@retrivora-ai/rag-engine 2.4.2 → 2.4.5
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/{BatchProcessor-zbQMaZHm.d.mts → BatchProcessor-Csy_K1qQ.d.mts} +2 -0
- package/dist/{BatchProcessor-D42E706a.d.ts → BatchProcessor-Dg18zZGk.d.ts} +2 -0
- package/dist/handlers/index.js +31 -5
- package/dist/handlers/index.mjs +31 -5
- package/dist/index.d.mts +3 -3
- package/dist/index.d.ts +3 -3
- package/dist/index.js +41 -21
- package/dist/index.min.css +2 -0
- package/dist/index.mjs +41 -21
- package/dist/server.d.mts +2 -2
- package/dist/server.d.ts +2 -2
- package/dist/server.js +32 -5
- package/dist/server.mjs +32 -5
- package/package.json +5 -4
- package/src/LicensedRetrivora.ts +0 -278
- package/src/app/api/chat/route.ts +0 -4
- package/src/app/api/health/route.ts +0 -5
- package/src/app/api/ingest/route.ts +0 -27
- package/src/app/api/suggestions/route.ts +0 -5
- package/src/app/api/upload/route.ts +0 -5
- package/src/app/constants.tsx +0 -270
- package/src/app/favicon.ico +0 -0
- package/src/app/globals.css +0 -449
- package/src/app/layout.tsx +0 -22
- package/src/app/page.tsx +0 -64
- package/src/app/types.ts +0 -30
- package/src/browser.tsx +0 -139
- package/src/components/ChatWidget.tsx +0 -393
- package/src/components/ChatWindow.tsx +0 -952
- package/src/components/CodeViewer.tsx +0 -56
- package/src/components/ConfigProvider.tsx +0 -123
- package/src/components/DocumentUpload.tsx +0 -354
- package/src/components/DynamicChart.tsx +0 -297
- package/src/components/HourglassLoader.tsx +0 -87
- package/src/components/MarkdownComponents.tsx +0 -186
- package/src/components/MessageBubble.tsx +0 -498
- package/src/components/ObservabilityPanel.tsx +0 -374
- package/src/components/ProductCard.tsx +0 -107
- package/src/components/ProductCarousel.tsx +0 -97
- package/src/components/SourceCard.tsx +0 -61
- package/src/components/ThemeProvider.tsx +0 -8
- package/src/components/ThemeToggle.tsx +0 -29
- package/src/components/ThinkingBlock.tsx +0 -75
- package/src/components/UIDispatcher.tsx +0 -342
- package/src/components/VisualizationRenderer.tsx +0 -621
- package/src/components/constants.tsx +0 -503
- package/src/config/ConfigBuilder.ts +0 -285
- package/src/config/EmbeddingStrategy.ts +0 -148
- package/src/config/LicenseConfigResolver.ts +0 -298
- package/src/config/RagConfig.ts +0 -328
- package/src/config/UniversalProfiles.ts +0 -82
- package/src/config/constants.ts +0 -64
- package/src/config/serverConfig.ts +0 -378
- package/src/config/uiConstants.ts +0 -47
- package/src/core/BatchProcessor.ts +0 -385
- package/src/core/CircuitBreaker.ts +0 -118
- package/src/core/ConfigFetcher.ts +0 -220
- package/src/core/ConfigResolver.ts +0 -165
- package/src/core/ConfigValidator.ts +0 -197
- package/src/core/DatabaseStorage.ts +0 -526
- package/src/core/FreeTierLimitsGuard.ts +0 -281
- package/src/core/LLMRouter.ts +0 -79
- package/src/core/LangChainAgent.ts +0 -132
- package/src/core/LicenseValidator.ts +0 -284
- package/src/core/LicenseVerifier.ts +0 -261
- package/src/core/MultiAgentCoordinator.ts +0 -239
- package/src/core/Pipeline.ts +0 -1279
- package/src/core/PluginManager.ts +0 -277
- package/src/core/ProjectSettings.ts +0 -247
- package/src/core/ProviderHealthCheck.ts +0 -133
- package/src/core/ProviderInterfaces.ts +0 -37
- package/src/core/ProviderRegistry.ts +0 -156
- package/src/core/QueryProcessor.ts +0 -399
- package/src/core/Retrivora.ts +0 -98
- package/src/core/VectorPlugin.ts +0 -193
- package/src/core/mcp.ts +0 -261
- package/src/exceptions/index.ts +0 -131
- package/src/handlers/index.ts +0 -1625
- package/src/hooks/useHydrated.ts +0 -13
- package/src/hooks/useProjectSettings.ts +0 -105
- package/src/hooks/useRagChat.ts +0 -436
- package/src/hooks/useStoredMessages.ts +0 -64
- package/src/index.css +0 -4269
- package/src/index.ts +0 -137
- package/src/lib/plugin.ts +0 -24
- package/src/llm/ILLMProvider.ts +0 -60
- package/src/llm/LLMFactory.ts +0 -141
- package/src/llm/providers/AnthropicProvider.ts +0 -215
- package/src/llm/providers/GeminiProvider.ts +0 -311
- package/src/llm/providers/GroqProvider.ts +0 -184
- package/src/llm/providers/OllamaProvider.ts +0 -232
- package/src/llm/providers/OpenAIProvider.ts +0 -188
- package/src/llm/providers/QwenProvider.ts +0 -199
- package/src/llm/providers/UniversalLLMAdapter.ts +0 -422
- package/src/llm/utils.ts +0 -46
- package/src/providers/graphdb/BaseGraphProvider.ts +0 -43
- package/src/providers/graphdb/SimpleGraphProvider.ts +0 -66
- package/src/providers/vectordb/BaseVectorProvider.ts +0 -98
- package/src/providers/vectordb/ChromaDBProvider.ts +0 -130
- package/src/providers/vectordb/MilvusProvider.ts +0 -134
- package/src/providers/vectordb/MongoDBProvider.ts +0 -250
- package/src/providers/vectordb/MultiTablePostgresProvider.ts +0 -461
- package/src/providers/vectordb/PineconeProvider.ts +0 -295
- package/src/providers/vectordb/PostgreSQLProvider.ts +0 -227
- package/src/providers/vectordb/QdrantProvider.ts +0 -294
- package/src/providers/vectordb/RedisProvider.ts +0 -105
- package/src/providers/vectordb/UniversalVectorProvider.ts +0 -221
- package/src/providers/vectordb/WeaviateProvider.ts +0 -172
- package/src/rag/DocumentChunker.ts +0 -144
- package/src/rag/EntityExtractor.ts +0 -90
- package/src/rag/LlamaIndexIngestor.ts +0 -60
- package/src/rag/Reranker.ts +0 -106
- package/src/rendering/IntentClassifier.ts +0 -167
- package/src/rendering/RendererRegistry.ts +0 -164
- package/src/rendering/RuleEngine.ts +0 -67
- package/src/rendering/VisualizationDecisionEngine.ts +0 -164
- package/src/rendering/__tests__/VisualizationDecisionEngine.test.ts +0 -237
- package/src/rendering/index.ts +0 -12
- package/src/rendering/rules/Rule1SpecificInfoRule.ts +0 -28
- package/src/rendering/rules/Rule2ComparisonRule.ts +0 -24
- package/src/rendering/rules/Rule3ProductDiscoveryRule.ts +0 -28
- package/src/rendering/rules/Rule4AnalyticalRule.ts +0 -38
- package/src/rendering/rules/Rule5MixedResponseRule.ts +0 -29
- package/src/rendering/rules/Rule6SmallResultSetRule.ts +0 -24
- package/src/rendering/rules/Rule7LargeTableRule.ts +0 -35
- package/src/rendering/types.ts +0 -96
- package/src/server.ts +0 -156
- package/src/tailwind.css +0 -194
- package/src/telemetry/TelemetryService.ts +0 -330
- package/src/types/chat.ts +0 -89
- package/src/types/index.ts +0 -278
- package/src/types/pdf-parse.d.ts +0 -13
- package/src/types/props.ts +0 -169
- package/src/utils/DocumentParser.ts +0 -111
- package/src/utils/ProductExtractor.ts +0 -350
- package/src/utils/SchemaMapper.ts +0 -176
- package/src/utils/UITransformer.ts +0 -1812
- package/src/utils/synonyms.ts +0 -80
- package/src/utils/templateUtils.ts +0 -78
- package/src/version.ts +0 -6
|
@@ -1,422 +0,0 @@
|
|
|
1
|
-
import axios, { AxiosInstance } from 'axios';
|
|
2
|
-
import { ILLMProvider } from '../ILLMProvider';
|
|
3
|
-
import { ChatMessage } from '../../types';
|
|
4
|
-
import { LLMConfig, EmbeddingConfig } from '../../config/RagConfig';
|
|
5
|
-
import { resolvePath, buildPayload } from '../../utils/templateUtils';
|
|
6
|
-
import { LLM_PROFILES } from '../../config/UniversalProfiles';
|
|
7
|
-
|
|
8
|
-
interface UniversalLLMOptions {
|
|
9
|
-
baseUrl?: string;
|
|
10
|
-
headers?: Record<string, string>;
|
|
11
|
-
chatPath?: string;
|
|
12
|
-
embedPath?: string;
|
|
13
|
-
pingPath?: string;
|
|
14
|
-
chatPayloadTemplate?: string;
|
|
15
|
-
embedPayloadTemplate?: string;
|
|
16
|
-
responseExtractPath?: string;
|
|
17
|
-
embedExtractPath?: string;
|
|
18
|
-
timeout?: number;
|
|
19
|
-
}
|
|
20
|
-
|
|
21
|
-
export function extractContent(obj: unknown): string | undefined {
|
|
22
|
-
if (!obj || typeof obj !== 'object') return undefined;
|
|
23
|
-
const choice = (obj as any).choices?.[0];
|
|
24
|
-
if (!choice) return undefined;
|
|
25
|
-
if (typeof choice.message?.content === 'string') return choice.message.content;
|
|
26
|
-
if (typeof choice.delta?.content === 'string') return choice.delta.content;
|
|
27
|
-
if (typeof choice.text === 'string') return choice.text;
|
|
28
|
-
if (typeof choice.content === 'string') return choice.content;
|
|
29
|
-
return undefined;
|
|
30
|
-
}
|
|
31
|
-
|
|
32
|
-
export class UniversalLLMAdapter implements ILLMProvider {
|
|
33
|
-
private readonly http: AxiosInstance;
|
|
34
|
-
private readonly model: string;
|
|
35
|
-
private readonly opts: UniversalLLMOptions;
|
|
36
|
-
private readonly systemPrompt: string;
|
|
37
|
-
private readonly maxTokens: number;
|
|
38
|
-
private readonly temperature: number;
|
|
39
|
-
private readonly baseUrl: string;
|
|
40
|
-
private readonly apiKey?: string;
|
|
41
|
-
private readonly resolvedHeaders: Record<string, string>;
|
|
42
|
-
|
|
43
|
-
constructor(config: LLMConfig | EmbeddingConfig) {
|
|
44
|
-
this.model = config.model;
|
|
45
|
-
|
|
46
|
-
const llmConfig = config as LLMConfig;
|
|
47
|
-
const options = (llmConfig.options as Record<string, unknown>) ?? {};
|
|
48
|
-
let profile: Record<string, unknown> = {};
|
|
49
|
-
|
|
50
|
-
if (typeof options.profile === 'string') {
|
|
51
|
-
profile = (LLM_PROFILES as Record<string, Record<string, unknown>>)[options.profile] || {};
|
|
52
|
-
} else if (typeof options.profile === 'object' && options.profile !== null) {
|
|
53
|
-
profile = options.profile as Record<string, unknown>;
|
|
54
|
-
}
|
|
55
|
-
|
|
56
|
-
this.opts = { ...profile, ...(llmConfig.options ?? {}) } as UniversalLLMOptions;
|
|
57
|
-
this.systemPrompt = llmConfig.systemPrompt ?? 'You are a helpful AI assistant. Use the provided context to answer the user.';
|
|
58
|
-
this.maxTokens = llmConfig.maxTokens ?? 1024;
|
|
59
|
-
this.temperature = llmConfig.temperature ?? 0;
|
|
60
|
-
this.apiKey = config.apiKey;
|
|
61
|
-
|
|
62
|
-
this.baseUrl = llmConfig.baseUrl ?? this.opts.baseUrl ?? '';
|
|
63
|
-
if (!this.baseUrl) {
|
|
64
|
-
throw new Error('[UniversalLLMAdapter] baseUrl is required in config or config.options');
|
|
65
|
-
}
|
|
66
|
-
|
|
67
|
-
this.resolvedHeaders = {
|
|
68
|
-
'Content-Type': 'application/json',
|
|
69
|
-
...(config.apiKey ? { Authorization: `Bearer ${config.apiKey}` } : {}),
|
|
70
|
-
...(this.opts.headers || {}),
|
|
71
|
-
};
|
|
72
|
-
|
|
73
|
-
// console.log('[UniversalLLMAdapter] Initialized with:', {
|
|
74
|
-
// model: this.model,
|
|
75
|
-
// baseUrl: this.baseUrl,
|
|
76
|
-
// hasApiKey: !!this.apiKey,
|
|
77
|
-
// apiKeyPrefix: this.apiKey ? `${this.apiKey.slice(0, 12)}...` : undefined,
|
|
78
|
-
// });
|
|
79
|
-
|
|
80
|
-
this.http = axios.create({
|
|
81
|
-
baseURL: this.baseUrl,
|
|
82
|
-
headers: this.resolvedHeaders,
|
|
83
|
-
timeout: this.opts.timeout ?? 60_000,
|
|
84
|
-
});
|
|
85
|
-
}
|
|
86
|
-
|
|
87
|
-
async chat(messages: ChatMessage[], context?: string): Promise<string> {
|
|
88
|
-
const path = this.opts.chatPath ?? '/chat/completions';
|
|
89
|
-
|
|
90
|
-
const safeMessages = (Array.isArray(messages) ? messages : [])
|
|
91
|
-
.filter((m): m is ChatMessage => Boolean(m && typeof m === 'object'))
|
|
92
|
-
.map((m) => ({
|
|
93
|
-
role: m.role || 'user',
|
|
94
|
-
content: typeof m.content === 'string' ? m.content : (m.content != null ? String(m.content) : ''),
|
|
95
|
-
}));
|
|
96
|
-
|
|
97
|
-
const formattedMessages = [
|
|
98
|
-
{ role: 'system', content: `${this.systemPrompt}\n\nContext:\n${context ?? 'None'}` },
|
|
99
|
-
...safeMessages,
|
|
100
|
-
];
|
|
101
|
-
|
|
102
|
-
let payload: unknown;
|
|
103
|
-
if (this.opts.chatPayloadTemplate) {
|
|
104
|
-
payload = buildPayload(this.opts.chatPayloadTemplate, {
|
|
105
|
-
model: this.model,
|
|
106
|
-
messages: formattedMessages,
|
|
107
|
-
maxTokens: this.maxTokens,
|
|
108
|
-
temperature: this.temperature,
|
|
109
|
-
});
|
|
110
|
-
} else {
|
|
111
|
-
payload = {
|
|
112
|
-
model: this.model,
|
|
113
|
-
messages: formattedMessages,
|
|
114
|
-
max_tokens: this.maxTokens,
|
|
115
|
-
temperature: this.temperature,
|
|
116
|
-
};
|
|
117
|
-
}
|
|
118
|
-
|
|
119
|
-
// Tier 1: Remote HTTP POST to Retrivora Cloud Proxy (e.g. https://www.retrivora.com/api/v1/chat/completions)
|
|
120
|
-
try {
|
|
121
|
-
console.debug(`[UniversalLLMAdapter] POST ${this.baseUrl}${path}`);
|
|
122
|
-
console.debug(`[UniversalLLMAdapter] Model: ${this.model}`);
|
|
123
|
-
console.debug(`[UniversalLLMAdapter] Auth: Bearer ${this.apiKey?.substring(0, 20) || 'none'}...`);
|
|
124
|
-
|
|
125
|
-
const { data } = await this.http.post(path, payload);
|
|
126
|
-
const extractPath = this.opts.responseExtractPath ?? 'choices[0].message.content';
|
|
127
|
-
const result = resolvePath(data, extractPath) ?? extractContent(data);
|
|
128
|
-
if (result !== undefined) {
|
|
129
|
-
return String(result);
|
|
130
|
-
}
|
|
131
|
-
} catch (httpErr: any) {
|
|
132
|
-
console.error('[UniversalLLMAdapter] ❌ HTTP chat request failed:', {
|
|
133
|
-
status: httpErr?.response?.status,
|
|
134
|
-
statusText: httpErr?.response?.statusText,
|
|
135
|
-
url: httpErr?.config?.url,
|
|
136
|
-
errorData: httpErr?.response?.data,
|
|
137
|
-
message: httpErr?.message,
|
|
138
|
-
});
|
|
139
|
-
|
|
140
|
-
// Tier 2: In-process fallback (only if HTTP call fails and server registered dispatch on globalThis)
|
|
141
|
-
const _g = globalThis as any;
|
|
142
|
-
const dispatch = _g.__retrivoraDispatchChat;
|
|
143
|
-
|
|
144
|
-
if (typeof dispatch === 'function') {
|
|
145
|
-
console.debug('[UniversalLLMAdapter] 🔄 Attempting in-process fallback dispatch');
|
|
146
|
-
const res = await dispatch({
|
|
147
|
-
model: this.model,
|
|
148
|
-
messages: formattedMessages as any,
|
|
149
|
-
max_tokens: this.maxTokens,
|
|
150
|
-
temperature: this.temperature,
|
|
151
|
-
}, this.apiKey);
|
|
152
|
-
|
|
153
|
-
const content = extractContent(res?.response);
|
|
154
|
-
if (content !== undefined) {
|
|
155
|
-
return content;
|
|
156
|
-
}
|
|
157
|
-
}
|
|
158
|
-
throw httpErr;
|
|
159
|
-
}
|
|
160
|
-
|
|
161
|
-
const { data } = await this.http.post(path, payload);
|
|
162
|
-
const extractPath = this.opts.responseExtractPath ?? 'choices[0].message.content';
|
|
163
|
-
const result = resolvePath(data, extractPath) ?? extractContent(data);
|
|
164
|
-
|
|
165
|
-
if (result === undefined) {
|
|
166
|
-
throw new Error(`[UniversalLLMAdapter] Could not extract text from path '${extractPath}' in response.`);
|
|
167
|
-
}
|
|
168
|
-
|
|
169
|
-
return String(result);
|
|
170
|
-
}
|
|
171
|
-
|
|
172
|
-
/**
|
|
173
|
-
* Streaming chat using native fetch + ReadableStream.
|
|
174
|
-
* Parses OpenAI-compatible SSE frames: `data: {...}\n\n`
|
|
175
|
-
*/
|
|
176
|
-
async * chatStream(messages: ChatMessage[], context?: string): AsyncIterable<string> {
|
|
177
|
-
const path = this.opts.chatPath ?? '/chat/completions';
|
|
178
|
-
const url = `${this.baseUrl.replace(/\/$/, '')}${path}`;
|
|
179
|
-
const extractPath = (this.opts.responseExtractPath ?? 'choices[0].message.content')
|
|
180
|
-
.replace('message.content', 'delta.content');
|
|
181
|
-
|
|
182
|
-
const safeMessages = (Array.isArray(messages) ? messages : [])
|
|
183
|
-
.filter((m): m is ChatMessage => Boolean(m && typeof m === 'object'))
|
|
184
|
-
.map((m) => ({
|
|
185
|
-
role: m.role || 'user',
|
|
186
|
-
content: typeof m.content === 'string' ? m.content : (m.content != null ? String(m.content) : ''),
|
|
187
|
-
}));
|
|
188
|
-
|
|
189
|
-
const formattedMessages = [
|
|
190
|
-
{ role: 'system', content: `${this.systemPrompt}\n\nContext:\n${context ?? 'None'}` },
|
|
191
|
-
...safeMessages,
|
|
192
|
-
];
|
|
193
|
-
|
|
194
|
-
let payload: unknown;
|
|
195
|
-
if (this.opts.chatPayloadTemplate) {
|
|
196
|
-
payload = buildPayload(this.opts.chatPayloadTemplate, {
|
|
197
|
-
model: this.model,
|
|
198
|
-
messages: formattedMessages,
|
|
199
|
-
maxTokens: this.maxTokens,
|
|
200
|
-
temperature: this.temperature,
|
|
201
|
-
});
|
|
202
|
-
if (typeof payload === 'object' && payload !== null) {
|
|
203
|
-
(payload as Record<string, unknown>).stream = true;
|
|
204
|
-
}
|
|
205
|
-
} else {
|
|
206
|
-
payload = {
|
|
207
|
-
model: this.model,
|
|
208
|
-
messages: formattedMessages,
|
|
209
|
-
max_tokens: this.maxTokens,
|
|
210
|
-
temperature: this.temperature,
|
|
211
|
-
stream: true,
|
|
212
|
-
};
|
|
213
|
-
}
|
|
214
|
-
|
|
215
|
-
let streamBody: ReadableStream<Uint8Array> | null = null;
|
|
216
|
-
|
|
217
|
-
// ── Tier 1: HTTP Stream Request to Retrivora Gateway Proxy ─────────────────
|
|
218
|
-
try {
|
|
219
|
-
const response = await fetch(url, {
|
|
220
|
-
method: 'POST',
|
|
221
|
-
headers: this.resolvedHeaders,
|
|
222
|
-
body: typeof payload === 'string' ? payload : JSON.stringify(payload),
|
|
223
|
-
});
|
|
224
|
-
|
|
225
|
-
if (response.ok) {
|
|
226
|
-
const contentType = response.headers.get('content-type') || '';
|
|
227
|
-
if (contentType.includes('application/json')) {
|
|
228
|
-
const json = await response.json();
|
|
229
|
-
const text = (resolvePath(json, extractPath) as string | undefined) ?? extractContent(json);
|
|
230
|
-
if (text && typeof text === 'string') {
|
|
231
|
-
yield text;
|
|
232
|
-
return;
|
|
233
|
-
}
|
|
234
|
-
} else if (response.body) {
|
|
235
|
-
streamBody = response.body;
|
|
236
|
-
}
|
|
237
|
-
} else {
|
|
238
|
-
const errorText = await response.text().catch(() => response.statusText);
|
|
239
|
-
console.warn(`[UniversalLLMAdapter] Remote HTTP stream returned ${response.status}: ${errorText}. Attempting in-process fallback...`);
|
|
240
|
-
}
|
|
241
|
-
} catch (fetchErr: any) {
|
|
242
|
-
console.warn(`[UniversalLLMAdapter] Remote HTTP stream fetch warning: ${fetchErr?.message}. Attempting in-process fallback...`);
|
|
243
|
-
}
|
|
244
|
-
|
|
245
|
-
// ── Tier 2: In-process Dispatch Fallback (if remote HTTP fails) ────────────
|
|
246
|
-
if (!streamBody) {
|
|
247
|
-
const _g = globalThis as any;
|
|
248
|
-
const dispatch = _g.__retrivoraDispatchChat;
|
|
249
|
-
|
|
250
|
-
if (typeof dispatch === 'function') {
|
|
251
|
-
try {
|
|
252
|
-
const res = await dispatch({
|
|
253
|
-
model: this.model,
|
|
254
|
-
messages: formattedMessages as any,
|
|
255
|
-
max_tokens: this.maxTokens,
|
|
256
|
-
temperature: this.temperature,
|
|
257
|
-
stream: true,
|
|
258
|
-
}, this.apiKey);
|
|
259
|
-
|
|
260
|
-
if (res?.stream) {
|
|
261
|
-
streamBody = res.stream as ReadableStream<Uint8Array>;
|
|
262
|
-
} else {
|
|
263
|
-
const content = extractContent(res?.response);
|
|
264
|
-
if (content !== undefined) {
|
|
265
|
-
yield content;
|
|
266
|
-
return;
|
|
267
|
-
}
|
|
268
|
-
}
|
|
269
|
-
} catch (dispatchErr) {
|
|
270
|
-
console.warn('[UniversalLLMAdapter] In-process dispatch error:', dispatchErr);
|
|
271
|
-
}
|
|
272
|
-
}
|
|
273
|
-
}
|
|
274
|
-
|
|
275
|
-
if (!streamBody) {
|
|
276
|
-
throw new Error(`[UniversalLLMAdapter] Streaming request failed. Ensure RETRIVORA_LICENSE_KEY is valid or LLM service is available.`);
|
|
277
|
-
}
|
|
278
|
-
|
|
279
|
-
const reader = streamBody.getReader();
|
|
280
|
-
const decoder = new TextDecoder('utf-8');
|
|
281
|
-
let buffer = '';
|
|
282
|
-
|
|
283
|
-
try {
|
|
284
|
-
while (true) {
|
|
285
|
-
const { done, value } = await reader.read();
|
|
286
|
-
if (done) break;
|
|
287
|
-
|
|
288
|
-
buffer += decoder.decode(value, { stream: true });
|
|
289
|
-
const lines = buffer.split('\n');
|
|
290
|
-
buffer = lines.pop() ?? '';
|
|
291
|
-
|
|
292
|
-
for (const line of lines) {
|
|
293
|
-
const trimmed = line.trim();
|
|
294
|
-
if (!trimmed || trimmed === 'data: [DONE]') continue;
|
|
295
|
-
if (!trimmed.startsWith('data:')) continue;
|
|
296
|
-
|
|
297
|
-
try {
|
|
298
|
-
const json = JSON.parse(trimmed.slice(5).trim());
|
|
299
|
-
const text = (resolvePath(json, extractPath) as string | undefined) ?? extractContent(json);
|
|
300
|
-
if (text && typeof text === 'string') yield text;
|
|
301
|
-
} catch {
|
|
302
|
-
// Skip malformed SSE lines
|
|
303
|
-
}
|
|
304
|
-
}
|
|
305
|
-
}
|
|
306
|
-
|
|
307
|
-
if (buffer.trim() && buffer.trim() !== 'data: [DONE]') {
|
|
308
|
-
const jsonStr = buffer.replace(/^data:\s*/, '').trim();
|
|
309
|
-
try {
|
|
310
|
-
const json = JSON.parse(jsonStr);
|
|
311
|
-
const text = (resolvePath(json, extractPath) as string | undefined) ?? extractContent(json);
|
|
312
|
-
if (text && typeof text === 'string') yield text;
|
|
313
|
-
} catch { /* ignore final partial line */ }
|
|
314
|
-
}
|
|
315
|
-
} finally {
|
|
316
|
-
reader.releaseLock();
|
|
317
|
-
}
|
|
318
|
-
}
|
|
319
|
-
|
|
320
|
-
async embed(text: string): Promise<number[]> {
|
|
321
|
-
const path = this.opts.embedPath ?? '/embeddings';
|
|
322
|
-
const payload = this.opts.embedPayloadTemplate
|
|
323
|
-
? buildPayload(this.opts.embedPayloadTemplate, { input: text, model: this.model })
|
|
324
|
-
: { input: text, model: this.model };
|
|
325
|
-
|
|
326
|
-
// First attempt: try HTTP call with fallback
|
|
327
|
-
try {
|
|
328
|
-
// console.log(`[UniversalLLMAdapter] Posting to embedding endpoint: ${this.baseUrl}${path}`, {
|
|
329
|
-
// model: this.model,
|
|
330
|
-
// inputLength: typeof text === 'string' ? text.length : 'unknown',
|
|
331
|
-
// });
|
|
332
|
-
const { data } = await this.http.post(path, payload);
|
|
333
|
-
const extractPath = this.opts.embedExtractPath ?? 'data[0].embedding';
|
|
334
|
-
const vector = resolvePath(data, extractPath) as number[] | undefined;
|
|
335
|
-
if (Array.isArray(vector)) {
|
|
336
|
-
//console.log(`[UniversalLLMAdapter] Embedding HTTP request successful. Vector size: ${vector.length}`);
|
|
337
|
-
return vector;
|
|
338
|
-
}
|
|
339
|
-
|
|
340
|
-
// If we get here, the API succeeded but the response format was invalid
|
|
341
|
-
console.warn(
|
|
342
|
-
`[UniversalLLMAdapter] Embedding API returned unexpected format for path '${extractPath}'. Response:`,
|
|
343
|
-
data
|
|
344
|
-
);
|
|
345
|
-
} catch (err: any) {
|
|
346
|
-
console.warn(
|
|
347
|
-
`[UniversalLLMAdapter] HTTP embedding request failed: ${err?.message || err}`,
|
|
348
|
-
{ path, model: this.model, response: err?.response?.data }
|
|
349
|
-
);
|
|
350
|
-
|
|
351
|
-
// In-process fallback for embeddings
|
|
352
|
-
const _g = globalThis as any;
|
|
353
|
-
const dispatchEmbed = _g.__retrivoraDispatchEmbedding;
|
|
354
|
-
|
|
355
|
-
if (typeof dispatchEmbed === 'function') {
|
|
356
|
-
try {
|
|
357
|
-
const res = await dispatchEmbed({ input: text, model: this.model }, this.apiKey);
|
|
358
|
-
if (res?.data?.[0]?.embedding) {
|
|
359
|
-
return res.data[0].embedding;
|
|
360
|
-
}
|
|
361
|
-
} catch (dispatchErr: any) {
|
|
362
|
-
console.warn('[UniversalLLMAdapter] In-process dispatch embedding failed:', dispatchErr);
|
|
363
|
-
}
|
|
364
|
-
}
|
|
365
|
-
}
|
|
366
|
-
|
|
367
|
-
// Second attempt: retry the HTTP call
|
|
368
|
-
try {
|
|
369
|
-
//console.log(`[UniversalLLMAdapter] Retrying embedding request to: ${this.baseUrl}${path}`);
|
|
370
|
-
const { data } = await this.http.post(path, payload);
|
|
371
|
-
const extractPath = this.opts.embedExtractPath ?? 'data[0].embedding';
|
|
372
|
-
const vector = resolvePath(data, extractPath) as number[] | undefined;
|
|
373
|
-
|
|
374
|
-
if (Array.isArray(vector)) {
|
|
375
|
-
//console.log(`[UniversalLLMAdapter] Retry embedding HTTP request successful. Vector size: ${vector.length}`);
|
|
376
|
-
return vector;
|
|
377
|
-
}
|
|
378
|
-
|
|
379
|
-
console.error(
|
|
380
|
-
`[UniversalLLMAdapter] Retry embedding failed: unexpected response format for path '${extractPath}'. Response:`,
|
|
381
|
-
data
|
|
382
|
-
);
|
|
383
|
-
throw new Error(
|
|
384
|
-
`[UniversalLLMAdapter] Could not extract embedding vector. ` +
|
|
385
|
-
`Expected array at '${extractPath}', got: ${typeof vector}. ` +
|
|
386
|
-
`Response keys: ${Object.keys(data || {})}.`
|
|
387
|
-
);
|
|
388
|
-
} catch (err: any) {
|
|
389
|
-
console.error(
|
|
390
|
-
`[UniversalLLMAdapter] Both embedding attempts failed. Path: ${path}, Model: ${this.model}`,
|
|
391
|
-
{ error: err?.message, statusCode: err?.response?.status, response: err?.response?.data }
|
|
392
|
-
);
|
|
393
|
-
throw new Error(
|
|
394
|
-
`[UniversalLLMAdapter] Failed to generate embedding after 2 attempts: ${err?.message || err}`
|
|
395
|
-
);
|
|
396
|
-
}
|
|
397
|
-
}
|
|
398
|
-
|
|
399
|
-
async batchEmbed(texts: string[]): Promise<number[][]> {
|
|
400
|
-
const results: number[][] = [];
|
|
401
|
-
for (const text of texts) {
|
|
402
|
-
try {
|
|
403
|
-
const vector = await this.embed(text);
|
|
404
|
-
results.push(vector);
|
|
405
|
-
} catch (err) {
|
|
406
|
-
console.warn(`[UniversalLLMAdapter] batchEmbed failed for text (first 50 chars): ${text.slice(0, 50)}...`, err);
|
|
407
|
-
// Don't push anything - the error will propagate
|
|
408
|
-
throw err;
|
|
409
|
-
}
|
|
410
|
-
}
|
|
411
|
-
return results;
|
|
412
|
-
}
|
|
413
|
-
|
|
414
|
-
async ping(): Promise<boolean> {
|
|
415
|
-
try {
|
|
416
|
-
await this.embed('ping');
|
|
417
|
-
return true;
|
|
418
|
-
} catch {
|
|
419
|
-
return false;
|
|
420
|
-
}
|
|
421
|
-
}
|
|
422
|
-
}
|
package/src/llm/utils.ts
DELETED
|
@@ -1,46 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* LLM utility helpers shared across all provider implementations.
|
|
3
|
-
*
|
|
4
|
-
* These are pure functions with no external dependencies so they can
|
|
5
|
-
* be imported by any provider without creating circular references.
|
|
6
|
-
*/
|
|
7
|
-
|
|
8
|
-
/**
|
|
9
|
-
* Builds the final system message content by injecting the RAG context
|
|
10
|
-
* into a system prompt using one of two strategies:
|
|
11
|
-
*
|
|
12
|
-
* 1. **Placeholder substitution** — if the prompt contains `{{context}}`,
|
|
13
|
-
* that token is replaced with the retrieved context string.
|
|
14
|
-
*
|
|
15
|
-
* 2. **Appendage** — if there is no placeholder, the context is appended
|
|
16
|
-
* after a separator so the LLM always receives the retrieved evidence,
|
|
17
|
-
* regardless of whether the caller supplied a per-call `systemPrompt`
|
|
18
|
-
* override.
|
|
19
|
-
*
|
|
20
|
-
* This function is intentionally strict: it never silently drops context.
|
|
21
|
-
* If `context` is empty or the sentinel "No relevant context found." the
|
|
22
|
-
* prompt is returned as-is so the LLM is not polluted with an empty block.
|
|
23
|
-
*
|
|
24
|
-
* @param systemPrompt - The base system prompt (from config or per-call override).
|
|
25
|
-
* @param context - The retrieved RAG context string.
|
|
26
|
-
* @returns - The resolved system message content ready to send.
|
|
27
|
-
*/
|
|
28
|
-
export function buildSystemContent(systemPrompt: string, context: string): string {
|
|
29
|
-
const noContext =
|
|
30
|
-
!context ||
|
|
31
|
-
context.trim() === '' ||
|
|
32
|
-
context.trim() === 'No relevant context found.';
|
|
33
|
-
|
|
34
|
-
// Replace {{context}} placeholder regardless of whether context is empty
|
|
35
|
-
// so the prompt is never sent with a raw placeholder token.
|
|
36
|
-
if (systemPrompt.includes('{{context}}')) {
|
|
37
|
-
return systemPrompt.replace('{{context}}', noContext ? '' : context);
|
|
38
|
-
}
|
|
39
|
-
|
|
40
|
-
// Without a placeholder: only append context when there is actual content.
|
|
41
|
-
if (noContext) {
|
|
42
|
-
return systemPrompt;
|
|
43
|
-
}
|
|
44
|
-
|
|
45
|
-
return `${systemPrompt}\n\n---\nContext:\n${context}`;
|
|
46
|
-
}
|
|
@@ -1,43 +0,0 @@
|
|
|
1
|
-
import { GraphDBConfig } from '../../config/RagConfig';
|
|
2
|
-
import { GraphNode, Edge, GraphSearchResult } from '../../types';
|
|
3
|
-
|
|
4
|
-
/**
|
|
5
|
-
* BaseGraphProvider — Abstract base class for Graph DB providers.
|
|
6
|
-
*/
|
|
7
|
-
export abstract class BaseGraphProvider {
|
|
8
|
-
protected config: GraphDBConfig;
|
|
9
|
-
|
|
10
|
-
constructor(config: GraphDBConfig) {
|
|
11
|
-
this.config = config;
|
|
12
|
-
}
|
|
13
|
-
|
|
14
|
-
/**
|
|
15
|
-
* Initialise connection to the graph database.
|
|
16
|
-
*/
|
|
17
|
-
abstract initialize(): Promise<void>;
|
|
18
|
-
|
|
19
|
-
/**
|
|
20
|
-
* Add nodes to the graph.
|
|
21
|
-
*/
|
|
22
|
-
abstract addNodes(nodes: GraphNode[]): Promise<void>;
|
|
23
|
-
|
|
24
|
-
/**
|
|
25
|
-
* Add edges (relationships) to the graph.
|
|
26
|
-
*/
|
|
27
|
-
abstract addEdges(edges: Edge[]): Promise<void>;
|
|
28
|
-
|
|
29
|
-
/**
|
|
30
|
-
* Query the graph for relevant nodes and edges.
|
|
31
|
-
*/
|
|
32
|
-
abstract query(queryText: string, limit?: number): Promise<GraphSearchResult>;
|
|
33
|
-
|
|
34
|
-
/**
|
|
35
|
-
* Check if the database is reachable.
|
|
36
|
-
*/
|
|
37
|
-
abstract ping(): Promise<boolean>;
|
|
38
|
-
|
|
39
|
-
/**
|
|
40
|
-
* Close connection.
|
|
41
|
-
*/
|
|
42
|
-
abstract disconnect(): Promise<void>;
|
|
43
|
-
}
|
|
@@ -1,66 +0,0 @@
|
|
|
1
|
-
import { BaseGraphProvider } from './BaseGraphProvider';
|
|
2
|
-
import { GraphNode, Edge, GraphSearchResult } from '../../types';
|
|
3
|
-
|
|
4
|
-
/**
|
|
5
|
-
* SimpleGraphProvider — In-memory graph storage for demonstration and local dev.
|
|
6
|
-
*/
|
|
7
|
-
export class SimpleGraphProvider extends BaseGraphProvider {
|
|
8
|
-
private nodes: Map<string, GraphNode> = new Map();
|
|
9
|
-
private edges: Edge[] = [];
|
|
10
|
-
|
|
11
|
-
async initialize(): Promise<void> {
|
|
12
|
-
//console.log('[SimpleGraphProvider] Initialised in-memory graph store.');
|
|
13
|
-
}
|
|
14
|
-
|
|
15
|
-
async addNodes(nodes: GraphNode[]): Promise<void> {
|
|
16
|
-
for (const node of nodes) {
|
|
17
|
-
this.nodes.set(node.id, node);
|
|
18
|
-
}
|
|
19
|
-
}
|
|
20
|
-
|
|
21
|
-
async addEdges(edges: Edge[]): Promise<void> {
|
|
22
|
-
this.edges.push(...edges);
|
|
23
|
-
}
|
|
24
|
-
|
|
25
|
-
async query(queryText: string, limit = 5): Promise<GraphSearchResult> {
|
|
26
|
-
const q = queryText.toLowerCase();
|
|
27
|
-
|
|
28
|
-
// Simple keyword matching for demonstration
|
|
29
|
-
const matchedNodes = Array.from(this.nodes.values()).filter(node =>
|
|
30
|
-
node.id.toLowerCase().includes(q) ||
|
|
31
|
-
node.label.toLowerCase().includes(q) ||
|
|
32
|
-
JSON.stringify(node.properties).toLowerCase().includes(q)
|
|
33
|
-
).slice(0, limit);
|
|
34
|
-
|
|
35
|
-
const matchedNodeIds = new Set(matchedNodes.map(n => n.id));
|
|
36
|
-
const matchedEdges = this.edges.filter(edge =>
|
|
37
|
-
matchedNodeIds.has(edge.source) || matchedNodeIds.has(edge.target)
|
|
38
|
-
);
|
|
39
|
-
|
|
40
|
-
// Also include connected nodes
|
|
41
|
-
for (const edge of matchedEdges) {
|
|
42
|
-
if (!matchedNodeIds.has(edge.source)) {
|
|
43
|
-
const source = this.nodes.get(edge.source);
|
|
44
|
-
if (source) matchedNodes.push(source);
|
|
45
|
-
}
|
|
46
|
-
if (!matchedNodeIds.has(edge.target)) {
|
|
47
|
-
const target = this.nodes.get(edge.target);
|
|
48
|
-
if (target) matchedNodes.push(target);
|
|
49
|
-
}
|
|
50
|
-
}
|
|
51
|
-
|
|
52
|
-
return {
|
|
53
|
-
nodes: Array.from(new Set(matchedNodes)),
|
|
54
|
-
edges: matchedEdges
|
|
55
|
-
};
|
|
56
|
-
}
|
|
57
|
-
|
|
58
|
-
async ping(): Promise<boolean> {
|
|
59
|
-
return true;
|
|
60
|
-
}
|
|
61
|
-
|
|
62
|
-
async disconnect(): Promise<void> {
|
|
63
|
-
this.nodes.clear();
|
|
64
|
-
this.edges = [];
|
|
65
|
-
}
|
|
66
|
-
}
|
|
@@ -1,98 +0,0 @@
|
|
|
1
|
-
import { VectorDBConfig } from '../../config/RagConfig';
|
|
2
|
-
import { VectorMatch, UpsertDocument } from '../../types';
|
|
3
|
-
import { IProviderValidator, IProviderHealthChecker } from '../../core/ProviderInterfaces';
|
|
4
|
-
|
|
5
|
-
/**
|
|
6
|
-
* BaseVectorProvider — Abstract base class for all vector database providers.
|
|
7
|
-
* Each provider (Pinecone, Milvus, Redis, etc.) must extend this class.
|
|
8
|
-
*/
|
|
9
|
-
export abstract class BaseVectorProvider {
|
|
10
|
-
protected readonly config: VectorDBConfig;
|
|
11
|
-
protected indexName: string;
|
|
12
|
-
protected projectId?: string;
|
|
13
|
-
|
|
14
|
-
constructor(config: VectorDBConfig) {
|
|
15
|
-
this.config = config;
|
|
16
|
-
this.indexName = config.indexName || 'default';
|
|
17
|
-
}
|
|
18
|
-
|
|
19
|
-
/**
|
|
20
|
-
* Initialise the connection (create tables, verify index, etc.)
|
|
21
|
-
*/
|
|
22
|
-
abstract initialize(): Promise<void>;
|
|
23
|
-
|
|
24
|
-
/**
|
|
25
|
-
* Insert or update a single vector + content pair.
|
|
26
|
-
*/
|
|
27
|
-
abstract upsert(doc: UpsertDocument, namespace?: string): Promise<void>;
|
|
28
|
-
|
|
29
|
-
/**
|
|
30
|
-
* Batch upsert for efficient ingestion of many documents.
|
|
31
|
-
*/
|
|
32
|
-
abstract batchUpsert(docs: UpsertDocument[], namespace?: string): Promise<void>;
|
|
33
|
-
|
|
34
|
-
/**
|
|
35
|
-
* Find the top-K most similar vectors to the query vector.
|
|
36
|
-
*/
|
|
37
|
-
abstract query(
|
|
38
|
-
vector: number[],
|
|
39
|
-
topK: number,
|
|
40
|
-
namespace?: string,
|
|
41
|
-
filter?: Record<string, unknown>
|
|
42
|
-
): Promise<VectorMatch[]>;
|
|
43
|
-
|
|
44
|
-
/**
|
|
45
|
-
* Delete a stored vector by ID.
|
|
46
|
-
*/
|
|
47
|
-
abstract delete(id: string | number, namespace?: string): Promise<void>;
|
|
48
|
-
|
|
49
|
-
/**
|
|
50
|
-
* Delete all vectors in a namespace (full data reset for a project).
|
|
51
|
-
*/
|
|
52
|
-
abstract deleteNamespace(namespace: string): Promise<void>;
|
|
53
|
-
|
|
54
|
-
/**
|
|
55
|
-
* Check if the underlying DB is reachable.
|
|
56
|
-
*/
|
|
57
|
-
abstract ping(): Promise<boolean>;
|
|
58
|
-
|
|
59
|
-
/**
|
|
60
|
-
* Gracefully close the connection.
|
|
61
|
-
*/
|
|
62
|
-
abstract disconnect(): Promise<void>;
|
|
63
|
-
|
|
64
|
-
/**
|
|
65
|
-
* Returns a validator for this provider's configuration.
|
|
66
|
-
*/
|
|
67
|
-
static getValidator?(): IProviderValidator;
|
|
68
|
-
|
|
69
|
-
/**
|
|
70
|
-
* Returns a health checker for this provider.
|
|
71
|
-
*/
|
|
72
|
-
static getHealthChecker?(): IProviderHealthChecker;
|
|
73
|
-
|
|
74
|
-
/**
|
|
75
|
-
* Remove internal keys (starting with __) and queryText from a filter object
|
|
76
|
-
* before passing it to an external provider that might not support them.
|
|
77
|
-
*/
|
|
78
|
-
protected sanitizeFilter(filter?: Record<string, unknown>): Record<string, unknown> {
|
|
79
|
-
if (!filter) return {};
|
|
80
|
-
|
|
81
|
-
const result: Record<string, unknown> = {};
|
|
82
|
-
|
|
83
|
-
for (const [key, value] of Object.entries(filter)) {
|
|
84
|
-
if (key.startsWith('__') || key === 'queryText' || key === 'keywords') {
|
|
85
|
-
continue;
|
|
86
|
-
}
|
|
87
|
-
|
|
88
|
-
if (key === 'metadata' && typeof value === 'object' && value !== null) {
|
|
89
|
-
// Merge metadata fields into the root for providers that expect flat filters
|
|
90
|
-
Object.assign(result, value);
|
|
91
|
-
} else {
|
|
92
|
-
result[key] = value;
|
|
93
|
-
}
|
|
94
|
-
}
|
|
95
|
-
|
|
96
|
-
return result;
|
|
97
|
-
}
|
|
98
|
-
}
|