refactor: make AI runtime z.ai-only and default to glm-5
This commit is contained in:
216
lib/server/ai.ts
216
lib/server/ai.ts
@@ -1,9 +1,8 @@
|
||||
import { createOpenAI } from '@ai-sdk/openai';
|
||||
import { generateText } from 'ai';
|
||||
import { createZhipu } from 'zhipu-ai-provider';
|
||||
|
||||
type AiWorkload = 'report' | 'extraction';
|
||||
type AiProvider = 'zhipu' | 'ollama';
|
||||
type AiProvider = 'zhipu';
|
||||
|
||||
type AiConfig = {
|
||||
provider: AiProvider;
|
||||
@@ -39,9 +38,6 @@ type RunAiAnalysisOptions = GetAiConfigOptions & {
|
||||
};
|
||||
|
||||
const CODING_API_BASE_URL = 'https://api.z.ai/api/coding/paas/v4';
|
||||
const OLLAMA_BASE_URL = 'http://127.0.0.1:11434';
|
||||
const OLLAMA_MODEL = 'qwen3:8b';
|
||||
const OLLAMA_API_KEY = 'ollama';
|
||||
|
||||
let warnedIgnoredZhipuBaseUrl = false;
|
||||
|
||||
@@ -80,128 +76,13 @@ function warnIgnoredZhipuBaseUrl(env: EnvSource, warn: (message: string) => void
|
||||
);
|
||||
}
|
||||
|
||||
function fallbackResponse(prompt: string) {
|
||||
const clipped = prompt.split('\n').slice(0, 6).join(' ').slice(0, 260);
|
||||
|
||||
return [
|
||||
'AI SDK fallback mode is active (live model configuration is missing or unavailable).',
|
||||
'Thesis: Portfolio remains analyzable with local heuristics until live model access is configured.',
|
||||
'Risk scan: Concentration and filing sentiment should be monitored after each sync cycle.',
|
||||
`Context digest: ${clipped}`
|
||||
].join('\n\n');
|
||||
}
|
||||
|
||||
function toOpenAiCompatibleBaseUrl(baseUrl: string) {
|
||||
const normalized = baseUrl.endsWith('/')
|
||||
? baseUrl.slice(0, -1)
|
||||
: baseUrl;
|
||||
|
||||
return normalized.endsWith('/v1')
|
||||
? normalized
|
||||
: `${normalized}/v1`;
|
||||
}
|
||||
|
||||
function asErrorMessage(error: unknown) {
|
||||
if (error instanceof Error && error.message) {
|
||||
return error.message;
|
||||
}
|
||||
|
||||
return String(error);
|
||||
}
|
||||
|
||||
function errorSearchText(error: unknown) {
|
||||
const chunks: string[] = [];
|
||||
const seen = new Set<unknown>();
|
||||
|
||||
const visit = (value: unknown) => {
|
||||
if (value === null || value === undefined) {
|
||||
return;
|
||||
}
|
||||
|
||||
if (typeof value === 'string') {
|
||||
const normalized = value.trim();
|
||||
if (normalized.length > 0) {
|
||||
chunks.push(normalized);
|
||||
}
|
||||
|
||||
return;
|
||||
}
|
||||
|
||||
if (typeof value !== 'object') {
|
||||
chunks.push(String(value));
|
||||
return;
|
||||
}
|
||||
|
||||
if (seen.has(value)) {
|
||||
return;
|
||||
}
|
||||
seen.add(value);
|
||||
|
||||
if (value instanceof Error) {
|
||||
if (value.message) {
|
||||
chunks.push(value.message);
|
||||
}
|
||||
|
||||
const withCause = value as Error & { cause?: unknown };
|
||||
if (withCause.cause !== undefined) {
|
||||
visit(withCause.cause);
|
||||
}
|
||||
return;
|
||||
}
|
||||
|
||||
const record = value as Record<string, unknown>;
|
||||
visit(record.message);
|
||||
visit(record.error);
|
||||
visit(record.reason);
|
||||
visit(record.detail);
|
||||
visit(record.details);
|
||||
visit(record.cause);
|
||||
};
|
||||
|
||||
visit(error);
|
||||
return chunks.join('\n');
|
||||
}
|
||||
|
||||
const REPORT_FALLBACK_ERROR_PATTERNS: RegExp[] = [
|
||||
/insufficient balance/i,
|
||||
/no resource package/i,
|
||||
/insufficient quota/i,
|
||||
/quota exceeded/i,
|
||||
/insufficient credit/i,
|
||||
/invalid api key/i,
|
||||
/authentication/i,
|
||||
/unauthorized/i,
|
||||
/forbidden/i,
|
||||
/payment required/i,
|
||||
/recharge/i,
|
||||
/unable to connect/i,
|
||||
/network/i,
|
||||
/timeout/i,
|
||||
/timed out/i,
|
||||
/econnrefused/i
|
||||
];
|
||||
|
||||
function shouldFallbackReportError(error: unknown) {
|
||||
const searchText = errorSearchText(error) || asErrorMessage(error);
|
||||
return REPORT_FALLBACK_ERROR_PATTERNS.some((pattern) => pattern.test(searchText));
|
||||
}
|
||||
|
||||
function defaultCreateModel(config: AiConfig) {
|
||||
if (config.provider === 'zhipu') {
|
||||
const zhipu = createZhipu({
|
||||
apiKey: config.apiKey,
|
||||
baseURL: config.baseUrl
|
||||
});
|
||||
|
||||
return zhipu(config.model);
|
||||
}
|
||||
|
||||
const openai = createOpenAI({
|
||||
apiKey: config.apiKey ?? OLLAMA_API_KEY,
|
||||
baseURL: toOpenAiCompatibleBaseUrl(config.baseUrl)
|
||||
const zhipu = createZhipu({
|
||||
apiKey: config.apiKey,
|
||||
baseURL: config.baseUrl
|
||||
});
|
||||
|
||||
return openai.chat(config.model);
|
||||
return zhipu(config.model);
|
||||
}
|
||||
|
||||
async function defaultGenerate(input: AiGenerateInput): Promise<AiGenerateOutput> {
|
||||
@@ -228,21 +109,16 @@ export function getReportAiConfig(options?: GetAiConfigOptions) {
|
||||
provider: 'zhipu',
|
||||
apiKey: envValue('ZHIPU_API_KEY', env),
|
||||
baseUrl: CODING_API_BASE_URL,
|
||||
model: envValue('ZHIPU_MODEL', env) ?? 'glm-4.7-flashx',
|
||||
model: envValue('ZHIPU_MODEL', env) ?? 'glm-5',
|
||||
temperature: parseTemperature(envValue('AI_TEMPERATURE', env))
|
||||
} satisfies AiConfig;
|
||||
}
|
||||
|
||||
export function getExtractionAiConfig(options?: GetAiConfigOptions) {
|
||||
const env = options?.env ?? process.env;
|
||||
|
||||
return {
|
||||
provider: 'ollama',
|
||||
apiKey: envValue('OLLAMA_API_KEY', env) ?? OLLAMA_API_KEY,
|
||||
baseUrl: envValue('OLLAMA_BASE_URL', env) ?? OLLAMA_BASE_URL,
|
||||
model: envValue('OLLAMA_MODEL', env) ?? OLLAMA_MODEL,
|
||||
...getReportAiConfig(options),
|
||||
temperature: 0
|
||||
} satisfies AiConfig;
|
||||
};
|
||||
}
|
||||
|
||||
export function isAiConfigured(options?: GetAiConfigOptions) {
|
||||
@@ -256,70 +132,32 @@ export async function runAiAnalysis(prompt: string, systemPrompt?: string, optio
|
||||
? getExtractionAiConfig(options)
|
||||
: getReportAiConfig(options);
|
||||
|
||||
if (workload === 'report' && !config.apiKey) {
|
||||
return {
|
||||
provider: 'local-fallback',
|
||||
model: config.model,
|
||||
text: fallbackResponse(prompt)
|
||||
};
|
||||
if (!config.apiKey) {
|
||||
throw new Error('ZHIPU_API_KEY is required for AI workloads');
|
||||
}
|
||||
|
||||
const createModel = options?.createModel ?? defaultCreateModel;
|
||||
const generate = options?.generate ?? defaultGenerate;
|
||||
const warn = options?.warn ?? console.warn;
|
||||
const model = createModel(config);
|
||||
|
||||
try {
|
||||
const model = createModel(config);
|
||||
const result = await generate({
|
||||
model,
|
||||
system: systemPrompt,
|
||||
prompt,
|
||||
temperature: config.temperature,
|
||||
maxRetries: 0
|
||||
});
|
||||
|
||||
const result = await generate({
|
||||
model,
|
||||
system: systemPrompt,
|
||||
prompt,
|
||||
temperature: config.temperature,
|
||||
maxRetries: 0
|
||||
});
|
||||
|
||||
const text = result.text.trim();
|
||||
if (!text) {
|
||||
if (workload === 'extraction') {
|
||||
return {
|
||||
provider: 'local-fallback',
|
||||
model: config.model,
|
||||
text: fallbackResponse(prompt)
|
||||
};
|
||||
}
|
||||
|
||||
throw new Error('AI SDK returned an empty response');
|
||||
}
|
||||
|
||||
return {
|
||||
provider: config.provider,
|
||||
model: config.model,
|
||||
text
|
||||
};
|
||||
} catch (error) {
|
||||
if (workload === 'report' && shouldFallbackReportError(error)) {
|
||||
warn(`[AI SDK] Report fallback activated: ${asErrorMessage(error)}`);
|
||||
|
||||
return {
|
||||
provider: 'local-fallback',
|
||||
model: config.model,
|
||||
text: fallbackResponse(prompt)
|
||||
};
|
||||
}
|
||||
|
||||
if (workload === 'extraction') {
|
||||
warn(`[AI SDK] Extraction fallback activated: ${asErrorMessage(error)}`);
|
||||
|
||||
return {
|
||||
provider: 'local-fallback',
|
||||
model: config.model,
|
||||
text: fallbackResponse(prompt)
|
||||
};
|
||||
}
|
||||
|
||||
throw error;
|
||||
const text = result.text.trim();
|
||||
if (!text) {
|
||||
throw new Error('AI SDK returned an empty response');
|
||||
}
|
||||
|
||||
return {
|
||||
provider: config.provider,
|
||||
model: config.model,
|
||||
text
|
||||
};
|
||||
}
|
||||
|
||||
export function __resetAiWarningsForTests() {
|
||||
|
||||
Reference in New Issue
Block a user