaining = ctx.buffer;
-  ctx.buffer = "";
-
-  if (ctx.insideThink) {
-    return { reasoningDelta: remaining || null, contentDelta: null };
-  }
-  return { reasoningDelta: null, contentDelta: remaining || null };
-}

package/open-sse/utils/toolCallArguments.js

@@ -1,12 +0,0 @@
-function appendToolCallArgumentDelta(current, incoming) {
-  const existing = typeof current === "string" ? current : "";
-  const next = typeof incoming === "string" ? incoming : "";
-  if (!existing) return next;
-  if (!next) return existing;
-  if (next === existing) return existing;
-  if (next.startsWith(existing)) return next;
-  return existing + next;
-}
-export {
-  appendToolCallArgumentDelta
-};

package/open-sse/utils/toolDeduper.js

@@ -1,49 +0,0 @@
-/**
- * Strip built-in/duplicate tools when equivalent MCP tools are present.
- * Goal: reduce tool definitions token bloat for Claude clients.
- */
-
-const DEDUP_RULES = [
-  {
-    // Exa MCP present → drop built-in web tools (Exa is preferred).
-    triggers: ["mcp__exa__web_search_exa", "mcp__exa__web_fetch_exa"],
-    strip: ["WebSearch", "WebFetch", "mcp__workspace__web_fetch"],
-  },
-  {
-    // Tavily MCP present → drop built-in web tools.
-    triggers: ["mcp__tavily__tavily_search", "mcp__tavily__tavily_extract"],
-    strip: ["WebSearch", "WebFetch", "mcp__workspace__web_fetch"],
-  },
-  {
-    // Browser MCP present → drop Cowork's duplicate Claude_in_Chrome connector.
-    triggers: [/^mcp__browsermcp__/],
-    strip: [/^mcp__Claude_in_Chrome__/],
-  },
-];
-
-function getToolName(t) {
-  return t?.name || t?.function?.name || "";
-}
-
-function matches(name, pattern) {
-  if (typeof pattern === "string") return name === pattern;
-  return pattern instanceof RegExp ? pattern.test(name) : false;
-}
-
-function dedupeTools(tools) {
-  if (!Array.isArray(tools) || tools.length === 0) return { tools, stripped: [] };
-  const names = tools.map(getToolName);
-  const toStrip = new Set();
-  for (const rule of DEDUP_RULES) {
-    const hasTrigger = names.some((n) => rule.triggers.some((p) => matches(n, p)));
-    if (!hasTrigger) continue;
-    for (const n of names) {
-      if (rule.strip.some((p) => matches(n, p))) toStrip.add(n);
-    }
-  }
-  if (toStrip.size === 0) return { tools, stripped: [] };
-  const out = tools.filter((t) => !toStrip.has(getToolName(t)));
-  return { tools: out, stripped: Array.from(toStrip) };
-}
-
-export { dedupeTools };

package/open-sse/utils/toolSources.js

@@ -1,31 +0,0 @@
-const MAX_VISIBLE_NAMES = 80;
-function getToolName(tool) {
-  return tool?.name || tool?.function?.name || tool?.type || "unknown";
-}
-function getToolSource(name) {
-  if (name.startsWith("mcp__")) {
-    const parts = name.split("__");
-    return parts[1] ? `mcp:${parts[1]}` : "mcp";
-  }
-  if (name.startsWith("web_search") || name.startsWith("web_fetch")) return "hosted:web";
-  if (name.startsWith("computer_") || name.startsWith("str_replace_")) return "hosted:computer";
-  return "client";
-}
-function summarizeToolSources(tools) {
-  if (!Array.isArray(tools) || tools.length === 0) return null;
-  const names = tools.map((tool) => getToolName(tool));
-  const sourceCounts = /* @__PURE__ */ new Map();
-  for (const name of names) {
-    const source = getToolSource(name);
-    sourceCounts.set(source, (sourceCounts.get(source) || 0) + 1);
-  }
-  const sources = Array.from(sourceCounts.entries()).map(([source, count]) => `${source}=${count}`).join(", ");
-  const visibleNames = names.slice(0, MAX_VISIBLE_NAMES).join(", ");
-  const suffix = names.length > MAX_VISIBLE_NAMES ? `, ... +${names.length - MAX_VISIBLE_NAMES} more` : "";
-  return `${tools.length} tools | sources: ${sources} | names: ${visibleNames}${suffix}`;
-}
-export {
-  getToolName,
-  getToolSource,
-  summarizeToolSources
-};

package/open-sse/utils/upstreamResponseHeaders.js

@@ -1,25 +0,0 @@
-const STRIP_HEADER_NAMES = /* @__PURE__ */ new Set([
-  "content-encoding",
-  "content-length",
-  "transfer-encoding"
-]);
-function stripStaleEncodingHeaders(input) {
-  const out = new Headers(input);
-  for (const name of STRIP_HEADER_NAMES) out.delete(name);
-  return out;
-}
-function filterUpstreamResponseHeaderEntries(entries, extraToStrip = []) {
-  const drop = new Set(STRIP_HEADER_NAMES);
-  for (const h of extraToStrip) drop.add(h.toLowerCase());
-  const result = [];
-  for (const [k, v] of entries) {
-    if (!drop.has(k.toLowerCase())) result.push([k, v]);
-  }
-  return result;
-}
-const STRIP_UPSTREAM_HEADER_NAMES = STRIP_HEADER_NAMES;
-export {
-  STRIP_UPSTREAM_HEADER_NAMES,
-  filterUpstreamResponseHeaderEntries,
-  stripStaleEncodingHeaders
-};

package/open-sse/utils/urlSanitize.js

@@ -1,15 +0,0 @@
-function stripTrailingSlashes(value) {
-  let end = value.length;
-  while (end > 0 && value.charCodeAt(end - 1) === 47) {
-    end--;
-  }
-  return end === value.length ? value : value.slice(0, end);
-}
-function normalizeBaseUrl(value) {
-  const str = typeof value === "string" ? value : "";
-  return stripTrailingSlashes(str.trim());
-}
-export {
-  normalizeBaseUrl,
-  stripTrailingSlashes
-};

package/open-sse/utils/usageTracking.js

@@ -1,347 +0,0 @@
-/**
- * Token Usage Tracking - Extract, normalize, estimate and log token usage
- */
-
-import { saveRequestUsage, appendRequestLog } from '../../../dist/lib/usageDb.js';
-import { FORMATS } from "../translator/formats.js";
-
-// ANSI color codes
-export const COLORS = {
-  reset: "\x1b[0m",
-  red: "\x1b[31m",
-  green: "\x1b[32m",
-  yellow: "\x1b[33m",
-  blue: "\x1b[34m",
-  cyan: "\x1b[36m"
-};
-
-// Buffer tokens to prevent context errors
-const BUFFER_TOKENS = 2000;
-
-// Get HH:MM:SS timestamp
-function getTimeString() {
-  return new Date().toLocaleTimeString("en-US", { hour12: false, hour: "2-digit", minute: "2-digit", second: "2-digit" });
-}
-
-/**
- * Add buffer tokens to usage to prevent context errors
- * @param {object} usage - Usage object (any format)
- * @returns {object} Usage with buffer added
- */
-export function addBufferToUsage(usage) {
-  if (!usage || typeof usage !== "object") return usage;
-
-  const result = { ...usage };
-
-  // Claude format
-  if (result.input_tokens !== undefined) {
-    result.input_tokens += BUFFER_TOKENS;
-  }
-
-  // OpenAI format
-  if (result.prompt_tokens !== undefined) {
-    result.prompt_tokens += BUFFER_TOKENS;
-  }
-
-  // Calculate or update total_tokens
-  if (result.total_tokens !== undefined) {
-    result.total_tokens += BUFFER_TOKENS;
-  } else if (result.prompt_tokens !== undefined && result.completion_tokens !== undefined) {
-    // Calculate total_tokens if not exists
-    result.total_tokens = result.prompt_tokens + result.completion_tokens;
-  }
-
-  return result;
-}
-
-export function filterUsageForFormat(usage, targetFormat) {
-  if (!usage || typeof usage !== "object") return usage;
-
-  // Helper to pick only defined fields from usage
-  const pickFields = (fields) => {
-    const filtered = {};
-    for (const field of fields) {
-      if (usage[field] !== undefined) {
-        filtered[field] = usage[field];
-      }
-    }
-    return filtered;
-  };
-
-  // Define allowed fields for each format
-  const formatFields = {
-    [FORMATS.CLAUDE]: [
-      'input_tokens', 'output_tokens', 
-      'cache_read_input_tokens', 'cache_creation_input_tokens',
-      'estimated'
-    ],
-    [FORMATS.GEMINI]: [
-      'promptTokenCount', 'candidatesTokenCount', 'totalTokenCount',
-      'cachedContentTokenCount', 'thoughtsTokenCount',
-      'estimated'
-    ],
-    [FORMATS.OPENAI_RESPONSES]: [
-      'input_tokens', 'output_tokens',
-      'input_tokens_details', 'output_tokens_details',
-      'estimated'
-    ],
-    // OpenAI format (default for OPENAI, CODEX, KIRO, etc.)
-    default: [
-      'prompt_tokens', 'completion_tokens', 'total_tokens',
-      'cached_tokens', 'reasoning_tokens',
-      'prompt_tokens_details', 'completion_tokens_details',
-      'estimated'
-    ]
-  };
-
-  // Get fields for target format
-  let fields = formatFields[targetFormat];
-  
-  // Use same fields for similar formats
-  if (targetFormat === FORMATS.GEMINI_CLI || targetFormat === FORMATS.ANTIGRAVITY) {
-    fields = formatFields[FORMATS.GEMINI];
-  } else if (targetFormat === FORMATS.OPENAI_RESPONSE) {
-    fields = formatFields[FORMATS.OPENAI_RESPONSES];
-  } else if (!fields) {
-    fields = formatFields.default;
-  }
-
-  return pickFields(fields);
-}
-
-/**
- * Normalize usage object - ensure all values are valid numbers
- */
-export function normalizeUsage(usage) {
-  if (!usage || typeof usage !== "object" || Array.isArray(usage)) return null;
-
-  const normalized = {};
-  const assignNumber = (key, value) => {
-    if (value === undefined || value === null) return;
-    const numeric = Number(value);
-    if (Number.isFinite(numeric)) normalized[key] = numeric;
-  };
-
-  assignNumber("prompt_tokens", usage?.prompt_tokens);
-  assignNumber("completion_tokens", usage?.completion_tokens);
-  assignNumber("total_tokens", usage?.total_tokens);
-  assignNumber("cache_read_input_tokens", usage?.cache_read_input_tokens);
-  assignNumber("cache_creation_input_tokens", usage?.cache_creation_input_tokens);
-  assignNumber("cached_tokens", usage?.cached_tokens);
-  assignNumber("reasoning_tokens", usage?.reasoning_tokens);
-
-  // Preserve nested details objects for OpenAI format forwarding
-  if (usage?.prompt_tokens_details && typeof usage.prompt_tokens_details === "object") {
-    normalized.prompt_tokens_details = usage.prompt_tokens_details;
-  }
-  if (usage?.completion_tokens_details && typeof usage.completion_tokens_details === "object") {
-    normalized.completion_tokens_details = usage.completion_tokens_details;
-  }
-
-  if (Object.keys(normalized).length === 0) return null;
-  return normalized;
-}
-
-/**
- * Check if usage has valid token data
- * Valid = has at least one token field with value > 0
- * Invalid = empty object {}, null, undefined, no token fields, or all zeros
- */
-export function hasValidUsage(usage) {
-  if (!usage || typeof usage !== "object") return false;
-
-  // Check for any known token field with value > 0
-  const tokenFields = [
-    "prompt_tokens", "completion_tokens", "total_tokens",  // OpenAI
-    "input_tokens", "output_tokens",                        // Claude
-    "promptTokenCount", "candidatesTokenCount"              // Gemini
-  ];
-
-  for (const field of tokenFields) {
-    if (typeof usage[field] === "number" && usage[field] > 0) {
-      return true;
-    }
-  }
-
-  return false;
-}
-
-/**
- * Extract usage from any format (Claude, OpenAI, Gemini, Responses API)
- */
-export function extractUsage(chunk) {
-  if (!chunk || typeof chunk !== "object") return null;
-
-  // Claude format (message_delta event)
-  if (chunk.type === "message_delta" && chunk.usage && typeof chunk.usage === "object") {
-    return normalizeUsage({
-      prompt_tokens: chunk.usage.input_tokens || 0,
-      completion_tokens: chunk.usage.output_tokens || 0,
-      cache_read_input_tokens: chunk.usage.cache_read_input_tokens,
-      cache_creation_input_tokens: chunk.usage.cache_creation_input_tokens
-    });
-  }
-
-  // OpenAI Responses API format (response.completed or response.done)
-  if ((chunk.type === "response.completed" || chunk.type === "response.done") && chunk.response?.usage && typeof chunk.response.usage === "object") {
-    const usage = chunk.response.usage;
-    const cachedTokens = usage.input_tokens_details?.cached_tokens;
-    return normalizeUsage({
-      prompt_tokens: usage.input_tokens || usage.prompt_tokens || 0,
-      completion_tokens: usage.output_tokens || usage.completion_tokens || 0,
-      cached_tokens: cachedTokens,
-      reasoning_tokens: usage.output_tokens_details?.reasoning_tokens,
-      prompt_tokens_details: cachedTokens ? { cached_tokens: cachedTokens } : undefined
-    });
-  }
-
-  // OpenAI format (also covers DeepSeek which uses prompt_cache_hit_tokens)
-  if (chunk.usage && typeof chunk.usage === "object" && chunk.usage.prompt_tokens !== undefined) {
-    return normalizeUsage({
-      prompt_tokens: chunk.usage.prompt_tokens,
-      completion_tokens: chunk.usage.completion_tokens || 0,
-      cached_tokens: chunk.usage.prompt_tokens_details?.cached_tokens || chunk.usage.prompt_cache_hit_tokens,
-      reasoning_tokens: chunk.usage.completion_tokens_details?.reasoning_tokens,
-      prompt_tokens_details: chunk.usage.prompt_tokens_details,
-      completion_tokens_details: chunk.usage.completion_tokens_details
-    });
-  }
-
-  // Gemini format (Antigravity)
-  // Antigravity wraps usageMetadata inside response: { response: { usageMetadata: {...} } }
-  const usageMeta = chunk.usageMetadata || chunk.response?.usageMetadata;
-  if (usageMeta && typeof usageMeta === "object") {
-    return normalizeUsage({
-      prompt_tokens: usageMeta.promptTokenCount || 0,
-      completion_tokens: usageMeta.candidatesTokenCount || 0,
-      total_tokens: usageMeta.totalTokenCount,
-      cached_tokens: usageMeta.cachedContentTokenCount,
-      reasoning_tokens: usageMeta.thoughtsTokenCount
-    });
-  }
-
-  // Ollama NDJSON format (raw from provider, before translation)
-  // Ollama sends: {"model":"...","done":true,"prompt_eval_count":N,"eval_count":M}
-  if (chunk.done === true && typeof chunk.prompt_eval_count === "number") {
-    return normalizeUsage({
-      prompt_tokens: chunk.prompt_eval_count || 0,
-      completion_tokens: chunk.eval_count || 0,
-      total_tokens: (chunk.prompt_eval_count || 0) + (chunk.eval_count || 0)
-    });
-  }
-
-  return null;
-}
-
-/**
- * Estimate input tokens from request body
- * Calculate total body size for more accurate estimation
- */
-export function estimateInputTokens(body) {
-  if (!body || typeof body !== "object") return 0;
-
-  try {
-    // Calculate total body size (includes messages, tools, system, thinking config, etc.)
-    const bodyStr = JSON.stringify(body);
-    const totalChars = bodyStr.length;
-
-    // Estimate: ~4 chars per token (rough average across all tokenizers)
-    return Math.ceil(totalChars / 4);
-  } catch (err) {
-    // Fallback if stringify fails
-    return 0;
-  }
-}
-
-/**
- * Estimate output tokens from content length
- */
-export function estimateOutputTokens(contentLength) {
-  if (!contentLength || contentLength <= 0) return 0;
-  return Math.max(1, Math.floor(contentLength / 4));
-}
-
-/**
- * Format usage object based on target format
- * @param {number} inputTokens - Input/prompt tokens
- * @param {number} outputTokens - Output/completion tokens
- * @param {string} targetFormat - Target format from FORMATS
- */
-export function formatUsage(inputTokens, outputTokens, targetFormat) {
-  // Claude format uses input_tokens/output_tokens
-  if (targetFormat === FORMATS.CLAUDE) {
-    return addBufferToUsage({ 
-      input_tokens: inputTokens, 
-      output_tokens: outputTokens, 
-      estimated: true 
-    });
-  }
-
-  // Default: OpenAI format (works for openai, gemini, responses, etc.)
-  return addBufferToUsage({
-    prompt_tokens: inputTokens,
-    completion_tokens: outputTokens,
-    total_tokens: inputTokens + outputTokens,
-    estimated: true
-  });
-}
-
-/**
- * Estimate full usage when provider doesn't return it
- * @param {object} body - Request body for input token estimation
- * @param {number} contentLength - Content length for output token estimation
- * @param {string} targetFormat - Target format from FORMATS constant
- */
-export function estimateUsage(body, contentLength, targetFormat = FORMATS.OPENAI) {
-  return formatUsage(
-    estimateInputTokens(body),
-    estimateOutputTokens(contentLength),
-    targetFormat
-  );
-}
-
-/**
- * Log usage with cache info (green color)
- */
-export function logUsage(provider, usage, model = null, connectionId = null, apiKey = null) {
-  if (!usage || typeof usage !== "object") return;
-
-  const p = provider?.toUpperCase() || "UNKNOWN";
-
-  // Support both formats:
-  // - OpenAI: prompt_tokens, completion_tokens
-  // - Claude: input_tokens, output_tokens
-  const inTokens = usage?.prompt_tokens || usage?.input_tokens || 0;
-  const outTokens = usage?.completion_tokens || usage?.output_tokens || 0;
-  const accountPrefix = connectionId ? connectionId.slice(0, 8) + "..." : "unknown";
-
-  let msg = `[${getTimeString()}] 📊 ${COLORS.green}[USAGE] ${p} | in=${inTokens} | out=${outTokens} | account=${accountPrefix}${COLORS.reset}`;
-
-  // Add estimated flag if present
-  if (usage.estimated) {
-    msg += ` ${COLORS.yellow}(estimated)${COLORS.reset}`;
-  }
-
-  // Add cache info if present (unified from different formats)
-  const cacheRead = usage.cache_read_input_tokens || usage.cached_tokens || usage.prompt_tokens_details?.cached_tokens;
-  if (cacheRead) msg += ` | cache_read=${cacheRead}`;
-
-  const cacheCreation = usage.cache_creation_input_tokens;
-  if (cacheCreation) msg += ` | cache_create=${cacheCreation}`;
-
-  const reasoning = usage.reasoning_tokens;
-  if (reasoning) msg += ` | reasoning=${reasoning}`;
-
-  console.log(msg);
-
-  // Save to usage DB
-  const tokens = {
-    prompt_tokens: inTokens,
-    completion_tokens: outTokens,
-    cache_read_input_tokens: cacheRead || 0,
-    cache_creation_input_tokens: cacheCreation || 0,
-    reasoning_tokens: reasoning || 0
-  };
-  saveRequestUsage({ model, provider, connectionId, tokens, apiKey: apiKey || undefined }).catch(() => { });
-  appendRequestLog({ model, provider, connectionId, tokens, status: "200 OK" }).catch(() => { });
-}