nothumanallowed 9.7.2 → 9.8.1

package/src/services/conversations.mjs

@@ -0,0 +1,277 @@
+/**
+ * Multi-conversation manager for NHA Chat.
+ *
+ * Stores conversations in ~/.nha/conversations/ as JSON files.
+ * Each conversation has an ID, auto-generated title, and message history.
+ *
+ * Zero npm dependencies.
+ */
+
+import fs from 'fs';
+import path from 'path';
+import crypto from 'crypto';
+import { NHA_DIR } from '../constants.mjs';
+
+const CONVERSATIONS_DIR = path.join(NHA_DIR, 'conversations');
+const ACTIVE_FILE = path.join(CONVERSATIONS_DIR, '.active');
+const MAX_TITLE_LENGTH = 60;
+const MAX_CONVERSATIONS = 100;
+
+// ── Helpers ──────────────────────────────────────────────────────────────────
+
+function ensureDir() {
+  fs.mkdirSync(CONVERSATIONS_DIR, { recursive: true });
+}
+
+function convPath(id) {
+  return path.join(CONVERSATIONS_DIR, `${id}.json`);
+}
+
+function generateId() {
+  return crypto.randomUUID().slice(0, 8);
+}
+
+/**
+ * Auto-generate a title from the first user message.
+ * Takes the first ~50 chars, trims to last word boundary.
+ */
+function autoTitle(firstMessage) {
+  if (!firstMessage) return 'New Chat';
+  let title = firstMessage.replace(/\s+/g, ' ').trim();
+  if (title.length > MAX_TITLE_LENGTH) {
+    title = title.slice(0, MAX_TITLE_LENGTH);
+    const lastSpace = title.lastIndexOf(' ');
+    if (lastSpace > 20) title = title.slice(0, lastSpace);
+    title += '...';
+  }
+  return title;
+}
+
+// ── CRUD ─────────────────────────────────────────────────────────────────────
+
+/**
+ * Create a new conversation and set it as active.
+ * @returns {{ id: string, title: string, messages: Array, createdAt: string, updatedAt: string }}
+ */
+export function createConversation(title = '') {
+  ensureDir();
+  const id = generateId();
+  const now = new Date().toISOString();
+  const conv = {
+    id,
+    title: title || 'New Chat',
+    messages: [],
+    createdAt: now,
+    updatedAt: now,
+  };
+  fs.writeFileSync(convPath(id), JSON.stringify(conv, null, 2) + '\n', 'utf-8');
+  setActiveId(id);
+  return conv;
+}
+
+/**
+ * Load a conversation by ID.
+ * @returns {object|null}
+ */
+export function loadConversation(id) {
+  try {
+    const data = fs.readFileSync(convPath(id), 'utf-8');
+    return JSON.parse(data);
+  } catch {
+    return null;
+  }
+}
+
+/**
+ * Save a conversation (full overwrite).
+ */
+export function saveConversation(conv) {
+  ensureDir();
+  conv.updatedAt = new Date().toISOString();
+  fs.writeFileSync(convPath(conv.id), JSON.stringify(conv, null, 2) + '\n', 'utf-8');
+}
+
+/**
+ * Delete a conversation by ID.
+ * If it was active, clears active state.
+ * @returns {boolean} true if deleted
+ */
+export function deleteConversation(id) {
+  const filePath = convPath(id);
+  if (!fs.existsSync(filePath)) return false;
+  fs.unlinkSync(filePath);
+  if (getActiveId() === id) {
+    clearActiveId();
+  }
+  return true;
+}
+
+/**
+ * List all conversations, sorted by updatedAt (newest first).
+ * Returns summary objects (no messages).
+ * @returns {Array<{ id: string, title: string, messageCount: number, createdAt: string, updatedAt: string }>}
+ */
+export function listConversations() {
+  ensureDir();
+  const files = fs.readdirSync(CONVERSATIONS_DIR)
+    .filter(f => f.endsWith('.json') && !f.startsWith('.'));
+
+  const convs = [];
+  for (const f of files) {
+    try {
+      const data = JSON.parse(fs.readFileSync(path.join(CONVERSATIONS_DIR, f), 'utf-8'));
+      convs.push({
+        id: data.id,
+        title: data.title || 'Untitled',
+        messageCount: (data.messages || []).length,
+        createdAt: data.createdAt,
+        updatedAt: data.updatedAt,
+      });
+    } catch { /* skip corrupt files */ }
+  }
+
+  convs.sort((a, b) => new Date(b.updatedAt) - new Date(a.updatedAt));
+
+  // Auto-prune old conversations beyond limit
+  if (convs.length > MAX_CONVERSATIONS) {
+    for (const old of convs.slice(MAX_CONVERSATIONS)) {
+      try { fs.unlinkSync(convPath(old.id)); } catch {}
+    }
+    return convs.slice(0, MAX_CONVERSATIONS);
+  }
+
+  return convs;
+}
+
+// ── Active Conversation ──────────────────────────────────────────────────────
+
+/**
+ * Get the active conversation ID.
+ * @returns {string|null}
+ */
+export function getActiveId() {
+  try {
+    return fs.readFileSync(ACTIVE_FILE, 'utf-8').trim() || null;
+  } catch {
+    return null;
+  }
+}
+
+/**
+ * Set the active conversation ID.
+ */
+export function setActiveId(id) {
+  ensureDir();
+  fs.writeFileSync(ACTIVE_FILE, id, 'utf-8');
+}
+
+/**
+ * Clear the active conversation.
+ */
+export function clearActiveId() {
+  try { fs.unlinkSync(ACTIVE_FILE); } catch {}
+}
+
+/**
+ * Get or create the active conversation.
+ * If none exists, creates a new one.
+ * @returns {object} conversation object
+ */
+export function getOrCreateActive() {
+  const activeId = getActiveId();
+  if (activeId) {
+    const conv = loadConversation(activeId);
+    if (conv) return conv;
+  }
+  return createConversation();
+}
+
+/**
+ * Add a message pair (user + assistant) to a conversation.
+ * Auto-titles the conversation from the first user message.
+ */
+export function addMessages(conv, userContent, assistantContent) {
+  conv.messages.push({ role: 'user', content: userContent });
+  conv.messages.push({ role: 'assistant', content: assistantContent });
+
+  // Auto-title from first user message
+  if (conv.title === 'New Chat' && conv.messages.length === 2) {
+    conv.title = autoTitle(userContent);
+  }
+
+  saveConversation(conv);
+}
+
+/**
+ * Get the message history from a conversation, capped at maxTurns pairs.
+ * @returns {Array<{role: string, content: string}>}
+ */
+export function getHistory(conv, maxTurns = 20) {
+  const messages = conv.messages || [];
+  return messages.slice(-(maxTurns * 2));
+}
+
+// ── Export ────────────────────────────────────────────────────────────────────
+
+/**
+ * Export conversation as Markdown.
+ */
+export function exportAsMarkdown(conv) {
+  const lines = [
+    `# ${conv.title}`,
+    `*Created: ${new Date(conv.createdAt).toLocaleString()}*`,
+    `*Messages: ${conv.messages.length}*`,
+    '',
+    '---',
+    '',
+  ];
+
+  for (const msg of conv.messages) {
+    if (msg.role === 'user') {
+      lines.push(`### You`);
+      lines.push(msg.content);
+    } else {
+      lines.push(`### NHA`);
+      lines.push(msg.content);
+    }
+    lines.push('');
+  }
+
+  return lines.join('\n');
+}
+
+/**
+ * Export conversation as JSON.
+ */
+export function exportAsJson(conv) {
+  return JSON.stringify({
+    id: conv.id,
+    title: conv.title,
+    createdAt: conv.createdAt,
+    updatedAt: conv.updatedAt,
+    messages: conv.messages,
+  }, null, 2);
+}
+
+// ── Migration ────────────────────────────────────────────────────────────────
+
+/**
+ * Migrate old single-file chat history to multi-conversation format.
+ * Called once on first run if old history exists.
+ */
+export function migrateOldHistory() {
+  const oldFile = path.join(NHA_DIR, 'memory', 'chat-history.json');
+  if (!fs.existsSync(oldFile)) return;
+
+  try {
+    const messages = JSON.parse(fs.readFileSync(oldFile, 'utf-8'));
+    if (!Array.isArray(messages) || messages.length === 0) return;
+
+    const conv = createConversation('Previous Chat');
+    conv.messages = messages;
+    saveConversation(conv);
+
+    // Rename old file to avoid re-migration
+    fs.renameSync(oldFile, oldFile + '.migrated');
+  } catch { /* migration failed — non-critical */ }
+}

package/src/services/llm.mjs

@@ -276,110 +276,141 @@ export async function callLLM(config, systemPrompt, userMessage, opts = {}) {
 }
 
 /**
- * Call LLM with multimodal (vision) messages — supports image content.
- * Uses the provider's native vision format.
- * @param {object} config
- * @param {Array} messages - Array of { role, content } where content can be string or array of content blocks
- * @returns {Promise<string>}
+ * Call an LLM provider with streaming enabled.
+ * Calls onToken(chunk) for each token, returns full text at the end.
+ * @returns {Promise<string>} The full LLM response text.
  */
-export async function callLLMVision(config, messages) {
-  const provider = config.llm.provider || 'anthropic';
-  const model = config.llm.model || null;
+export async function callLLMStream(config, systemPrompt, userMessage, onToken, opts = {}) {
+  const provider = opts.provider || config.llm.provider || 'anthropic';
+  const model = opts.model || config.llm.model || null;
   const apiKey = getApiKey(config, provider);
-  if (!apiKey) throw new Error(`No API key for ${provider}. Vision requires Claude, GPT-4, or Gemini.`);
+  if (!apiKey) throw new Error(`No API key for ${provider}`);
 
-  if (provider === 'anthropic') {
-    // Anthropic format: system separate, messages with content blocks
-    const systemMsg = messages.find(m => m.role === 'system');
-    const userMsgs = messages.filter(m => m.role !== 'system');
-
-    // Convert OpenAI-style image_url to Anthropic format
-    const anthropicMessages = userMsgs.map(m => {
-      if (typeof m.content === 'string') return m;
-      const blocks = m.content.map(block => {
-        if (block.type === 'text') return block;
-        if (block.type === 'image_url') {
-          const url = block.image_url.url;
-          const match = url.match(/^data:image\/(png|jpeg|gif|webp);base64,(.+)$/);
-          if (match) {
-            return { type: 'image', source: { type: 'base64', media_type: `image/${match[1]}`, data: match[2] } };
-          }
-        }
-        return block;
-      });
-      return { role: m.role, content: blocks };
-    });
+  const callFn = getProviderCall(provider);
+  if (!callFn) throw new Error(`Unknown provider: ${provider}`);
+
+  // Gemini and Cohere don't support streaming — fall back to non-streaming
+  if (provider === 'gemini' || provider === 'cohere') {
+    const text = await callFn(apiKey, model, systemPrompt, userMessage, false);
+    if (onToken) onToken(text);
+    return text;
+  }
+
+  const format = provider === 'anthropic' ? 'anthropic' : 'openai';
+  const body = buildRequestBody(provider, model, systemPrompt, userMessage, true);
+  const url = getProviderUrl(provider, model, apiKey);
+  const headers = getProviderHeaders(provider, apiKey);
+
+  const res = await fetch(url, {
+    method: 'POST',
+    headers,
+    body: JSON.stringify(body),
+  });
+  if (!res.ok) {
+    const err = await res.text();
+    throw new Error(`${provider} ${res.status}: ${err}`);
+  }
 
-    const body = {
+  return streamSSEWithCallback(res, format, onToken);
+}
+
+/** Build request body for a provider */
+function buildRequestBody(provider, model, systemPrompt, userMessage, stream) {
+  if (provider === 'anthropic') {
+    return {
       model: model || 'claude-sonnet-4-20250514',
-      max_tokens: 4096,
-      system: systemMsg?.content || '',
-      messages: anthropicMessages,
+      max_tokens: 8192,
+      system: systemPrompt,
+      messages: [{ role: 'user', content: userMessage }],
+      stream,
     };
-    const res = await fetch('https://api.anthropic.com/v1/messages', {
-      method: 'POST',
-      headers: { 'Content-Type': 'application/json', 'x-api-key': apiKey, 'anthropic-version': '2023-06-01' },
-      body: JSON.stringify(body),
-    });
-    if (!res.ok) throw new Error(`Anthropic vision ${res.status}: ${await res.text()}`);
-    const data = await res.json();
-    return data.content?.[0]?.text || '';
   }
+  // OpenAI-compatible format (OpenAI, DeepSeek, Grok, Mistral)
+  const modelDefaults = {
+    openai: 'gpt-4o',
+    deepseek: 'deepseek-chat',
+    grok: 'grok-3-latest',
+    mistral: 'mistral-large-latest',
+  };
+  return {
+    model: model || modelDefaults[provider] || 'gpt-4o',
+    max_tokens: 8192,
+    messages: [
+      { role: 'system', content: systemPrompt },
+      { role: 'user', content: userMessage },
+    ],
+    stream,
+  };
+}
+
+/** Get provider API URL */
+function getProviderUrl(provider, model, apiKey) {
+  const urls = {
+    anthropic: 'https://api.anthropic.com/v1/messages',
+    openai: 'https://api.openai.com/v1/chat/completions',
+    deepseek: 'https://api.deepseek.com/v1/chat/completions',
+    grok: 'https://api.x.ai/v1/chat/completions',
+    mistral: 'https://api.mistral.ai/v1/chat/completions',
+  };
+  return urls[provider] || urls.openai;
+}
 
-  if (provider === 'openai' || provider === 'deepseek' || provider === 'grok' || provider === 'mistral') {
-    // OpenAI-compatible format — works with GPT-4V, DeepSeek VL, etc.
-    const url = provider === 'openai' ? 'https://api.openai.com/v1/chat/completions'
-              : provider === 'deepseek' ? 'https://api.deepseek.com/chat/completions'
-              : provider === 'grok' ? 'https://api.x.ai/v1/chat/completions'
-              : 'https://api.mistral.ai/v1/chat/completions';
-
-    const visionModel = model || (provider === 'openai' ? 'gpt-4o' : model);
-    const body = { model: visionModel, max_tokens: 4096, messages };
-    const res = await fetch(url, {
-      method: 'POST',
-      headers: { 'Content-Type': 'application/json', 'Authorization': `Bearer ${apiKey}` },
-      body: JSON.stringify(body),
-    });
-    if (!res.ok) throw new Error(`${provider} vision ${res.status}: ${await res.text()}`);
-    const data = await res.json();
-    return data.choices?.[0]?.message?.content || '';
+/** Get provider request headers */
+function getProviderHeaders(provider, apiKey) {
+  if (provider === 'anthropic') {
+    return {
+      'Content-Type': 'application/json',
+      'x-api-key': apiKey,
+      'anthropic-version': '2023-06-01',
+    };
   }
+  return {
+    'Content-Type': 'application/json',
+    'Authorization': `Bearer ${apiKey}`,
+  };
+}
+
+/** SSE stream parser with onToken callback (does NOT write to stdout directly) */
+async function streamSSEWithCallback(res, format, onToken) {
+  const reader = res.body.getReader();
+  const decoder = new TextDecoder();
+  let buffer = '';
+  let fullText = '';
+
+  while (true) {
+    const { done, value } = await reader.read();
+    if (done) break;
+
+    buffer += decoder.decode(value, { stream: true });
+    const lines = buffer.split('\n');
+    buffer = lines.pop() || '';
+
+    for (const line of lines) {
+      if (!line.startsWith('data: ')) continue;
+      const data = line.slice(6).trim();
+      if (data === '[DONE]') continue;
+
+      try {
+        const json = JSON.parse(data);
+        let chunk = '';
 
-  if (provider === 'gemini') {
-    // Gemini format — inline_data with base64
-    const systemMsg = messages.find(m => m.role === 'system');
-    const userMsgs = messages.filter(m => m.role !== 'system');
-    const parts = [];
-    for (const msg of userMsgs) {
-      if (typeof msg.content === 'string') {
-        parts.push({ text: msg.content });
-      } else {
-        for (const block of msg.content) {
-          if (block.type === 'text') parts.push({ text: block.text });
-          if (block.type === 'image_url') {
-            const match = block.image_url.url.match(/^data:image\/(.*?);base64,(.+)$/);
-            if (match) parts.push({ inline_data: { mime_type: `image/${match[1]}`, data: match[2] } });
+        if (format === 'anthropic') {
+          if (json.type === 'content_block_delta') {
+            chunk = json.delta?.text || '';
           }
+        } else {
+          chunk = json.choices?.[0]?.delta?.content || '';
         }
-      }
-    }
 
-    const geminiModel = model || 'gemini-2.0-flash';
-    const body = {
-      contents: [{ parts }],
-      systemInstruction: systemMsg ? { parts: [{ text: systemMsg.content }] } : undefined,
-    };
-    const res = await fetch(`https://generativelanguage.googleapis.com/v1beta/models/${geminiModel}:generateContent?key=${apiKey}`, {
-      method: 'POST',
-      headers: { 'Content-Type': 'application/json' },
-      body: JSON.stringify(body),
-    });
-    if (!res.ok) throw new Error(`Gemini vision ${res.status}: ${await res.text()}`);
-    const data = await res.json();
-    return data.candidates?.[0]?.content?.parts?.[0]?.text || '';
+        if (chunk) {
+          fullText += chunk;
+          if (onToken) onToken(chunk);
+        }
+      } catch {}
+    }
   }
 
-  throw new Error(`Vision not supported for provider: ${provider}. Use Claude, GPT-4, or Gemini.`);
+  return fullText;
 }
 
 /**