100 lines
2.9 KiB
JavaScript
100 lines
2.9 KiB
JavaScript
import { NextResponse } from 'next/server';
|
|
import LLMClient from '@/lib/llm/core/index';
|
|
import { getModelConfigById } from '@/lib/db/model-config';
|
|
|
|
async function resolveLatestModelConfig(projectId, incomingModel = {}) {
|
|
const modelId = incomingModel?.id;
|
|
if (!modelId) {
|
|
return incomingModel;
|
|
}
|
|
|
|
try {
|
|
const latestModelConfig = await getModelConfigById(modelId);
|
|
if (!latestModelConfig) {
|
|
return incomingModel;
|
|
}
|
|
if (String(latestModelConfig.projectId) !== String(projectId)) {
|
|
return incomingModel;
|
|
}
|
|
|
|
// Keep transient client-only fields, but force endpoint/auth/model fields to latest DB values.
|
|
return {
|
|
...incomingModel,
|
|
...latestModelConfig
|
|
};
|
|
} catch (error) {
|
|
console.error('Failed to resolve latest model config:', String(error));
|
|
return incomingModel;
|
|
}
|
|
}
|
|
|
|
export async function POST(request, { params }) {
|
|
try {
|
|
const { projectId } = params;
|
|
|
|
// Validate project ID.
|
|
if (!projectId) {
|
|
return NextResponse.json({ error: 'The project ID cannot be empty' }, { status: 400 });
|
|
}
|
|
|
|
// Read request payload.
|
|
const { model, messages } = await request.json();
|
|
const resolvedModel = await resolveLatestModelConfig(projectId, model);
|
|
|
|
// Validate request parameters.
|
|
if (!resolvedModel) {
|
|
return NextResponse.json({ error: 'The model parameters cannot be empty' }, { status: 400 });
|
|
}
|
|
|
|
if (!Array.isArray(messages) || messages.length === 0) {
|
|
return NextResponse.json({ error: 'The message list cannot be empty' }, { status: 400 });
|
|
}
|
|
|
|
// Use custom LLM client.
|
|
const llmClient = new LLMClient(resolvedModel);
|
|
|
|
// Normalize message payload for text + vision models.
|
|
const formattedMessages = messages.map(msg => {
|
|
// Plain text message.
|
|
if (typeof msg.content === 'string') {
|
|
return {
|
|
role: msg.role,
|
|
content: msg.content
|
|
};
|
|
}
|
|
// Multimodal message (e.g. image parts).
|
|
if (Array.isArray(msg.content)) {
|
|
return {
|
|
role: msg.role,
|
|
content: msg.content
|
|
};
|
|
}
|
|
// Fallback.
|
|
return {
|
|
role: msg.role,
|
|
content: msg.content
|
|
};
|
|
});
|
|
|
|
// Call LLM API.
|
|
let response = '';
|
|
try {
|
|
const { answer, cot } = await llmClient.getResponseWithCOT(formattedMessages.filter(f => f.role !== 'error'));
|
|
response = `<think>${cot}</think>${answer}`;
|
|
} catch (error) {
|
|
console.error('Failed to call LLM API:', String(error));
|
|
return NextResponse.json(
|
|
{
|
|
error: `Failed to call ${resolvedModel.modelId || resolvedModel.modelName || 'unknown'} model: ${error.message}`
|
|
},
|
|
{ status: 500 }
|
|
);
|
|
}
|
|
|
|
return NextResponse.json({ response });
|
|
} catch (error) {
|
|
console.error('Failed to process chat request:', String(error));
|
|
return NextResponse.json({ error: `Failed to process chat request: ${error.message}` }, { status: 500 });
|
|
}
|
|
}
|