evandro/OllaHub
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

wip: snapshot das alterações não commitadas (2026-10-20)
release.yml (push) Invalid workflow file
Details

Browse Source 
Origem: D:/Documents/Github/OllaHub
Arquivos omitidos (segredos, bancos de dados, logs, >50 MB): 0
This commit is contained in:
evandro committed 2026-10-20 00:49:46 -03:00
1 parent a7cbd1a3b4
commit 861af0bb9f
217 files changed
+39421 -1340
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

hooks/use-chat.ts
+49 -49
View File
@@ -4,7 +4,7 @@ import { listen } from '@tauri-apps/api/event';
import { removeMetadataNoise } from '@/lib/metadata';
import { useSettingsStore } from '@/store/settings-store';
export type ThinkingStepType =
export type ThinkingStepType =
| 'preprocessing'
| 'web-research'
| 'sources-found'
@@ -27,7 +27,7 @@ export interface ThinkingMessageMetadata {
timestamp: number;
duration?: number;
activeQueries?: Array<{ query: string; source: string; round?: number; startedAt: number }>;
activeUrls?: Array<{ url: string; title?: string; status: string; source?: string; startedAt: number; duration?: number }>;
activeUrls?: Array<{ url: string; title?: string; status: 'started' | 'completed' | 'failed' | 'cached'; source?: string; startedAt: number; duration?: number }>;
}
export interface Message {
@@ -46,15 +46,15 @@ function parseMetadata(fullContent: string): { content: string, metadata?: any }
if (!fullContent || fullContent.trim() === '') {
return { content: fullContent || '' };
}
// More flexible regex that handles whitespace and line breaks (case-insensitive)
const metadataRegex = /<metadata>\s*([\s\S]*?)\s*<\/metadata>/gi;
// First, extract all metadata blocks
const metadataMatches: Array<{ match: string, json: any }> = [];
let match;
let lastIndex = 0;
while ((match = metadataRegex.exec(fullContent)) !== null) {
const metadataStr = match[1].trim();
try {
@@ -66,11 +66,11 @@ function parseMetadata(fullContent: string): { content: string, metadata?: any }
}
lastIndex = metadataRegex.lastIndex;
}
// Remove ALL metadata blocks from content
let content = fullContent.replace(/<metadata>\s*[\s\S]*?\s*<\/metadata>/gi, '').trim();
content = removeMetadataNoise(content);
// Also remove common prefixes if they appear at the end of the content
const prefixes = [
'Bloco oculto de metadados:',
@@ -80,30 +80,30 @@ function parseMetadata(fullContent: string): { content: string, metadata?: any }
'JSON metadata:',
'---'
];
for (const prefix of prefixes) {
const regex = new RegExp(`${prefix}\\s*$`, 'i');
if (regex.test(content)) {
content = content.replace(regex, '').trim();
}
}
// Ensure we don't return empty content if there was content before metadata
if (!content && fullContent.length > 0) {
// If all content was metadata, keep nothing (shouldn't happen but safe fallback)
content = '';
}
// Return first valid metadata (if any)
const metadata = metadataMatches.length > 0 ? metadataMatches[0].json : undefined;
return { content, metadata };
}
// Helper to strip metadata in real-time during streaming (for display)
function stripMetadataTag(content: string): string {
if (!content) return '';
// Common prefixes that models might output before the metadata block
const prefixes = [
'Bloco oculto de metadados:',
@@ -113,13 +113,13 @@ function stripMetadataTag(content: string): string {
'JSON metadata:',
'---'
];
let cleaned = content;
// Remove any metadata tags (complete or incomplete) for real-time display
// Case-insensitive regex to catch variations
cleaned = cleaned.replace(/<metadata>[\s\S]*?$/gi, '').replace(/<metadata>[\s\S]*?<\/metadata>/gi, '').trim();
// Aggressively strip known prefixes if they appear at the very end of the content
// This helps hide "Bloco oculto de metadados:" while the tag is being generated
for (const prefix of prefixes) {
@@ -128,18 +128,18 @@ function stripMetadataTag(content: string): string {
cleaned = cleaned.replace(regex, '').trim();
}
}
return removeMetadataNoise(cleaned);
}
// Helper to detect tool calls in LLM response
export function detectToolCalls(content: string, availableTools: any[]): ToolCall[] {
const toolCalls: ToolCall[] = [];
// Simple pattern matching for tool calls
// Pattern: "use tool X with args Y" or "execute X(args)"
// This is a basic implementation - can be enhanced with better NLP
// Look for common patterns
const patterns = [
// Pattern 1: "use [tool_name] with [args]" or "use [tool_name] (args)"
@@ -147,13 +147,13 @@ export function detectToolCalls(content: string, availableTools: any[]): ToolCal
// Pattern 2: "[tool_name](args)"
/([a-z_]+)\s*\(([^)]+)\)/gi,
];
for (const pattern of patterns) {
let match;
while ((match = pattern.exec(content)) !== null) {
const potentialToolName = match[1].toLowerCase().trim();
const potentialArgs = match[2]?.trim() || '{}';
// Find matching tool in available tools
for (const toolInfo of availableTools) {
if (toolInfo.tool.name.toLowerCase() === potentialToolName) {
@@ -179,7 +179,7 @@ export function detectToolCalls(content: string, availableTools: any[]): ToolCal
args = {};
}
}
toolCalls.push({
server_name: toolInfo.server_name,
tool_name: toolInfo.tool.name,
@@ -190,7 +190,7 @@ export function detectToolCalls(content: string, availableTools: any[]): ToolCal
}
}
}
return toolCalls;
}
@@ -202,7 +202,7 @@ export async function executeToolCall(toolCall: ToolCall): Promise<string> {
toolName: toolCall.tool_name,
arguments: toolCall.arguments,
});
// Format result for display
if (typeof result === 'string') {
return result;
@@ -244,7 +244,7 @@ export function useChat() {
const lastTokenRef = useRef<string>(''); // Track last token to detect duplicates
const renderScheduledRef = useRef<boolean>(false);
const rafIdRef = useRef<number | null>(null);
// Função para atualizar o estado visual com o conteúdo acumulado
const flushToState = () => {
renderScheduledRef.current = false;
@@ -265,13 +265,13 @@ export function useChat() {
rafIdRef.current = requestAnimationFrame(flushToState);
}
};
// Listener para eventos do Rust
useEffect(() => {
let unlistenCreated: (() => void) | null = null;
let unlistenToken: (() => void) | null = null;
let unlistenError: (() => void) | null = null;
// Listener para chat-created (título gerado)
listen<ChatCreatedEvent>('chat-created', (event) => {
const { session_id, title, emoji } = event.payload;
@@ -281,11 +281,11 @@ export function useChat() {
}).then(unlisten => {
unlistenCreated = unlisten;
});
// Listener para chat-token (streaming de tokens)
listen<ChatTokenEvent>('chat-token', (event) => {
const { content, done } = event.payload;
if (content && content.length > 0) {
// Detecção de duplicatas (MANTER LÓGICA ORIGINAL)
if (content === lastTokenRef.current && rawContentRef.current.endsWith(content)) {
@@ -294,14 +294,14 @@ export function useChat() {
}
return;
}
lastTokenRef.current = content;
rawContentRef.current += content;
// MUDANÇA: Agendar flush em vez de chamar setMessages diretamente
scheduleFlush();
}
if (done) {
// Cancelar qualquer flush pendente
if (rafIdRef.current !== null) {
@@ -309,7 +309,7 @@ export function useChat() {
rafIdRef.current = null;
renderScheduledRef.current = false;
}
// Flush IMEDIATO com processamento de metadata
setMessages(prev => {
const last = prev[prev.length - 1];
@@ -327,7 +327,7 @@ export function useChat() {
}).then(unlisten => {
unlistenToken = unlisten;
});
// Listener para chat-error
listen<ChatErrorEvent>('chat-error', (event) => {
const { error } = event.payload;
@@ -337,7 +337,7 @@ export function useChat() {
}).then(unlisten => {
unlistenError = unlisten;
});
// Cleanup
return () => {
unlistenCreated?.();
@@ -362,14 +362,14 @@ export function useChat() {
}
) => {
setIsLoading(true);
// Verificar se a última mensagem já é do usuário (para evitar duplicação)
// Se for, usar as mensagens existentes; caso contrário, adicionar nova mensagem do usuário
const lastMessage = messages[messages.length - 1];
const newMessages: Message[] = lastMessage?.role === 'user' && lastMessage?.content === content
? [...messages] // Já existe, não adicionar novamente
: [...messages, { role: 'user', content }]; // Adicionar nova mensagem do usuário
setMessages(newMessages);
// Prepare messages for API (include system prompt if present)
@@ -385,40 +385,40 @@ export function useChat() {
const settings = useSettingsStore.getState();
const maxTokens = settings.contextWindow || 4096;
const maxMessages = 10; // Máximo de mensagens a manter
// Estimar tokens (aproximação: 1 token ≈ 4 caracteres)
function estimateTokens(text: string): number {
return Math.ceil(text.length / 4);
}
// Aplicar janela deslizante: manter últimas N mensagens que cabem no limite de tokens
let truncatedMessages = [...newMessages];
let totalTokens = systemPrompt ? estimateTokens(systemPrompt) : 0;
// Sempre manter a última mensagem do usuário
const lastUserMessage = truncatedMessages[truncatedMessages.length - 1];
truncatedMessages = truncatedMessages.slice(0, -1);
// Adicionar mensagens de trás para frente até atingir o limite
const selectedMessages: Message[] = [];
for (let i = truncatedMessages.length - 1; i >= 0; i--) {
const msg = truncatedMessages[i];
const msgTokens = estimateTokens(msg.content);
// Limitar por número de mensagens E por tokens
if (selectedMessages.length >= maxMessages - 1) break; // -1 porque vamos adicionar a última depois
if (totalTokens + msgTokens > maxTokens * 0.9) break; // 90% do limite para margem de segurança
selectedMessages.unshift(msg);
totalTokens += msgTokens;
}
// Adicionar a última mensagem do usuário
selectedMessages.push(lastUserMessage);
totalTokens += estimateTokens(lastUserMessage.content);
// Constrói mensagens para payload, aplicando override apenas à última mensagem do usuário
const baseMessages = systemPrompt
const baseMessages = systemPrompt
? [{ role: 'system', content: systemPrompt }, ...selectedMessages]
: selectedMessages;
@@ -458,7 +458,7 @@ export function useChat() {
// Reset raw content buffer for this message
rawContentRef.current = '';
lastTokenRef.current = ''; // Reset last token tracker
// Add empty assistant message to start streaming into
setMessages(prev => {
const last = prev[prev.length - 1];
@@ -493,14 +493,14 @@ export function useChat() {
systemPrompt: systemPrompt || null,
enableRag: false, // TODO: Implementar RAG depois
});
// O streaming será processado pelos listeners de eventos
// Não precisamos fazer mais nada aqui, os eventos cuidam do resto
} catch (error: any) {
console.error('Chat error:', error);
let errorMessage = 'Erro ao obter resposta do Ollama.';
if (error.name === 'AbortError' || error.message === 'RequestTimeout') {
errorMessage = '**Timeout:** A requisição demorou muito para responder. Tente novamente ou use um modelo menor.';
} else if (error.message === 'OllamaNotRunning') {
@@ -522,7 +522,7 @@ export function useChat() {
const errorDetails = error.message || error.toString();
errorMessage = `**Erro:** Falha ao obter resposta do Ollama.\n\nDetalhes: ${errorDetails}`;
}
setMessages(prev => [...prev, { role: 'assistant', content: errorMessage }]);
} finally {
setIsLoading(false);