@lenylvt/pi-ai

Unified LLM client layer for the four built-in providers kept in this fork:

• anthropic
• github-copilot
• openai-codex
• openrouter

The package ships built-in model catalogs for those providers only. Custom models and custom providers are still possible through the generic Model<Api> and registry APIs.

Install

bun add @lenylvt/pi-ai

TypeBox exports are re-exported from @lenylvt/pi-ai: Type, Static, and TSchema.

Built-in Providers

| Provider | Auth | Notes | |----------|------|-------| | anthropic | OAuth or API key | Claude Pro/Max or ANTHROPIC_API_KEY | | github-copilot | OAuth or token | COPILOT_GITHUB_TOKEN, GH_TOKEN, or GITHUB_TOKEN | | openai-codex | OAuth | ChatGPT Plus/Pro subscription | | openrouter | API key | OPENROUTER_API_KEY |

Built-in APIs

• anthropic-messages
• openai-completions
• openai-responses
• openai-codex-responses

github-copilot uses a mix of anthropic-messages, openai-completions, and openai-responses depending on the selected model. openrouter uses openai-completions. openai-codex uses openai-codex-responses.

Quick Start

import { complete, getModel, Type, type Context, type Tool } from "@lenylvt/pi-ai";

const model = getModel("anthropic", "claude-sonnet-4-5");

const tools: Tool[] = [
	{
		name: "get_time",
		description: "Get the current time",
		parameters: Type.Object({
			timezone: Type.Optional(Type.String()),
		}),
	},
];

const context: Context = {
	systemPrompt: "You are a helpful assistant.",
	messages: [{ role: "user", content: "What time is it?", timestamp: Date.now() }],
	tools,
};

const message = await complete(model, context);
console.log(message.content);

Providers and Models

import { getModel, getModels, getProviders } from "@lenylvt/pi-ai";

console.log(getProviders());
// ["anthropic", "github-copilot", "openai-codex", "openrouter"]

const anthropicModels = getModels("anthropic");
const codex = getModel("openai-codex", "gpt-5.4");
const openRouter = getModel("openrouter", "openai/gpt-5.1-codex");

Environment Variables

| Provider | Environment Variable(s) | |----------|-------------------------| | anthropic | ANTHROPIC_API_KEY or ANTHROPIC_OAUTH_TOKEN | | github-copilot | COPILOT_GITHUB_TOKEN, GH_TOKEN, or GITHUB_TOKEN | | openrouter | OPENROUTER_API_KEY |

openai-codex does not use a static API key in this fork. Authenticate with OAuth.

You can inspect environment-based auth with getEnvApiKey():

import { getEnvApiKey } from "@lenylvt/pi-ai";

const anthropicKey = getEnvApiKey("anthropic");
const openRouterKey = getEnvApiKey("openrouter");

OAuth

OAuth helpers are exported from @lenylvt/pi-ai/oauth.

CLI Login

bunx @lenylvt/pi-ai login
bunx @lenylvt/pi-ai login anthropic
bunx @lenylvt/pi-ai login github-copilot
bunx @lenylvt/pi-ai login openai-codex

Programmatic OAuth

import {
	getOAuthApiKey,
	loginAnthropic,
	loginGitHubCopilot,
	loginOpenAICodex,
	type OAuthCredentials,
} from "@lenylvt/pi-ai/oauth";

Stored credentials are the caller's responsibility.

Thinking / Reasoning

Use streamSimple() or completeSimple() with a unified reasoning level:

import { completeSimple, getModel } from "@lenylvt/pi-ai";

const model = getModel("openrouter", "openai/gpt-5.1-codex");

const response = await completeSimple(
	model,
	{
		messages: [{ role: "user", content: "Refactor this function", timestamp: Date.now() }],
	},
	{
		reasoning: "medium",
	},
);

Streaming

import { getModel, stream } from "@lenylvt/pi-ai";

const model = getModel("github-copilot", "gpt-5");
const s = stream(model, {
	messages: [{ role: "user", content: "Summarize this diff", timestamp: Date.now() }],
});

for await (const event of s) {
	if (event.type === "text_delta") {
		process.stdout.write(event.delta);
	}
}

Custom Models

Custom models still work. Use openai-completions or anthropic-messages for the common proxy patterns supported by this fork.

import type { Model } from "@lenylvt/pi-ai";

const openRouterModel: Model<"openai-completions"> = {
	id: "openai/gpt-5.4-mini",
	name: "OpenRouter GPT-5.4 Mini",
	api: "openai-completions",
	provider: "openrouter",
	baseUrl: "https://openrouter.ai/api/v1",
	reasoning: true,
	input: ["text"],
	cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
	contextWindow: 128000,
	maxTokens: 16384,
	compat: {
		supportsDeveloperRole: true,
		supportsReasoningEffort: false,
	},
};

Public Exports

Provider-specific option types kept in this fork:

• AnthropicOptions
• OpenAICompletionsOptions
• OpenAIResponsesOptions
• OpenAICodexResponsesOptions

Notes

• openai-codex requires ChatGPT Plus or Pro.
• github-copilot can require enabling some models in VS Code before the token can use them.
• openrouter is the source of truth for its model availability; use the published model list bundled with each release.

License

MIT