agenix

A TypeScript SDK for building tool-using LLM agents with provider adapters, streaming support, memory/planner abstractions, and a composable execution loop.

Features

• OpenAI-compatible provider support
• Gemini-compatible provider support (OpenAI-compatible endpoint)
• Claude-compatible provider support (Anthropic Messages API)
• Tool execution loop with iterative reasoning
• Streaming-first runtime with automatic fallback
• Event bus for lifecycle + stream chunk events
• Logger with runtime toggle wrapper
• Pluggable memory and planner abstractions

Install

pnpm add agenix

Minimal Usage

import { Agent, OpenAICompatibleProvider } from "agenix";

const provider = new OpenAICompatibleProvider({
  apiKey: process.env.OPENAI_API_KEY ?? "",
});

const agent = new Agent({
  provider,
  model: "gpt-4o-mini",
  loggerEnabled: true,
});

const result = await agent.run("Give me 3 short TypeScript SDK testing tips.");

console.log(result.finishReason);
console.log(result.response?.message);

Provider Quick Setup

import {
  OpenAICompatibleProvider,
  GeminiCompatibleProvider,
  ClaudeCompatibleProvider,
} from "agenix";

const openai = new OpenAICompatibleProvider({
  apiKey: process.env.OPENAI_API_KEY ?? "",
});

const gemini = new GeminiCompatibleProvider({
  apiKey: process.env.GEMINI_API_KEY ?? "",
});

const claude = new ClaudeCompatibleProvider({
  apiKey: process.env.ANTHROPIC_API_KEY ?? "",
});

Tools + Loop

import { Agent, OpenAICompatibleProvider } from "agenix";

const agent = new Agent({
  provider: new OpenAICompatibleProvider({
    apiKey: process.env.OPENAI_API_KEY ?? "",
  }),
  model: "gpt-4o-mini",
  tools: [
    {
      name: "now",
      description: "Return current UTC time",
      async execute() {
        return { now: new Date().toISOString() };
      },
    },
  ],
});

const result = await agent.run("Call now and summarize the output.");
console.log(result.messages);

Streaming + Events

import { Agent, AgentEventBus, OpenAICompatibleProvider } from "agenix";

const events = new AgentEventBus();

events.on("provider:chunk", ({ data }) => {
  const payload = data as { chunk?: { delta?: { text?: string } } };
  if (payload.chunk?.delta?.text) {
    process.stdout.write(payload.chunk.delta.text);
  }
});

const agent = new Agent({
  provider: new OpenAICompatibleProvider({
    apiKey: process.env.OPENAI_API_KEY ?? "",
  }),
  model: "gpt-4o-mini",
  events,
  preferStreaming: true,
});

await agent.run("Explain event-driven architecture in one paragraph.");

Logging Toggle

const agent = new Agent({
  provider,
  model: "gpt-4o-mini",
  loggerEnabled: false,
});

Documentation

• docs/README.md
• docs/quickstart.md
• docs/providers.md
• docs/architecture.md
• docs/streaming.md
• docs/events-logging.md

Examples

• examples/README.md
• examples/minimal-openai.ts
• examples/providers-switch.ts
• examples/with-tools.ts
• examples/streaming-events.ts
• examples/custom-memory-planner.ts

Development

pnpm clean
pnpm build
pnpm test -- --run
pnpm typecheck

Publish Gate

prepublishOnly enforces:

• clean
• build
• tests
• typecheck