Consider it done. The open-source AI agent that works out of the box · 想到,就能做到。开源、开箱即用的 AI Agent。
TypeScript agentai-agentai-assistantandroidclaude-codecodexelectroniosllmmacosreact-nativetypescriptwindows
import type {
CustomProviderConfig,
CustomProviderRuntimeConfig,
ProviderRuntimeModelConfig,
} from '@cindy/model-providers';
import {
LOCAL_ADVANCED_PRESET_IDS,
MANAGED_LMSTUDIO_PROVIDER_ID,
} from './localModelRuntime.js';
function loopbackOrigin(baseUrl: string): string | null {
try {
const url = new URL(baseUrl);
if (url.protocol !== 'http:' && url.protocol !== 'https:') return null;
if (url.hostname !== '127.0.0.1' && url.hostname !== 'localhost') return null;
return `${url.protocol}//${url.host}`;
} catch {
return null;
}
}
function copyModels(models: readonly ProviderRuntimeModelConfig[] = []): ProviderRuntimeModelConfig[] {
return models.map((model) => ({
id: model.id,
name: model.name,
...(model.contextWindow ? { contextWindow: model.contextWindow } : {}),
...(model.supportsImageInput ? { supportsImageInput: true } : {}),
}));
}
function isPlainChatRuntime(
runtime: CustomProviderRuntimeConfig | undefined,
): runtime is CustomProviderRuntimeConfig {
if (!runtime?.baseUrl) return false;
if (runtime.wireProtocol !== undefined && runtime.wireProtocol !== 'openai-chat') return false;
if (runtime.headers && Object.keys(runtime.headers).length > 0) return false;
if (runtime.modelsUrl || runtime.requestPath || runtime.piCatalogProviderId) return false;
return loopbackOrigin(runtime.baseUrl) !== null;
}
/**
* 旧本机预设只有 Pi。把未改过指纹的 LM Studio / llama.cpp / vLLM
* 补上现在预设里的 harness,不改用户手改过的供应商。
*/
export function migrateLocalConnectProvider(
existing: CustomProviderConfig,
): CustomProviderConfig | null {
if ((existing.auth?.method ?? 'none') !== 'none') return null;
const runtimeKeys = Object.keys(existing.runtimes);
if (runtimeKeys.some((key) => key !== 'pi' && key !== 'claude-code' && key !== 'codex')) {
return null;
}
const pi = existing.runtimes.pi;
if (!isPlainChatRuntime(pi)) return null;
const isLmStudio = existing.id === MANAGED_LMSTUDIO_PROVIDER_ID;
const isKnownPreset =
isLmStudio ||
(LOCAL_ADVANCED_PRESET_IDS as readonly string[]).includes(existing.id) ||
/^(llama-cpp|vllm)-\d+$/.test(existing.id);
if (!isKnownPreset) return null;
const models = copyModels(pi.models);
const origin = loopbackOrigin(pi.baseUrl);
if (!origin) return null;
const runtimes = { ...existing.runtimes };
let changed = false;
if (isLmStudio && !runtimes['claude-code']) {
runtimes['claude-code'] = {
baseUrl: origin,
wireProtocol: 'anthropic-messages',
models: [...models],
};
changed = true;
}
if (!runtimes.codex) {
runtimes.codex = {
baseUrl: pi.baseUrl,
wireProtocol: 'openai-chat',
models: [...models],
};
changed = true;
} else if (runtimes.codex.wireProtocol === 'openai-responses') {
runtimes.codex = {
...runtimes.codex,
wireProtocol: 'openai-chat',
};
changed = true;
}
return changed ? { ...existing, runtimes } : existing;
}