apunkt/mike
1
0
Fork 0
mirror of https://github.com/willchen96/mike.git synced 2026-06-30 21:59:37 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

refactor: enhance error handling and streamline API key management in LLM modules

Browse source
This commit is contained in:
willchen96 2026-05-14 23:20:28 +08:00
parent f39f175273
commit a2368a7479
4 changed files with 135 additions and 114 deletions
Download patch file Download diff file Expand all files Collapse all files

14
backend/src/lib/llm/claude.ts
View file

@ -20,9 +20,19 @@ type NativeMessage = {
const MAX_TOKENS = 16384;
function apiKey(override?: string | null): string {
const key = override?.trim() || process.env.ANTHROPIC_API_KEY?.trim() || "";
if (!key) {
throw new Error(
"Anthropic API key is not configured. Set ANTHROPIC_API_KEY or add a user Anthropic key.",
);
}
return key;
}
function client(override?: string | null): Anthropic {
const apiKey = override?.trim() || process.env.ANTHROPIC_API_KEY || "";
return new Anthropic({ apiKey });
const apiKeyValue = apiKey(override);
return new Anthropic({ apiKey: apiKeyValue });
}
function toNativeMessages(

98
backend/src/lib/llm/gemini.ts
View file

@ -28,9 +28,6 @@ type GeminiContent = {
parts: GeminiPart[];
};
const RETRYABLE_STATUSES = new Set([429, 500, 502, 503, 504]);
const MAX_GEMINI_ATTEMPTS = 3;
function apiKey(override?: string | null): string {
const key = override?.trim() || process.env.GEMINI_API_KEY?.trim() || "";
if (!key) {
@ -45,49 +42,6 @@ function client(override?: string | null): GoogleGenAI {
return new GoogleGenAI({ apiKey: apiKey(override) });
}
function geminiStatus(err: unknown): number | null {
const status = (err as { status?: unknown })?.status;
return typeof status === "number" ? status : null;
}
function isRetryableGeminiError(err: unknown): boolean {
const status = geminiStatus(err);
if (status != null && RETRYABLE_STATUSES.has(status)) return true;
const message =
err instanceof Error ? err.message : typeof err === "string" ? err : "";
return /UNAVAILABLE|Service Unavailable|high demand|try again later/i.test(
message,
);
}
function retryDelayMs(attempt: number): number {
return 400 * 2 ** attempt;
}
async function sleep(ms: number): Promise<void> {
await new Promise((resolve) => setTimeout(resolve, ms));
}
async function withGeminiRetries<T>(operation: () => Promise<T>): Promise<T> {
let lastError: unknown;
for (let attempt = 0; attempt < MAX_GEMINI_ATTEMPTS; attempt++) {
try {
return await operation();
} catch (err) {
lastError = err;
const isLastAttempt = attempt === MAX_GEMINI_ATTEMPTS - 1;
if (isLastAttempt || !isRetryableGeminiError(err)) throw err;
console.warn("[gemini] transient error; retrying", {
attempt: attempt + 1,
status: geminiStatus(err),
});
await sleep(retryDelayMs(attempt));
}
}
throw lastError;
}
function toNativeContents(messages: StreamChatParams["messages"]): GeminiContent[] {
return messages.map((m) => ({
role: m.role === "assistant" ? "model" : "user",
@ -107,25 +61,23 @@ export async function streamGemini(
let fullText = "";
for (let iter = 0; iter < maxIter; iter++) {
const stream = await withGeminiRetries(() =>
ai.models.generateContentStream({
model,
contents: contents as never,
config: {
systemInstruction: systemPrompt,
tools: functionDeclarations.length
? [{ functionDeclarations } as never]
: undefined,
// When enabled, ask Gemini to surface thought summaries.
// When disabled, explicitly zero the thinking budget so the
// model skips thinking entirely (saves tokens and latency
// for bulk extraction jobs).
thinkingConfig: enableThinking
? { includeThoughts: true }
: { thinkingBudget: 0 },
},
}),
);
const stream = await ai.models.generateContentStream({
model,
contents: contents as never,
config: {
systemInstruction: systemPrompt,
tools: functionDeclarations.length
? [{ functionDeclarations } as never]
: undefined,
// When enabled, ask Gemini to surface thought summaries.
// When disabled, explicitly zero the thinking budget so the
// model skips thinking entirely (saves tokens and latency
// for bulk extraction jobs).
thinkingConfig: enableThinking
? { includeThoughts: true }
: { thinkingBudget: 0 },
},
});
// Per-iteration accumulators.
const textParts: string[] = [];
@ -207,14 +159,12 @@ export async function completeGeminiText(params: {
apiKeys?: { gemini?: string | null };
}): Promise<string> {
const ai = client(params.apiKeys?.gemini);
const resp = await withGeminiRetries(() =>
ai.models.generateContent({
model: params.model,
contents: [{ role: "user", parts: [{ text: params.user }] }],
config: params.systemPrompt
? { systemInstruction: params.systemPrompt }
: undefined,
}),
);
const resp = await ai.models.generateContent({
model: params.model,
contents: [{ role: "user", parts: [{ text: params.user }] }],
config: params.systemPrompt
? { systemInstruction: params.systemPrompt }
: undefined,
});
return resp.text ?? "";
}

12
backend/src/lib/llm/openai.ts
View file

@ -36,7 +36,13 @@ type ResponseStreamEvent = {
};
function apiKey(override?: string | null): string {
return override?.trim() || process.env.OPENAI_API_KEY?.trim() || "";
const key = override?.trim() || process.env.OPENAI_API_KEY?.trim() || "";
if (!key) {
throw new Error(
"OpenAI API key is not configured. Set OPENAI_API_KEY or add a user OpenAI key.",
);
}
return key;
}
function toResponseTools(tools: OpenAIToolSchema[]): ResponseFunctionTool[] {
@ -131,9 +137,11 @@ async function createResponse(params: {
if (!response.ok) {
const text = await response.text().catch(() => "");
throw new Error(
const err = new Error(
`OpenAI request failed (${response.status}): ${text || response.statusText}`,
);
(err as { status?: number }).status = response.status;
throw err;
}
return response;