@sanqianx/claude-ai-workbench 0.13.2 → 0.13.4

package/src/backend/lib/llm-client.js

@@ -1,176 +1,176 @@
-// _site/lib/llm-client.js
-//
-// Tiny Anthropic Messages API client. Zero npm deps; supports HTTPS and HTTP
-// Anthropic-compatible endpoints. Connection settings are resolved from an AI
-// profile first, then environment variables as fallback:
-//
-//   profile.baseUrl       or ANTHROPIC_BASE_URL
-//   profile.apiKey        or ANTHROPIC_AUTH_TOKEN
-//   profile.mainModel     or ANTHROPIC_MODEL
-//   profile.timeoutMs     or API_TIMEOUT_MS
-//
-// This lets the local UI persist model settings in ai-profiles.json while still
-// keeping env vars useful for scripts and temporary overrides.
-
-const fs = require('fs');
-const path = require('path');
-const http = require('http');
-const https = require('https');
-const { URL } = require('url');
-
-const DEFAULT_BASE = 'https://api.anthropic.com';
-const DEFAULT_VERSION = '2023-06-01';
-const DEFAULT_MODEL = 'claude-haiku-4-5';
-const { getDataDir } = require('./data-dir');
-
-function aiProfilesPath() {
-  return path.join(getDataDir(), 'ai-profiles.json');
-}
-
-function readProfilesFile() {
-  try {
-    const cfg = JSON.parse(fs.readFileSync(aiProfilesPath(), 'utf-8'));
-    return Array.isArray(cfg.profiles) ? cfg.profiles : [];
-  } catch {
-    return [];
-  }
-}
-
-function findProfile(profileId) {
-  if (!profileId) return null;
-  return readProfilesFile().find(profile => profile && profile.id === profileId) || null;
-}
-
-function readConfig(options = {}) {
-  const profileId = typeof options === 'string' ? options : options.profileId;
-  const inlineProfile = options && typeof options === 'object' && options.profile && typeof options.profile === 'object'
-    ? options.profile
-    : null;
-  const profile = inlineProfile || findProfile(profileId) || {};
-  const timeoutMs = Number(profile.timeoutMs || process.env.API_TIMEOUT_MS || 60_000);
-  return {
-    profileId: profile.id || profileId || null,
-    baseUrl: profile.baseUrl || profile.apiBaseUrl || profile.anthropicBaseUrl || process.env.ANTHROPIC_BASE_URL || DEFAULT_BASE,
-    apiKey: profile.apiKey || profile.authToken || profile.anthropicAuthToken || process.env.ANTHROPIC_AUTH_TOKEN || '',
-    model: profile.mainModel || profile.model || process.env.ANTHROPIC_MODEL || DEFAULT_MODEL,
-    version: DEFAULT_VERSION,
-    timeoutMs: Number.isFinite(timeoutMs) && timeoutMs > 0 ? timeoutMs : 60_000,
-  };
-}
-
-function buildRequestPath(base, p) {
-  if (!p.startsWith('/')) return p;
-  const basePath = base.pathname.endsWith('/') ? base.pathname.slice(0, -1) : base.pathname;
-  return basePath + p;
-}
-
-function postJson({ baseUrl, path: p, body, apiKey, version, timeoutMs = 60_000 }) {
-  return new Promise((resolve, reject) => {
-    try {
-      const base = new URL(baseUrl);
-      const transport = base.protocol === 'http:' ? http : https;
-      const data = JSON.stringify(body);
-      const req = transport.request({
-        host: base.hostname,
-        port: base.port || (base.protocol === 'http:' ? 80 : 443),
-        path: buildRequestPath(base, p),
-        method: 'POST',
-        headers: {
-          'Content-Type': 'application/json',
-          'x-api-key': apiKey,
-          'anthropic-version': version,
-          'Content-Length': Buffer.byteLength(data),
-        },
-        timeout: timeoutMs,
-      }, res => {
-        const chunks = [];
-        res.on('data', c => chunks.push(c));
-        res.on('end', () => {
-          const text = Buffer.concat(chunks).toString('utf-8');
-          if (res.statusCode < 200 || res.statusCode >= 300) {
-            return reject(new Error(`HTTP ${res.statusCode}: ${text.slice(0, 500)}`));
-          }
-          try { resolve(JSON.parse(text)); }
-          catch (e) { reject(new Error(`bad JSON: ${e.message}; body head: ${text.slice(0, 200)}`)); }
-        });
-      });
-      req.on('error', reject);
-      req.on('timeout', () => req.destroy(new Error(`timeout after ${timeoutMs}ms`)));
-      req.write(data);
-      req.end();
-    } catch (e) {
-      reject(new Error(`bad baseUrl: ${baseUrl}: ${e.message}`));
-    }
-  });
-}
-
-async function completeText({ system, user, maxTokens = 512, model, profileId, profile }) {
-  const cfg = readConfig({ profileId, profile });
-  if (!cfg.apiKey) {
-    const id = profileId ? ` for profile ${profileId}` : '';
-    throw new Error(`API key not set${id}`);
-  }
-  const body = {
-    model: model || cfg.model,
-    max_tokens: maxTokens,
-    messages: [{ role: 'user', content: user }],
-  };
-  if (system) body.system = system;
-
-  const raw = await postJson({
-    baseUrl: cfg.baseUrl,
-    path: '/v1/messages',
-    body,
-    apiKey: cfg.apiKey,
-    version: cfg.version,
-    timeoutMs: cfg.timeoutMs,
-  });
-
-  const text = (raw.content || [])
-    .filter(block => block && block.type === 'text' && typeof block.text === 'string')
-    .map(block => block.text)
-    .join('\n')
-    .trim();
-
-  return {
-    text,
-    raw,
-    usage: raw.usage || null,
-    model: raw.model || body.model,
-    profileId: cfg.profileId,
-  };
-}
-
-async function completeJson({ system, user, schema, maxTokens = 2048, model, profileId, profile }) {
-  const schemaHint = schema ? `\n\nThe output MUST be a JSON object matching this schema:\n${schema}\n` : '';
-  const finalUser = `${user}${schemaHint}\n\nRespond with ONLY the JSON object, no prose, no markdown fences.`;
-  const result = await completeText({ system, user: finalUser, maxTokens, model, profileId, profile });
-  const text = result.text;
-
-  let parsed = null;
-  let parseError = null;
-  const candidates = [
-    text,
-    text.replace(/^```(?:json)?\s*/i, '').replace(/```\s*$/, ''),
-  ];
-  for (const candidate of candidates) {
-    if (!candidate) continue;
-    try {
-      parsed = JSON.parse(candidate);
-      parseError = null;
-      break;
-    } catch (e) {
-      parseError = e;
-    }
-  }
-
-  return { ...result, parsed, parseError };
-}
-
-module.exports = {
-  completeJson,
-  completeText,
-  readConfig,
-  findProfile,
-  aiProfilesPath,
-};
+// _site/lib/llm-client.js
+//
+// Tiny Anthropic Messages API client. Zero npm deps; supports HTTPS and HTTP
+// Anthropic-compatible endpoints. Connection settings are resolved from an AI
+// profile first, then environment variables as fallback:
+//
+//   profile.baseUrl       or ANTHROPIC_BASE_URL
+//   profile.apiKey        or ANTHROPIC_AUTH_TOKEN
+//   profile.mainModel     or ANTHROPIC_MODEL
+//   profile.timeoutMs     or API_TIMEOUT_MS
+//
+// This lets the local UI persist model settings in ai-profiles.json while still
+// keeping env vars useful for scripts and temporary overrides.
+
+const fs = require('fs');
+const path = require('path');
+const http = require('http');
+const https = require('https');
+const { URL } = require('url');
+
+const DEFAULT_BASE = 'https://api.anthropic.com';
+const DEFAULT_VERSION = '2023-06-01';
+const DEFAULT_MODEL = 'claude-haiku-4-5';
+const { getDataDir } = require('./data-dir');
+
+function aiProfilesPath() {
+  return path.join(getDataDir(), 'ai-profiles.json');
+}
+
+function readProfilesFile() {
+  try {
+    const cfg = JSON.parse(fs.readFileSync(aiProfilesPath(), 'utf-8'));
+    return Array.isArray(cfg.profiles) ? cfg.profiles : [];
+  } catch {
+    return [];
+  }
+}
+
+function findProfile(profileId) {
+  if (!profileId) return null;
+  return readProfilesFile().find(profile => profile && profile.id === profileId) || null;
+}
+
+function readConfig(options = {}) {
+  const profileId = typeof options === 'string' ? options : options.profileId;
+  const inlineProfile = options && typeof options === 'object' && options.profile && typeof options.profile === 'object'
+    ? options.profile
+    : null;
+  const profile = inlineProfile || findProfile(profileId) || {};
+  const timeoutMs = Number(profile.timeoutMs || process.env.API_TIMEOUT_MS || 60_000);
+  return {
+    profileId: profile.id || profileId || null,
+    baseUrl: profile.baseUrl || profile.apiBaseUrl || profile.anthropicBaseUrl || process.env.ANTHROPIC_BASE_URL || DEFAULT_BASE,
+    apiKey: profile.apiKey || profile.authToken || profile.anthropicAuthToken || process.env.ANTHROPIC_AUTH_TOKEN || '',
+    model: profile.mainModel || profile.model || process.env.ANTHROPIC_MODEL || DEFAULT_MODEL,
+    version: DEFAULT_VERSION,
+    timeoutMs: Number.isFinite(timeoutMs) && timeoutMs > 0 ? timeoutMs : 60_000,
+  };
+}
+
+function buildRequestPath(base, p) {
+  if (!p.startsWith('/')) return p;
+  const basePath = base.pathname.endsWith('/') ? base.pathname.slice(0, -1) : base.pathname;
+  return basePath + p;
+}
+
+function postJson({ baseUrl, path: p, body, apiKey, version, timeoutMs = 60_000 }) {
+  return new Promise((resolve, reject) => {
+    try {
+      const base = new URL(baseUrl);
+      const transport = base.protocol === 'http:' ? http : https;
+      const data = JSON.stringify(body);
+      const req = transport.request({
+        host: base.hostname,
+        port: base.port || (base.protocol === 'http:' ? 80 : 443),
+        path: buildRequestPath(base, p),
+        method: 'POST',
+        headers: {
+          'Content-Type': 'application/json',
+          'x-api-key': apiKey,
+          'anthropic-version': version,
+          'Content-Length': Buffer.byteLength(data),
+        },
+        timeout: timeoutMs,
+      }, res => {
+        const chunks = [];
+        res.on('data', c => chunks.push(c));
+        res.on('end', () => {
+          const text = Buffer.concat(chunks).toString('utf-8');
+          if (res.statusCode < 200 || res.statusCode >= 300) {
+            return reject(new Error(`HTTP ${res.statusCode}: ${text.slice(0, 500)}`));
+          }
+          try { resolve(JSON.parse(text)); }
+          catch (e) { reject(new Error(`bad JSON: ${e.message}; body head: ${text.slice(0, 200)}`)); }
+        });
+      });
+      req.on('error', reject);
+      req.on('timeout', () => req.destroy(new Error(`timeout after ${timeoutMs}ms`)));
+      req.write(data);
+      req.end();
+    } catch (e) {
+      reject(new Error(`bad baseUrl: ${baseUrl}: ${e.message}`));
+    }
+  });
+}
+
+async function completeText({ system, user, maxTokens = 512, model, profileId, profile }) {
+  const cfg = readConfig({ profileId, profile });
+  if (!cfg.apiKey) {
+    const id = profileId ? ` for profile ${profileId}` : '';
+    throw new Error(`API key not set${id}`);
+  }
+  const body = {
+    model: model || cfg.model,
+    max_tokens: maxTokens,
+    messages: [{ role: 'user', content: user }],
+  };
+  if (system) body.system = system;
+
+  const raw = await postJson({
+    baseUrl: cfg.baseUrl,
+    path: '/v1/messages',
+    body,
+    apiKey: cfg.apiKey,
+    version: cfg.version,
+    timeoutMs: cfg.timeoutMs,
+  });
+
+  const text = (raw.content || [])
+    .filter(block => block && block.type === 'text' && typeof block.text === 'string')
+    .map(block => block.text)
+    .join('\n')
+    .trim();
+
+  return {
+    text,
+    raw,
+    usage: raw.usage || null,
+    model: raw.model || body.model,
+    profileId: cfg.profileId,
+  };
+}
+
+async function completeJson({ system, user, schema, maxTokens = 2048, model, profileId, profile }) {
+  const schemaHint = schema ? `\n\nThe output MUST be a JSON object matching this schema:\n${schema}\n` : '';
+  const finalUser = `${user}${schemaHint}\n\nRespond with ONLY the JSON object, no prose, no markdown fences.`;
+  const result = await completeText({ system, user: finalUser, maxTokens, model, profileId, profile });
+  const text = result.text;
+
+  let parsed = null;
+  let parseError = null;
+  const candidates = [
+    text,
+    text.replace(/^```(?:json)?\s*/i, '').replace(/```\s*$/, ''),
+  ];
+  for (const candidate of candidates) {
+    if (!candidate) continue;
+    try {
+      parsed = JSON.parse(candidate);
+      parseError = null;
+      break;
+    } catch (e) {
+      parseError = e;
+    }
+  }
+
+  return { ...result, parsed, parseError };
+}
+
+module.exports = {
+  completeJson,
+  completeText,
+  readConfig,
+  findProfile,
+  aiProfilesPath,
+};

package/src/backend/lib/model-context-windows.js

@@ -1,82 +1,82 @@
-// Per-family context window sizes (in tokens) for common models.
-// Match strategy: lowercase the model name, then prefix-match — first hit wins.
-// Entry order matters: more specific prefixes (e.g. 'claude-opus-4') precede
-// broader ones (e.g. 'claude'). Keep specific entries above generic ones.
-const MODEL_CONTEXT_WINDOWS = [
-  // Anthropic Claude — full family 200K
-  ['claude-opus-4', 200000],
-  ['claude-sonnet-4', 200000],
-  ['claude-haiku-4', 200000],
-  ['claude-3-7', 200000],
-  ['claude-3-5', 200000],
-  ['claude-3', 200000],
-  ['claude', 200000],
-
-  // MiniMax
-  ['minimax-m3', 1048576],
-  ['minimax-m2', 1048576],
-  ['minimax', 1048576],
-
-  // Zhipu GLM
-  ['glm-5', 1048576],
-  ['glm-4.6', 1048576],
-  ['glm-4.5', 128000],
-  ['glm-4-plus', 128000],
-  ['glm-4', 128000],
-  ['glm', 128000],
-
-  // DeepSeek
-  ['deepseek-v3', 128000],
-  ['deepseek-r1', 128000],
-  ['deepseek', 64000],
-
-  // OpenAI
-  ['gpt-5', 200000],
-  ['gpt-4.1', 1047576],
-  ['gpt-4o', 128000],
-  ['gpt-4-turbo', 128000],
-  ['gpt-4', 8192],
-  ['o3', 200000],
-  ['o4-mini', 200000],
-  ['o1', 200000],
-
-  // Qwen
-  ['qwen3', 131072],
-  ['qwen-max', 32768],
-  ['qwen', 32768],
-
-  // Google Gemini
-  ['gemini-2.5', 1048576],
-  ['gemini-1.5', 1048576],
-  ['gemini', 32768],
-];
-
-function normalizeModelName(model) {
-  return String(model || '').trim().toLowerCase();
-}
-
-function lookupContextWindow(modelName) {
-  const name = normalizeModelName(modelName);
-  if (!name) return null;
-  for (const [prefix, size] of MODEL_CONTEXT_WINDOWS) {
-    if (name.startsWith(prefix)) return size;
-  }
-  return null;
-}
-
-// Resolution order: explicit profile override → model table → env var → fallback.
-function resolveContextWindow({ profileContextWindow, model, env = process.env, fallback = 200000 } = {}) {
-  const explicit = Number(profileContextWindow);
-  if (Number.isFinite(explicit) && explicit > 0) return Math.floor(explicit);
-  const tableHit = lookupContextWindow(model);
-  if (tableHit) return tableHit;
-  const envValue = Number(env && env.CLAUDE_CONTEXT_WINDOW);
-  if (Number.isFinite(envValue) && envValue > 0) return Math.floor(envValue);
-  return fallback;
-}
-
-module.exports = {
-  MODEL_CONTEXT_WINDOWS,
-  lookupContextWindow,
-  resolveContextWindow,
-};
+// Per-family context window sizes (in tokens) for common models.
+// Match strategy: lowercase the model name, then prefix-match — first hit wins.
+// Entry order matters: more specific prefixes (e.g. 'claude-opus-4') precede
+// broader ones (e.g. 'claude'). Keep specific entries above generic ones.
+const MODEL_CONTEXT_WINDOWS = [
+  // Anthropic Claude — full family 200K
+  ['claude-opus-4', 200000],
+  ['claude-sonnet-4', 200000],
+  ['claude-haiku-4', 200000],
+  ['claude-3-7', 200000],
+  ['claude-3-5', 200000],
+  ['claude-3', 200000],
+  ['claude', 200000],
+
+  // MiniMax
+  ['minimax-m3', 1048576],
+  ['minimax-m2', 1048576],
+  ['minimax', 1048576],
+
+  // Zhipu GLM
+  ['glm-5', 1048576],
+  ['glm-4.6', 1048576],
+  ['glm-4.5', 128000],
+  ['glm-4-plus', 128000],
+  ['glm-4', 128000],
+  ['glm', 128000],
+
+  // DeepSeek
+  ['deepseek-v3', 128000],
+  ['deepseek-r1', 128000],
+  ['deepseek', 64000],
+
+  // OpenAI
+  ['gpt-5', 200000],
+  ['gpt-4.1', 1047576],
+  ['gpt-4o', 128000],
+  ['gpt-4-turbo', 128000],
+  ['gpt-4', 8192],
+  ['o3', 200000],
+  ['o4-mini', 200000],
+  ['o1', 200000],
+
+  // Qwen
+  ['qwen3', 131072],
+  ['qwen-max', 32768],
+  ['qwen', 32768],
+
+  // Google Gemini
+  ['gemini-2.5', 1048576],
+  ['gemini-1.5', 1048576],
+  ['gemini', 32768],
+];
+
+function normalizeModelName(model) {
+  return String(model || '').trim().toLowerCase();
+}
+
+function lookupContextWindow(modelName) {
+  const name = normalizeModelName(modelName);
+  if (!name) return null;
+  for (const [prefix, size] of MODEL_CONTEXT_WINDOWS) {
+    if (name.startsWith(prefix)) return size;
+  }
+  return null;
+}
+
+// Resolution order: explicit profile override → model table → env var → fallback.
+function resolveContextWindow({ profileContextWindow, model, env = process.env, fallback = 200000 } = {}) {
+  const explicit = Number(profileContextWindow);
+  if (Number.isFinite(explicit) && explicit > 0) return Math.floor(explicit);
+  const tableHit = lookupContextWindow(model);
+  if (tableHit) return tableHit;
+  const envValue = Number(env && env.CLAUDE_CONTEXT_WINDOW);
+  if (Number.isFinite(envValue) && envValue > 0) return Math.floor(envValue);
+  return fallback;
+}
+
+module.exports = {
+  MODEL_CONTEXT_WINDOWS,
+  lookupContextWindow,
+  resolveContextWindow,
+};