Critical fix for Docker deployment where AI features were trying to connect to localhost:11434 instead of using configured provider (Ollama Docker service or OpenAI). Problems fixed: 1. Reformulation (clarify/shorten/improve) failing with ECONNREFUSED 127.0.0.1:11434 2. Auto-labels failing with same error 3. Notebook summaries failing 4. Could not switch from Ollama to OpenAI in admin Root cause: - Code had hardcoded fallback to 'http://localhost:11434' in multiple places - .env.docker file not created on server (gitignore'd) - No validation that required environment variables were set Changes: 1. lib/ai/factory.ts: - Remove hardcoded 'http://localhost:11434' fallback - Only use localhost for local development (NODE_ENV !== 'production') - Throw error if OLLAMA_BASE_URL not set in production 2. lib/ai/providers/ollama.ts: - Remove default parameter 'http://localhost:11434' from constructor - Require baseUrl to be explicitly passed - Throw error if baseUrl is missing 3. lib/ai/services/paragraph-refactor.service.ts: - Remove 'http://localhost:11434' fallback (2 locations) - Require OLLAMA_BASE_URL to be set - Throw clear error if not configured 4. app/(main)/admin/settings/admin-settings-form.tsx: - Add debug info showing current provider state - Display database config value for transparency - Help troubleshoot provider selection issues 5. DOCKER-SETUP.md: - Complete guide for Docker configuration - Instructions for .env.docker setup - Examples for Ollama Docker, OpenAI, and external Ollama - Troubleshooting common issues Usage: On server, create .env.docker with proper provider configuration: - Ollama in Docker: OLLAMA_BASE_URL="http://ollama:11434" - OpenAI: OPENAI_API_KEY="sk-..." - External Ollama: OLLAMA_BASE_URL="http://SERVER_IP:11434" Then in admin interface, users can independently configure: - Tags Provider (for auto-labels, AI features) - Embeddings Provider (for semantic search) Result: ✓ Clear errors if Ollama not configured ✓ Can switch to OpenAI freely in admin ✓ No more hardcoded localhost in production ✓ Proper separation between local dev and Docker production Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
138 lines
4.2 KiB
TypeScript
138 lines
4.2 KiB
TypeScript
import { AIProvider, TagSuggestion, TitleSuggestion } from '../types';
|
|
|
|
export class OllamaProvider implements AIProvider {
|
|
private baseUrl: string;
|
|
private modelName: string;
|
|
private embeddingModelName: string;
|
|
|
|
constructor(baseUrl: string, modelName: string = 'llama3', embeddingModelName?: string) {
|
|
if (!baseUrl) {
|
|
throw new Error('baseUrl is required for OllamaProvider')
|
|
}
|
|
// Ensure baseUrl ends with /api for Ollama API
|
|
this.baseUrl = baseUrl.endsWith('/api') ? baseUrl : `${baseUrl}/api`;
|
|
this.modelName = modelName;
|
|
this.embeddingModelName = embeddingModelName || modelName;
|
|
}
|
|
|
|
async generateTags(content: string): Promise<TagSuggestion[]> {
|
|
try {
|
|
const response = await fetch(`${this.baseUrl}/generate`, {
|
|
method: 'POST',
|
|
headers: { 'Content-Type': 'application/json' },
|
|
body: JSON.stringify({
|
|
model: this.modelName,
|
|
prompt: `Analyse la note suivante et extrais les concepts clés sous forme de tags courts (1-3 mots max).
|
|
|
|
Règles:
|
|
- Pas de mots de liaison (le, la, pour, et...).
|
|
- Garde les expressions composées ensemble (ex: "semaine prochaine", "New York").
|
|
- Normalise en minuscules sauf noms propres.
|
|
- Maximum 5 tags.
|
|
|
|
Réponds UNIQUEMENT sous forme de liste JSON d'objets : [{"tag": "string", "confidence": number}].
|
|
|
|
Contenu de la note: "${content}"`,
|
|
stream: false,
|
|
}),
|
|
});
|
|
|
|
if (!response.ok) throw new Error(`Ollama error: ${response.statusText}`);
|
|
|
|
const data = await response.json();
|
|
const text = data.response;
|
|
|
|
const jsonMatch = text.match(/\[\s*\{[\s\S]*\}\s*\]/);
|
|
if (jsonMatch) {
|
|
return JSON.parse(jsonMatch[0]);
|
|
}
|
|
|
|
// Support pour le format { "tags": [...] }
|
|
const objectMatch = text.match(/\{\s*"tags"\s*:\s*(\[[\s\S]*\])\s*\}/);
|
|
if (objectMatch && objectMatch[1]) {
|
|
return JSON.parse(objectMatch[1]);
|
|
}
|
|
|
|
return [];
|
|
} catch (e) {
|
|
console.error('Erreur API directe Ollama:', e);
|
|
return [];
|
|
}
|
|
}
|
|
|
|
async getEmbeddings(text: string): Promise<number[]> {
|
|
try {
|
|
const response = await fetch(`${this.baseUrl}/embeddings`, {
|
|
method: 'POST',
|
|
headers: { 'Content-Type': 'application/json' },
|
|
body: JSON.stringify({
|
|
model: this.embeddingModelName,
|
|
prompt: text,
|
|
}),
|
|
});
|
|
|
|
if (!response.ok) throw new Error(`Ollama error: ${response.statusText}`);
|
|
|
|
const data = await response.json();
|
|
return data.embedding;
|
|
} catch (e) {
|
|
console.error('Erreur embeddings directs Ollama:', e);
|
|
return [];
|
|
}
|
|
}
|
|
|
|
async generateTitles(prompt: string): Promise<TitleSuggestion[]> {
|
|
try {
|
|
const response = await fetch(`${this.baseUrl}/generate`, {
|
|
method: 'POST',
|
|
headers: { 'Content-Type': 'application/json' },
|
|
body: JSON.stringify({
|
|
model: this.modelName,
|
|
prompt: `${prompt}
|
|
|
|
Réponds UNIQUEMENT sous forme de tableau JSON : [{"title": "string", "confidence": number}]`,
|
|
stream: false,
|
|
}),
|
|
});
|
|
|
|
if (!response.ok) throw new Error(`Ollama error: ${response.statusText}`);
|
|
|
|
const data = await response.json();
|
|
const text = data.response;
|
|
|
|
// Extraire le JSON de la réponse
|
|
const jsonMatch = text.match(/\[\s*\{[\s\S]*\}\s*\]/);
|
|
if (jsonMatch) {
|
|
return JSON.parse(jsonMatch[0]);
|
|
}
|
|
|
|
return [];
|
|
} catch (e) {
|
|
console.error('Erreur génération titres Ollama:', e);
|
|
return [];
|
|
}
|
|
}
|
|
|
|
async generateText(prompt: string): Promise<string> {
|
|
try {
|
|
const response = await fetch(`${this.baseUrl}/generate`, {
|
|
method: 'POST',
|
|
headers: { 'Content-Type': 'application/json' },
|
|
body: JSON.stringify({
|
|
model: this.modelName,
|
|
prompt: prompt,
|
|
stream: false,
|
|
}),
|
|
});
|
|
|
|
if (!response.ok) throw new Error(`Ollama error: ${response.statusText}`);
|
|
|
|
const data = await response.json();
|
|
return data.response.trim();
|
|
} catch (e) {
|
|
console.error('Erreur génération texte Ollama:', e);
|
|
throw e;
|
|
}
|
|
}
|
|
}
|